This PDF file contains the front matter associated with SPIE Proceedings Volume 13184, including the Title Page, Copyright information, Table of Contents, and Conference Committee information

Submarine cable is an important channel for underwater power transmission and information transmission. However, due to the influence of ocean circulation and fishing vessel anchoring, the actual position of submarine cable is very easy to change. To solve the above problems, the active sound source is used to locate the submarine cable. The ship generates several signals along the submarine cable route, receives and demodulates the acoustic signals through the built-in optical fiber of the submarine cable, and uses the time difference of arrival (TDOA) algorithm to locate the position of a single channel, so as to determine the position of the entire submarine cable.

This paper proposes a detection method for missing small sized parts based on machine vision, which is able to determine whether small sized parts placed on a pallet are missing. Firstly, the high-resolution industrial camera is utilized to photograph the small parts placed on a pallet. Rectangle detection and template matching techniques are employed to estimate the pose of the pallet. Subsequently, based on the pallet template and the estimated pallet pose, the rectangle bounding boxes of the small sized parts are located. Using a pixel statistics method, it is determined whether there are parts within all the bounding boxes, and the missing parts are detected. Finally, a visual inspection system for detection of the missing small sized parts is set up and experiments are conducted. The results indicate that the method proposed in this paper can achieve the detection of missing small sized parts within the pallet.

An improved planar UWB antenna based on hexagonal structure is designed. By introducing the defect structure to the ground plane of the regular hexagonal antenna, the -10dB impedance bandwidth of the antenna is 3.0-19.2GHz, covering the frequency range of UWB communication and higher frequency range. The antenna has a simple structure and a size of 26mm*31mm*1.6mm. The pattern and impedance matching in the whole bandwidth range meet the requirements of UWB communication.

Based on the xfab-0.18um HV SOI process, a self-adaptive BUCK-type dual-channel synchronous DC-DC controller was designed, using the D-CAP ( Directly from the output capacitors) mode with adaptive conduction time for loop control. The D-CAP mode has high efficiency, fast response, and simple The D-CAP mode has high efficiency, fast response, and simple loop control, simplifying the circuit while improving the stability and reliability of the system circuit. synchronous buck controller includes three operating modes: PWM mode, Auto skip mode, and PSM mode, which can address applications with light loads and heavy loads. The final designed dual-channel synchronous buck controller includes three operating modes: PWM mode, Auto skip mode, and PSM mode, which can address applications with light loads and heavy loads. The controller supports an input voltage range of 5.528V and an output voltage range of 25.5V. efficiency with an input voltage of 12V and an output voltage of 5V. Under the highest frequency application, the ripple is less than 50mV, and the load transient response time is less than 3%. Under the highest frequency application, the ripple is less than 50mV, and the load transient response time is less than 3us. Finally, the overall circuit was simulated and analyzed, and the results met the expected targets.

The advancement of modern industry towards digitization and automation has led to the widespread utilization of electronic components in coal mining equipment. Various electronic components are extensively employed in industrial equipment, safety early warning systems, emergency equipment, and military products. These equipment, which have been in operation for an extended period, are susceptible to failure risks. This study focuses on investigating the failure issues of electronic components in coal mining equipment and conducting research on reliability evaluation technology. By subjecting these devices to accelerated tests, the vulnerabilities are identified, and the failure modes and mechanisms are determined. Subsequently, practical recommendations for quality enhancement are proposed. This study serves as a valuable resource for the reliability assessment of electronic components

As quantum computing technology matures, the availability and performance of quantum devices are steadily improving. However, in the NISQ (Noisy Intermediate-Scale Quantum) era, the quantum bit error rate caused by quantum noise remains a significant constraint on computational accuracy and reliability. In quantum operating systems, quantum task scheduling is a core technology crucial for fully leveraging the performance potential of quantum processors. Currently, most multi-processor scheduling algorithms focus on task load balancing and do not directly consider the characteristics of quantum circuits and the differentiated error rates of quantum computing backends. In quantum computing, the impact of noise errors in single-qubit gates, two-qubit gates, and measurement gates on the fidelity of quantum gate circuits is crucial. Therefore, intelligently matching quantum tasks with the characteristics of backend processors to optimize computational accuracy is an urgent problem to be addressed. This paper proposes a new quantum task scheduling algorithm, which matches tasks with backends based on the weight of single-qubit gates, two-qubit gates, and measurement gates in quantum circuits, as well as the error rate characteristics of each quantum computing backend. Experimental results show that the proposed scheduling algorithm for improving the fidelity of quantum gate circuits outperforms static load balancing scheduling algorithms, resulting in a 4.39% increase in the fidelity of quantum tasks.

Using computational fluid dynamics (CFD) simulations, this study examines airflow organization in modern ship meeting rooms. It investigates indoor and velocity distribution, calculates air age and Predicted Mean Vote (PMV), analyzes indoor thermal comfort, and proposes optimization strategies for ship meeting room air conditioning design.

For high-frequency periodic signals, this design is based on the principle of equivalent sampling, which greatly reduces the requirement for the sampling rate of the ADC chip and cuts down the cost of observing periodic signals. This design proposes a way to use STM32 to measure the frequency, control the AD9854 to output the clock signal of high-precision frequency to drive the ADS803 chip, and display the acquired signal on the serial screen. For frequency measurement, this design provides a high-precision temperature-compensated crystal for STM32 and AD9854, and uses equal-precision frequency measurement to measure the precise frequency of the signal. This accurate frequency is utilized to calculate the appropriate real-time sampling rate. In terms of signal conditioning, this design utilizes the OPA659 as an amplifier to make the 1Hz~20MHz signal attenuate and amplify normally. The measurement frequency range of this design is 10Hz~10MHz, the vertical sensitivity range is 10mV/div~2V/div, the real-time sampling rate is less than 1Msa/s, the equivalent sampling rate is more than 200MSa/s, and the scanning speed range is 100 ns/div~10ms/div.

Welding process is a critical part of industrial manufacturing, and the welding quality directly affects the final construction cycle and production cost. Therefore, to predict the welding quality efficiently and accurately. A welding quality prediction method based on real-time data is proposed in this paper. Firstly, a welding quality prediction model combining genetic algorithm and back propagation neural network is proposed to achieve accurate prediction of welding quality. Then, the experiment results demonstrated that this method had satisfactory performance and could be applied to real-time accurate prediction of welding quality. Compared with the traditional BP algorithm prediction model, the coefficient of determination is improved by 44.46%, and the prediction accuracy is over 91.236%.

Solving the small sample problem can not only save data acquisition costs, but also enhance the model's generalization ability to unknown data, thereby ensuring the accuracy and reliability of fault diagnosis results. Therefore, this paper proposes a small-sample bearing fault diagnosis method that combines the ECA attention mechanism and the DenseNet network (ECA-DenseNet). First, the two-dimensional DenseNet network and ECA attention mechanism are converted into one-dimensional networks to reduce computational complexity and information loss. Secondly, the size and position of the convolutional layer in the network are improved and adjusted, and the 1D-ECA module is introduced into the DenseNet model. Finally, we use the original signal as the input of ECA-DenseNet. After feature recognition, the network classifier completes fault feature classification. We verified the effect of the proposed model under different working conditions on the laboratory PT500 bearing data set, and compared it with other diagnostic models. Experimental results show that the proposed method has high fault identification accuracy and generalization ability under small sample conditions.

With the continuous evolution of communication systems, the application of directional transmission signals or directional reception signals is crucial in the field of signal processing, such as radar systems or satellite communication systems. In addressing the enhancement of beamforming performance for wideband signals, a modeling study on wideband frequency invariant beamforming is conducted, proposing a beamforming algorithm based on Principal Component Analysis (PCA). This algorithm, utilizing a designed uniform array, employs PCA to manipulate the weight matrix, altering the array position and weights based on the high-energy contributions to derive a comprehensive directional map. Simulation experiments demonstrate that the novel approach ensures the integrity of the frequency-invariant beamformer while enhancing its performance. By applying Principal Component Analysis to a uniformly spaced linear array input, the method addresses parameter estimation issues, enhancing accuracy while ensuring better sparsity.

In view of the technical bottlenecks of eutectic salt electric superheat detector used in aircraft duct leakage monitoring, such as poor electromagnetic compatibility, low measurement accuracy and slow response time, a linear fiber optical sensor for aircraft duct leakage detection is studied in this paper. The research demonstrates that the sensor is capable of adapting to the operational conditions of pipeline bending assembly, with a temperature sensing accuracy of 0.1m, positioning accuracy of 1m, temperature measurement accuracy within ±3℃, and a working temperature range from 0- 180℃. Moreover, it can replace the eutectic salt leak detection sensors to achieve remote distributed measurement.

Diffusion neural network is a neural network model based on diffusion process, which can effectively denoise and denoise data. In addition, Transformer neural network is a neural network model based on attention mechanism, which has been widely used in natural language processing. In other fields, it can effectively learn the dependency between different positions in the input sequence and obtain the positioning coordinates. Combining the functions of the two neural networks, an indoor Bluetooth positioning method based on the Transformer neural network is proposed. This method uses the Diffusion neural network to denoise the Bluetooth positioning data to obtain high-quality positioning data; Input it into the Transformer neural network to obtain the positioning coordinates, so as to realize indoor positioning. By combining these two neural networks, the accuracy and robustness of indoor localization can be improved. The experimental results show that the denoising method based on the Diffusion neural network can effectively reduce the impact of multipath effects on the data, reduce the overall error of the data to 1 meter, and the positioning method based on the Transformer neural network can reduce the positioning error is controlled within 0-2 meters, and the average error is 1 meter.

Power transformers are among the most important and expensive equipment in power system. In order to planning maintenance and capital expenditures efficiently, it is vital for firms to predict the life of transformers accurately. Though standard Weibull is applied widely, it fails to capture the situation where the hazard rates are not monotone or constant. In order to achieve these behaviors from a single distribution, the odd Weibull distribution is adopted and further truncated since the lifespan of a transformer is usually restricted by its technical life. Parameters are estimated by maximizing the logarithm of likelihood function. Empirical studies proves the adequacy of the proposed model.

The quality of products is getting more and more attention, which posing a serious threat to people's lives and property safety. We attaches great importance to product quality development. The failure of mechanical products sometimes leads to serious consequences. Therefore, it is necessary to research the reliability of mechanical products. This study focuses on actuator fault of nonlinear system. The soft fault diagnosis is a tough problem in the design of the system reliability. However, traditional fault estimation algorithms have poor performance on soft fault. Firstly, a novel observer is studied to estimate the early potential fault. Two typical faults fast-varying and slow-varying) are considered and estimated. Finally, the advantage of the proposed method is verified by simulation

The conventional maximum power tracking (MPPT) approach is inapplicable to the problem of the photovoltaic system's output power curve having numerous peaks and many steps as a result of local shading of the solar array under actual illumination. In order to maximize power point tracking, this research suggests an enhanced whale method for BP neural network optimization. Building upon the foundation of the original whale algorithm, this method introduces adaptive inertia weight strategy and chaotic initialization, hence improving the algorithm's ability to exit the local optimal position. Subsequently, the initial parameters are inserted into the training network to anticipate the maximum power point, and the BP neural network's initial connection weights and biases are optimized using the enhanced whale method. Using the enhanced whale algorithm to optimize the BP neural network to realize photovoltaic MPPT has good tracking speed and accuracy, and the oscillation at the maximum power point is clearly minimized, as demonstrated by MATLAB simulation verification.

The academic paper introduces a multi-strategy enhanced Sand Cat Swarm Algorithm (TL-SCSO) designed to tackle the challenges of limited population diversity, poor convergence accuracy, and susceptibility to local optima. The algorithm leverages Cubic chaos mapping to enrich the initial population distribution, ensuring a more diverse and representative starting point. Additionally, an adaptive t-distribution perturbation approach is employed to maintain population diversity throughout the optimization process. The incorporation of the golden ratio into the Levi's flight strategy enables the algorithm to effectively navigate away from local optima, enhancing its overall performance. Finally, the algorithm is utilized to optimize the hyperparameters of BiTCN-BiLSTM-Attention, thereby improving model training speed and prediction accuracy. The algorithm's efficacy is demonstrated through its application to data from a photovoltaic power station in central China, showcasing its superior stability and convergence accuracy under various weather conditions.

In the process of aircraft design, various experiments are required to validate design parameters under real flight conditions. Such experiments are costly, and currently there are limited solutions for effectively dealing with scenarios such as occasional faults during these experiments. This paper proposes an on-board remote high-definition collaborative troubleshooting and flight decision-making assistance device, aiming to achieve efficient test scenario replication and decision support through remote collaboration between ground personnel and flight crew. The device utilizes several image capture devices in the cockpit and employs algorithms such as action recognition and image recognition. It utilizes highspeed satellite communication technology to structure flight data, including pilot operations, faults, and aircraft status, into a timeline script. Ground personnel can use this output script to provide corresponding solutions. Flight crew members receive real-time guidance and decision support through smart glasses. It is evaluated that this approach may save a significant amount of travel expenses and time in ground tests, flight tests, field fault diagnosis, and crew training scenarios. The innovation of the proposed device lies in the application of cloud servers for timeline script generation and the use of remote retrieval by monitoring personnel for key operation fault diagnosis. Furthermore, it incorporates intelligent recognition technology for flight guidance, enabling efficient troubleshooting and reducing the number of flight tests. Therefore, research and development costs could be reduced, and engineering change progress could be accelerated.

Most air quality monitors and lighting fixtures are currently separate, with air quality monitors typically being stand-alone devices that require separate power supplies and mounting locations. This not only occupies valuable space but also complicates their use and management. The emergence of air quality monitoring ceiling lights in smart homes aims to address this issue. However, many existing ceiling light designs do not consider heat dissipation, which can lead to performance issues, reduced lifespan, and even fire hazards. To tackle these challenges, this study presents the design of an intelligent ceiling lamp for air quality monitoring utilizing air duct technology. Equipped with a seven-in-one air quality sensor, the ceiling lamp collects indoor air data and performs real-time intelligent control and anomaly identification through a cloud-side-end synergy approach. The incorporation of air ducts enhances the sensitivity and accuracy of air quality monitoring, facilitates thermal convection, and improves heat dissipation efficiency, ultimately enhancing performance and longevity. Physical product design and field tests conducted on the cloud-side-end platform demonstrate that the ceiling lamp's air quality monitoring and anomaly identification have an average inference time of less than 1 millisecond with an identification accuracy of over 98%. Furthermore, the air duct design improves thermal efficiency by approximately 28%. Overall, this intelligent ceiling lamp represents a practical, efficient, and intelligent multi-functional indoor lighting solution.

An ultra-wideband (UWB) antenna with tri-notch is designed, which size is 22mm×31.5mm×1.6mm. The antenna operates in the bandwidth of 2.7 to 12GHz. The antenna radiation patch consists of rectangular, circular and triangular patches, and the ground plane is a rectangle etched with slots. By etching C-shaped slots and open-rectangular-ring(ORR) slots on the radiating patch and adding ORR parasitic resonators on the ground plane, triple-notch waves are generated, which can suppress three bands. C-band downlink frequency band (3.7 to 4.2 GHz) and WLAN frequency band (5.15 to 5.85 GHz), UNII-2c (5.47 to 5.725GHz), WLAN uplink band (5.725 to 5.825GHz), X-band downlink frequency band (7.25 to 7.75 GHz) have been suppressed effectively. The antenna has good omnidirectional properties in the UWB band. Simulation results show that the antenna can be used in UWB communication system.

This paper designs and makes a high precision and miniature fiber Bragg grating (FBG) gas flow sensor for the low measurement accuracy and large volume of traditional electromagnetic flow sensor in complex working environment. The flow sensor is selected, and the high ratio diameter nozzle is used as the throttle element to produce the pressure difference on both sides of the nozzle. By pasting the FBG along the radial direction on the sensitive diaphragm, the deformation caused by the induction diaphragm due to the pressure difference, so as to associate the flow change with the FBG wavelength drift. The mathematical relationship between the measured gas flow and FBG wavelength values is derived from theoretical and simulation, and the dynamic flow characteristics of the flow sensor are obtained through experimental calibration. The experimental results show that the flow change and FBG wavelength drift, consistent with the theoretical analysis, the flow sensor dynamic flow measurement sensitivity is 11.04 NL/min. The FBG gas flow sensor designed in this paper has the advantages of high precision, miniaturization, strong anti-electromagnetic interference ability and easy integration in the complex working environment.

The emerging power system, predominantly reliant on renewable energy sources, drives the future diversification of enduse energy structures. However, within this new power system, business terminals encounter various access risks. Terminals within unmanned aerial vehicle (UAV) power inspection systems, distributed photovoltaic business, and vehicular networking are susceptible to network attacks, exposing them to risks of information leakage. To address the security threats prevalent in the emerging power industry, this paper proposes a Bayesian network-based threat modeling approach. Building upon this, a probabilistic graphical model-based threat risk assessment scheme is presented. Furthermore, typical interaction scenarios of novel power system operations are selected to demonstrate the effectiveness of the model by employing Bayesian networks for modeling and simulation validation.

Thermal runaway of lithium batteries is a matter of great concern to relevant scholars. The gas pressure during the charge and discharge control process of lithium batteries is directly related to the performance of lithium batteries. Relevant scholars have been pursuing a method that can safely monitor the internal gas pressure of lithium batteries. Based on the characteristics of optical fiber sensors such as small size, anti-electromagnetic interference, and intrinsic safety, this paper designs and prepares an optical fiber gas sensor that can be used to measure the internal gas of lithium batteries. The measurement range of the sensor is 0-240kPa, and the resolution of the sensor is 0.5kPa, the sensor's accuracy reaches 5kPa, which can be used as a feasible detection method for gas generated during the working process of lithium batteries.

Underwater acoustic (UWA) modem is an important part of underwater information transmission field, which is widely used in underwater wireless sensor network, coral reef and ocean monitoring. At present, most commercial modems are expensive and bulky, which limits their large-scale application. It is based on the low cost, miniaturization, robust UWA modem this demand, first of all, this paper takes NXP iMXRT1062 MCU as the core processor, designed the corresponding receiving and transmitting circuit, secondly, in the design of hardware based on the realization of UWA FH-2FSK communication algorithm. The actual lake test results show that the designed modem can achieve a communication rate of 100 bps in the complex shallow water area, and the communication distance is 70 m.

The accelerated storage life assessment method based on the reliable growth of assembly-level products exploits the equality of step-down-stress life testing and reliability growth testing, which tackles the problem that the accelerated model is difficult to determine when using the accelerated life assessment method based on the accelerated model. A segmented expression of the function of step-down accelerated model and accumulative step-down accelerated model are derived and the expression of accelerated factor is obtained by using the step-down accelerated model. Combined with the first step test result and the storage objective age limit, the ideal step-down accelerated curve and accumulative ideal step-down accelerated curve are presented. Through the comparison of the follow-up step test results and the first step test result, the objective age limit of the assembly-level products is confirmed. The technology introduced is validated with the resultant coincidence of evaluated field storage lives.

To solve the problem of reliability of mass production and long-term storage of spacecraft harness, this paper carries out systematic analysis and verification. Based on the composition of harness and the environmental stress during storage, the failure mode of stored harness and the occurrence probability of various faults are determined, and the wire is located as the weak point of harness. Then, the reliability verification model and life assessment model of harness were established, and relevant tests were carried out. The test results show that the performances and end-of-life reliability of the stored harness meet the application requirements of most spacecrafts. The conclusion provides technical support for the mass production and storage of spacecraft harness.

As aluminum profiles become increasingly popular in aerospace and aircraft manufacturing, the detection and quality management of surface defects in aluminum profiles are crucial. This study enhances the YOLOv5 algorithm for detecting surface defects in aluminum profiles, using images from a company's production line. It introduces K-Means++ to the adaptive anchor box algorithm for optimal initial center selection in clustering. Additionally, C3 was replaced with C2f, and the SE attention mechanism was introduced. Experiments demonstrate that the optimized YOLOv5 algorithm performs excellently. It not only achieves faster detection speeds but also improves detection accuracy, effectively addressing the challenge of low recall rates for small and slender targets.

As a non-destructive and real-time detection technology, acoustic emission (AE) testing has been widely applied in the state prediction and fault diagnosis in wire electrical discharge machining (WEDM). However, there has been limited research investigating the sources of interference affecting AE signals during the machining process. These sources of interference can induce noise, resulting in AE signals distortion. Therefore, this study designs a filtering algorithm based on discrete wavelet transform (DWT) to reduce the impact of noise on the AE signals. The experimental results indicate that noise mainly resides within the frequency ranges of [500,000Hz, 1,000,000Hz] and [0Hz, 15,600Hz], while AE signals generated by the workpiece are primarily distributed in the range of [32,300Hz, 500,000Hz]. The signal-to-noise ratio (SNR) of the filtered AE signal is 10.7dB, which is beneficial to analyze and extract features of signals in the follow-on work.

With the development of prefabricated construction, substation engineering construction is also shifting towards modular construction. Therefore, prefabricated steel structure substations have been widely used in power systems. However, the construction of prefabricated substations currently relies on a large number of on-site experiments and lacks accurate supervision and feedback methods. This article proposes a technical route for the construction of simulated prefabricated steel structure substations based on digital twin technology and carries out its application. Through BIM, IOT, and AI, the construction of digital twin models and data transmission are completed. This provides convenience for achieving digital expression of substation construction, simulating assembly processes, and optimizing construction processes.

The paper designed a permanent magnet vibration damping chair. Two magnetic rings with opposite poles are arranged on the base of the chair and the bottom of the chair body, and the entry is arranged on the side of the chair about the principle of magnetic levitation. The parametric model was established through the three-dimensional modeling software Solid-works. The magnetic field simulation analysis showed the magnetic field strength, distribution and repulsive force on the magnetic ring surface, and the first six order constrained mode analysis. The simulation results show that the chair can produce copper resistance adult weight of repulsive force.

With the rapid development of China's water conservancy and hydropower industry, the mining industry is also developing at a high speed, but with the continuous development of mineral resources, doing a good job of waterproofing and drainage is a prerequisite for ensuring the safety and normal production of open-pit mines. Especially the depressed open-pit mines, which objectively have the conditions to collect atmospheric precipitation, surface runoff, and underground water gusher, so during the whole process of open-pit mine production, Effective drainage measures must be taken even during construction. In this paper, the ground-to-air frequency domain electromagnetic method is used to detect leakage around the mine. After energization, an air-core coil is used to detect the measured area. During the experiment, three more serious leakage parts were found. As long as there is leakage and flowing water, etc. You can detect the occurrence of water flow and other situations. The effect is good and meets the required requirements.

Aiming at the diversity of marine pipe fittings production, the complexity of connection methods and production processes, this paper is devoted to the optimization of production and processing of pipe fittings workshop aiming at the minimum completion time. An improved cuckoo algorithm is introduced through mathematical modeling to solve the production and processing problems of the pipe fitting workshop. Based on the basic cuckoo search algorithm, an improved two-layer integer coding method is used, a Levy flight adaptive step factor is introduced, and a nest parasitic strategy based on two kinds of neighborhood local search is designed. The algorithm is applied to the enterprise production model in the experiment. The results show that the scheduling model algorithm can quickly obtain the optimal target scheduling scheme to achieve the purpose of optimizing the production and processing of ship fittings.

In order to solve the problem of low transformer fault recognition rate, this paper proposes a fault diagnosis method based on feature selection and improved bald eagle search algorithm (IBES) to optimize random forest (RF). Firstly, to solve the problem that the specific gravity of dissolved gas in transformer oil is difficult to diagnose efficiently, this paper proposes to use 7 ratio relations between five gases and the specific gravity of the original five gases as fault characteristics, and use RF for feature selection. Secondly, to solve the problem of low convergence accuracy of the Bald eagle search algorithm (BES), this paper uses Sine chaotic mapping, Levy flight and Cauchy Gaussian variation perturbation strategy to improve BES. Finally, in order to improve the diagnostic accuracy of RF model, IBES is used to optimize RF parameters and build IBES-RF model. The simulation results show that compared with BES-RF and PSO-RF models, IBES-RF has the best fault diagnosis effect, and the accuracy rate is 90.51%.

Analog circuits are an important part of modern electronic power systems. Accurate detection of analog circuit faults, especially soft faults, is of great significance for the maintenance and inspection of electronic systems. This paper proposes to apply the Boruta feature selection method to the field of soft fault diagnosis of analog circuits to screen out lowdimensional and efficient feature components from the high-dimensional time-domain statistical features and frequencydomain features of circuit responses. Then, the feature components are used as input to train the LightGBM classification model, and the Bayesian optimization method is introduced to optimize the model hyperparameters. Finally, the trained fault diagnosis model is verified in two typical experimental circuits, and satisfactory accuracy is obtained.

Sensor technology plays an important role in the field of measurement science and engineering. Fabry-Perot (F-P) sensor has always been the focus of research due to its high precision and sensitivity. Traditional F-P sensor demodulation algorithms have limitations in terms of demodulation time and precision. In order to pursue higher accuracy and achieve faster demodulation, we report a joint demodulation algorithm based on Fibonacci and cross-correlation. The Fibonacci search method is used to quickly search the cavity length corresponding to the maximum cross-correlation number, and the target accuracy is achieved through search iteration. Experimental results show that this algorithm is used for DualCavity Fabry-Perot Interferometer (DFPI) sensor demodulation, with a resolution of 10 pm and a cavity length demodulation time of less than 0.06 s. This study provides new insights for improving the demodulation efficiency and accuracy of F-P sensors.

Beam intensity is a fundamental parameter of beam flow. The beam intensity measurement system of the Heavy Ion Medical Machine (HIMM) requires a wide dynamic range current detection card to monitor the beam intensity. The card will receive the current signal and integrate it within the fully differential current frequency converter (FDCFC). A voltage comparator is employed to check the output against the threshold and control the current integrate direction. The input current signal is transformed into a TTL signal by integrating and changing current direction between the differential pairs. The card utilizes a high slew rate differential amplifier and a multi-switch, and the current is integrated in different branches of pairs. The laboratory tests show the following results: The fully differential current frequency converter performs effectively, with an input dynamic range covering 1nA to 10mA, a nonlinear error of less than 15%, and an output frequency ranging from 100Hz to 200kHz, meeting all the design goals.

Surface defects on transmission towers usually include cracks, corrosion, rust, and other issues, and these defects often have different shapes and sizes. Therefore, designing accurate identification methods to ensure that drones can reliably identify various types of defects is a challenge at this stage. Therefore, a new UAV tracking and recognition method for transmission tower surface defects is designed by combining edge computing method. Based on edge computing, the data collected by UAVs on the surface of transmission towers are classified accurately. Based on this, a multi strategy PSO is used to achieve unmanned aerial vehicle tracking of surface defects on transmission towers. Simulated annealing algorithm is introduced to improve particle swarm optimization, and a multi-strategy particle swarm optimization algorithm is formed. According to Sobel operator, UAV tracking and identification of surface defects of transmission towers are completed. The experimental results show that the research method can track all defects on the surface of transmission towers, and has higher accuracy in identifying defects.

With the increasing complexity of the electromagnetic environment and the severity of electromagnetic radiation problems, in order to accurately obtain the amplitude of spatial electric field strength and effectively reduce the harm caused by electromagnetic radiation, a method of using a square structure to load the antenna arm based on the lumped resistance loaded direct dipole antenna and fractal theory is proposed. The influence of the length of a single circle and different number of circles of the square structure on the frequency response and electric field strength deviation (i.e. flatness deviation) of the electric field probe antenna in the frequency band of 50KHz~1000KHz is studied. A new type of resistance loaded electric field probe sensing dipole antenna based on the square structure is designed and manufactured. The actual test results show that the frequency response of the antenna has been extended by using a square structure loading, effectively reducing the flatness deviation of the electric field probe antenna. The low-frequency operating frequency band of the existing electric field probe with flatness error within ± 3dB has been extended from 1000KHz to 70KHz, and the flatness deviation in the central frequency band from 200KHz to 1000KHz has been optimized from ± 2.53dB to ± 0.45dB. Effectively expanding the detection frequency band of the electric field probe provides a new sensing antenna for measuring electromagnetic radiation in lower frequency bands.

Human acupoints have dynamic characteristics, that is, the parameters of human acupoints change dynamically under physiological or pathological conditions, but acupoints do not exist in the fixed position of human body surface like the foothold of traditional acupuncture and moxibustion treatment. therefore, in order to explore the dynamic characteristics of acupoint impedance when human body state changes and analyze the performance of abnormal meridian acupoints, this study proposes to use human body impedance parameter distribution measurement and characteristic analysis. Identify the abnormal situation of human acupoints, directly display the measured electrical impedance by the way of color map, reflect the different information of acupoints, and then infer the health status of the human body.

The accelerometer signal is capable of providing crucial information regarding object motion, posture, and vibration. Therefore, it is of great significance to investigate noise suppression techniques for accelerometer signals. However, in complex industrial environments and under various equipment conditions, the noise generated exhibits a diverse nature, leading to suboptimal noise reduction outcomes when employing conventional denoising algorithms. This paper proposes a neural network model (AgentUNet) for denoising accelerometer signals. The network consists of a combination of UNet and Agent Attention, enhancing its ability to learn contextual information and achieve superior denoising performance. Additionally, instead of using clean signals as labeled data, we train the neural network using pairs of noisy data, thereby alleviating the challenge of collecting clean signals in industrial environments. Finally, we conduct experiments on experiment dataset. The experimental results indicate that AgentUNet achieves superior denoising performance compared to other baseline models. Furthermore, the denoising effectiveness of the neural network trained without using clean signals as labeled data is similar to that of the network trained using supervised learning methods.

In the field of machine vision, algorithms for estimating optical flow in sequential images have been proposed for many years. However, due to the advancements in image resolution and video frame rates of sensor, the optical flow algorithm running on a CPU or a GPU may struggle to meet real-time processing and low power consumption requirements. This paper presents an FPGA-based accelerated method for high-speed LK optical flow estimation. We implemented a serial input and multi-window output buffer structure to enhance the algorithm's parallel processing capability and improve processing speed. Additionally, we designed a multi-clock convolution structure to reduce resource consumption. Tested on the Middlebury dataset, the FPGA-based LK optical flow algorithm developed in this paper can achieve a processing speed of 330 frames per second for image sequences with a resolution of 640x480. It also maintains a relatively high level of accuracy and has low power consumption.

Airborne equipment suffers a large overloaded force frequently due to the high dynamic characteristic of the airplane. To improve the performance of vibration isolation for overloaded payloads, an adaptive negative stiffness and sky-hook damping (ANS-SHD) control strategy is proposed based on a hybrid actuating system. The configuration and dynamic model of the system is illustrated and a simulation model of vibration isolation for time-variant payload with immense extents is established to validate the ability of proposed strategy. The simulation results demonstrate that the proposed strategy has a significant superiority compared with the sky-hook damping (SHD) control at 2-50 Hz and the vibration can be attenuated by 92.2% under the random excitation.

Real-time simulation is an important means to verify the reliability of renewable power stations. Considering the deficiencies of the current real-time model and simulation platform of renewable energy stations, this paper develops a real-time simulation model for PV power stations, and proposes a CPU-FPGA real-time simulation interactive method suitable for CPU-FPGA hardware platform. First, based on the simulation characteristics of each part of the PV power station, the model is constructed on FPGA (Field Programmable Gate Array) and CPU respectively. The parallel simulation of PV power generation unit is carried out on FPGA. Then, the CPU-FPGA simulation platform is built, and the communication framework is built based on the Aurora optical port protocol to realize the data interaction of the size step. Finally, compared the RTDS cross-unit model, the correctness of the proposed simulation method and communication framework is verified.

High levels of vibration affect the performance limit of machines, such as speed and efficiency. Vibration suppression is very important for the application of motion control system in high-end equipment. The performance of many complex and precise mechanical systems can be improved by using command shaping techniques to suppress machine vibration. The input shaping technique, originated in MIT artificial intelligence laboratory, has proven to be highly effective and strongly robustness for flexible mechanical systems. This paper studied and derived the mathematical expression of the input shaper, listed and compared the characteristics and robustness of several main shapers, and simulated the vibration suppression effect of Extra insensitive shaper. And then, the vibration suppression effect of four kinds of shaper was verified carefully by experiments of a platform consisted of two axes linear motors. Appropriate comparison and discussion were carried out for the results. The conclusion had reference significance for practical engineering of motion control.

The energy storage battery primarily interacts with the microgrid through the power conversion system context of the rapid development of energy storage modules to regulate the comprehensive management of power quality inside the microgrid. Currently, the most popular power conversion system has a bipolar construction, with a front-stage DC/AC converter module for AC/DC conversion and a back-stage bidirectional DC/DC converter module for energy storage charging and discharging control. To begin, this paper chooses the architecture of the energy storage converter and develops its mathematical model using the two-phase rotating coordinate system. The adaptive trap with frequency tracking and filtering function is then used to extract the fundamental positive sequence voltage to obtain the synchronous rotation angle to inhibit the interference of the voltage unbalance distortion on the PLL; finally, the i_p-i_q current detection algorithm is proposed to take into account the active current calculation function and to complete the calculation of the active component and the integrated current det.

This paper presents a novel approach to improve the maintenance efficiency of expressway electromechanical devices through a preventive maintenance model based on a multi-factor model and the Analytic Hierarchy Process (AHP) method. We analyze eight commonly used devices in expressway maintenance projects across five dimensions to identify areas for preventive maintenance. Our proposed quantitative decision-making model provides a comprehensive calculation method to optimize equipment maintenance. The results show a significant improvement in overall maintenance efficiency for expressway electromechanical devices.

This study proposes a method for obtaining the enveloped energy spectrum using measured transient vibration time-history data. By combining the principles of random signal generation in control theory of vibration controller, we successfully constructed a time-domain waveform that possesses both the amplitude characteristics of the envelope energy spectrum and typical phase information from acquisited test data. using vibration table waveform reproduction technology, we conducted transient vibration environment testing on products. This method provides important technical support for future transient vibration vibration table test methods that directly utilize energy spectrum control.

Rapeseed oil is one of the most widely produced edible oils in the world and has a broad market prospect. The moisture content of rapeseed raw material directly affects the oil yield and oil quality. Microwave technology has been proved to be useful for measuring the moisture content of materials. An open microstrip resonant ring (MRR) sensor based on microwave technology was developed in this paper. The dielectric properties of the measured material are derived according to the S21-parameters of the sensor. A rapeseed moisture test platform consisting of a vector network analyzer, a MRR sensor and a container was built to verify the applicability of the density-independent algorithm in moisture of rapeseed measurement. The coefficient of determination of this density-independent algorithm is 0.982 The dielectric properties of rapeseed are divided by the moist bulk density of rapeseed and a density-independent function is obtained through a series of numerical fitting. The results show that the density-independent function can obtain accurate prediction results when the moisture content is 3 to 18%, and the standard performance errors are 0.821%. We designed a low-cost MRR sensor, derived the complex dielectric properties of the material based on the S21-parameters of this sensor, and extended the applicability of a density-independent algorithm, which together demonstrated the potential for accurate moisture measurement in rapeseed.

As a fundamental task in natural language processing, text classification plays an important role in various applications such as information retrieval, machine translation, and sentiment analysis. However, most deep models do not fully consider the rich information in the training instances during prediction, resulting in incomplete learned text features. To address this problem, this paper proposes the BERT-TextCNN model, which uses BERT to obtain word embeddings with sentence-level global features and uses the embeddings as inputs to TextCNN to capture local features, achieving high accuracy in text classification. To verify the superiority of the model, comparisons with several classic models were conducted, and the test results on the THUCNews dataset showed that the BERT-TextCNN model had improvements in accuracy, recall, and F1 score. Therefore, this model provides an effective solution for text classification tasks.

As employment pressure increases, university students face significant employment-related psychological stress. Music therapy is widely used to alleviate and adjust employment anxiety. However, accurately identifying and assessing the impact of music therapy on emotional states through electroencephalogram (EEG) sequence analysis poses a challenge due to the high dimensionality of EEG data and the complexity of time series. To address these issues, we propose an innovative Temporal Graph Convolutional Network (T-GCN) architecture that captures features in EEG data. Our method achieves a 93.52% Accuracy on the SEED dataset, validating the effectiveness of T-GCN and training strategy, and further supporting the feasibility of music therapy in relieving employment psychological barriers among college students, potentially informing personalized design for employment psychology adjustment and music therapy interventions.

Text sentiment analysis is a mainstream method in text information analysis, which could extract important data information from text, including customer needs and opinions on the product or service. Mainstream algorithms used in text sentiment analysis typically require large amounts of data to train models, making them less applicable in research areas with limited datasets, such as information technology (IT) services. Therefore, this paper proposes a text sentiment analysis algorithm specifically designed for small sample datasets—the Equal Sentiment Enhancement with Distribution (ESED) algorithm. The kernel of this algorithm is the improvement and innovation of the gradient descent algorithm in sentiment analysis with small sample datasets. This algorithm only requires that the number of mutually independent texts exceed the number of sentiment words, and it could achieve relatively accurate calculation of word sentiment values even under the condition of small sample datasets. The present study conducts simulation experiments to illustrate the convergence and usage condition of the ESED algorithm, demonstrating its feasibility in the application of text sentiment analysis with small sample datasets. To demonstrate the superiority of the ESED algorithm in text sentiment analysis with small sample datasets, it selects the problem of customers' purchase intention (CPI) prediction for IT services on freelance platforms for comparative experiments, comparing the performance of the ESED algorithm with four mainstream text sentiment analysis algorithms. Comparative experiment results show that compared to four baseline algorithms, the ESED algorithm achieves a decrease in Mean Squared Error (MSE) of predicting the CPI by 18.0%–51.5% in the dataset with 1196 samples. Conclusively, this paper contributes to the extended application of text-based sentiment analysis research in the field of research with small sample datasets and improves the prediction of CPI for IT services on freelance platforms.

The subtle differences between different kinds of fine-grained images make the task of fine-grained image recognition challenging. In this paper, we use DenseNet as the backbone network to do the following investigations in the image recognition task. We propose a multi-scale information fusion model based on attention mechanism for the fine-grained images task. The improved model uses the feature information of different scales generated at different network levels, and the information of each scale learns from each other, which improves the localization ability of the model. A multiscale information fusion fine-grained image recognition model based on contrast learning is proposed. The improved model introduces contrastive learning in the original training strategy, which constrains the model to acquire consistent common local information and complementary information, thus improves the model's ability to understand images and express semantics.

Spatial crowdsourcing demonstrates its efficient data collection capability and drives the development of numerous related applications, such as geographical environmental monitoring, food delivery, and traffic management. Task assignment is a core concern in spatial crowdsourcing, which requires a balance between efficiency (i.e., reducing the cost of assigned tasks) and effectiveness (achieving a greater number of tasks covered for execution). Therefore, we define the problem of optimal task quality in spatial crowdsourcing networks. In contrast to traditional task quality problems, we select only a subset of mobile workers to maximize the task quality with limited budgets. This problem is proven to be NP-hard. We first define the task quality function and prove that it is submodular and non-decreasing. By leveraging the properties of the function, we propose a (1-(1/e)) heuristic algorithm with a time complexity of 𝑂(n^k+2), where integer 𝑘≥3. Ultimately, experimental findings demonstrate that our method surpasses both random selection and a state-of-the-art technique in terms of overall quality.

It’s challenging to optimize offline handwritten Chinese recognition as a topic of pattern recognition as the consequence of the complex structures of Chinese characters. To solve this problem, many data pre-processing methods and neural network architectures have been optimized to improve the recognition accuracy. To this end, we propose a fully convolutional neural network architecture with text kernel for offline handwritten Chinese text recognition in this paper. To make the features of the image clearer, the text kernel which can align the characters of handwritten Chinese text is implemented. This novel fully convolutional neural network captures the feature maps in two branch and promote the efficiency of neural networks. Experimenting on the CASIA-HWDB 2.0-2.2 datasets, our proposed method achieves a competitive performance and the accuracy achieves 97.87% with language model.

The spread of Internet rumors is a key concern for information management researchers. In order to fully represent the main characteristics of the propagation process of network rumors, we introduced the factors that can accurately portray the influence of rumor propagation as parameters, and combined with the Logistic model, we constructed a network rumor propagation model. Numerical simulation of the model was carried out based on the Eulerian method, and the influence degree of each parameter was analyzed, which provided a theoretical basis for rumor propagation prediction and rumor governance.

The single-stage text-to-speech model has made significant advancements in recent years. A comparison between two prominent models, VITS and VITS2, reveals that training with the same individual's voice dataset led to the voiceprint model generated by VITS closely aligning with the original voiceprint; however, its segmentation and fluency were subpar. On the other hand, VITS2 showed the opposite characteristics. Through the implementation of Retrieval-based Speech (RVTS), involving the extraction of intonation and fragmented sentences from the VITS2 model followed by code utilization, we were able to achieve speech synthesis. Our experimental findings demonstrate that our approach surpasses both VITS and VITS2, leveraging the strengths of both models.

In intelligent unmanned ground vehicle systems, decision-making algorithms often face challenges in adapting to dynamically changing environments, and their generalization capabilities may be limited. Most existing decision-making algorithms can only achieve robust results in the original scene, but when transferred to new scenes under same task, the algorithm performance drops sharply. Moreover, when deploying decision-making algorithms to unmanned ground vehicles, they often struggle to achieve performance comparable to computer simulations. To tackle this challenge, this paper proposes Scene Semantic Reconstruction for Unmanned Ground Vehicle Virtual-Real Integration(S2RU). S2RU decomposes the scene into abstract entities with object semantic information and then combines these entities using compositional neural radiance fields to enhance the capabilities of the UGV agent. This means the decision-making process is divided into two stages. In the first stage, concrete entities in the original perceptual information are mapped to abstract entities and transformed into scene semantic maps. In the second stage, decisions are made based on scene semantic maps. We have validated in both simulation and real-world environments, showcasing robust transferability between these environments and enabling cross-scene transfer for the same task and validate the usability, completeness and stability of S2RU.Results demonstrate that our methods improve success rate of a particular task across different scenes by at least 20% compared to other virtual-real integration methods.

To address the challenge of recognizing specialized terminology in gas safety, we introduce a novel Named Entity Recognition (NER) model that integrates semantic similarity. Unlike traditional NER algorithms such as BERT+BiLSTM+CRF, this approach incorporates a word similarity weighting layer. This layer improves the recognition of entities related to Chinese urban gas pipeline standards by utilizing a proposed algorithm that maps inter-character similarity to vector weight values within the word embedding layer. Experimental results validate the model's superiority, exhibiting a notable improvement of 0.051 in the F1 metric.

In the task of cross-view image-based geolocation, many existing methods rely on coordinate transformations, which fail to expand the perception across dimensions and establish global correlations effectively. Moreover, these methods often require substantial computational resources, and the parameterization of models lacks effective control. This paper proposes a novel approach to cross-view geo-visual localization. Compared to existing methods, it achieves a retrieval accuracy of R@1 reaching 95.42% on the CVUSA dataset while reducing the parameter count by approximately 31.62% compared to TransGeo. Importantly, it does not require the assumption that the center positions of any query images are perfectly aligned. To expedite model inference and reduce parameter requirements, the algorithm's feature extraction backbone employs the PVT structure. Furthermore, it utilizes multi-scale feature maps for subsequent inference, which balances global and local features spatially. However, CNN methods tend to overlook long-range relationships between pixels or patches. Hence, the Mixer method is introduced, which utilizes spatial channel mixing algorithms to obtain more robust global image vector descriptors. Additionally, the unique flexibility of this module ensures long-term contextual relationships without the need for assumptions regarding feature alignment coordinate transformations.

With the increasing popularity of digital video applications, video restoration techniques have become increasingly important. This paper presents a flow-based video restoration method that aims to achieve high-quality video restoration by analyzing the spatio-temporal relationships between video frames. Specifically, we employ two key steps: flow preprocessing and inter-frame restoration network. In the flow preprocessing stage, frame differencing and optical flow estimation are utilized to obtain the appearance and motion features between frames. The inter-frame restoration network learns feature representations and restoration capabilities by compensating for flow estimation errors during the frame alignment process. We conducted extensive experimental evaluations on multiple video datasets. The experimental results demonstrate significant improvements in restoration quality, spatio-temporal coherence, and real-time performance.

The issue of blind spots in the field of vision during automobile driving has always been a significant concern. In response to the limitations of the stereo-matching algorithm based on census transforms, such as inadequate accuracy, insensitivity to textures, and excessive reliance on central grayscale values, an improved local stereo-matching algorithm based on census transform is proposed. The algorithm utilizes the maximum cross-domain variance method to calculate the matching window size, reducing mismatches in discontinuous disparity regions. The minimum cross-domain mean method is employed to determine the grayscale value of the central pixel, minimizing external influences on the central pixel. Additionally, a guided filter with edge-preserving characteristics is used for cost aggregation, further reducing mismatches in non-continuous domains. The experimental results illustrate that the introduced algorithm attains an average mismatch rate of 5.96% on the Middlebury benchmark dataset, marking a substantial enhancement when contrasted with the original census algorithm, which exhibited a mismatch rate of 16.2%.

The advantage of interactive segmentation is that it can be adjusted in real-time according to user requests and feedback, leading to more precise and expected segmentation results. This research provides a new interactive segmentation method to increase the accuracy of image segmentation. The proposed method uses deformable convolutions instead of traditional convolutions to better capture spatial variations in the target area. In addition, the Context Feature Calibration module is introduced to model various semantic context information for each pixel, thereby further enhancing the accuracy of segmentation results. The proposed method was verified using multiple public datasets. The results show that the method exhibits good performance. In summary, the proposed method offers a novel and effective solution to the interactive segmentation task.

The Doppler radar human action recognition methods based on deep learning theory have been developed rapidly. However, in practical applications, the offline batch training of traditional deep learning methods is difficult to adapt to the dynamic recognition needs of changing data. To solve the above problems, an incremental Doppler radar-based human action recognition class algorithm is proposed, allowing new action classes to be added gradually. Aiming at the problems of catastrophic forgetting and poor security of old class data storage, which are common in traditional incremental learning algorithms, a bias calibration method is proposed to compensate for the classifier's output bias on the old and new class data to inhibit the catastrophic forgetting effect; and a method that retains the training parameters of the generative adversarial network for generating the sample data of the old class, which avoids the risk of the leakage of the real data. Finally, the comparison experiments between the proposed algorithm and the traditional incremental learning algorithm for human action recognition are accomplished by using Doppler radar to collect the echo data of human actions, and the results verify the superior performance of the proposed method in learning new tasks while retaining the previous knowledge

When a single mobile robot is used in SLAM construction, the low efficiency of SLAM construction will occur due to the small number of mobile robots. When multi-mobile robots are used in SLAM construction, aiming at the two-dimensional raster local environment map generated by multiple mobile robots, this paper proposes an improved real-time fusion method of two-dimensional raster map based on feature point matching. The improved algorithm PROSAC is adopted to replace RANSAC algorithm and improve algorithm efficiency to eliminate mismatched feature points. Improve the efficiency of map fusion. First, two mobile robots were used to construct the local environment map by using the Gmapping mapping algorithm respectively. Then, the constructed two-dimensional raster map was processed in gray scale, and the feature points in the raster map after gray scale processing were extracted, filtered and matched, and finally the global map fusion was completed. Through gazebo simulation experiment, local maps constructed by multiple mobile robots were fused to verify the feasibility and effectiveness of the proposed method.

Natural Language Processing (NLP) is a growing area of artificial intelligence research with the goal of enabling computers to understand and interpret human language. In this study, we propose a new deep learning-based model that aims to improve machine understanding of natural language. We implemented a model that combines Recurrent Neural Networks (RNN) and Attention Mechanisms and tested it on multiple datasets, including sentiment analysis and text summarization. Experimental results show that our model outperforms existing techniques on several tasks.

Aiming at the problem that the use of Iterative Least Squares (ILS) algorithm and Extended Kalman filter (EKF) in the position solution of instantaneous Doppler localization for the localization of non-cooperative low-orbit communication satellites will bring errors in instantaneous Doppler localization, a further optimization method is adopted on the Improved cubature Kalman filter (ICKF) to achieve a higher accuracy of the position solution. The ICKF algorithm based on Interacting Multiple Model (IMM) is designed to model the influence of the ionosphere and other disturbing factors, reduce the influence of the estimation error of the ionosphere, and be able to adaptively estimate the noise; take the static positioning of the star chain satellite constellation for the ground station as an example, and verify the feasibility and validity of the algorithm by simulation through the simulation platform of STK software. The experimental results show that the accuracy of the ICKF algorithm based on Interacting Multiple Model (IMM) is significantly improved in the instantaneous Doppler localization solution compared with ICKF.

In most audio-visual speech separation models, audio and video features are generally directly merged. This approach may result in insufficient utilization of the interrelationships among cross-modal features. Therefore, joint attention is employed during the fusion of audio-visual features to achieve cross-modal integration. By jointly modeling both intra-modal and inter-modal relationships, features of each modality can focus on both another modality and themselves, thereby obtaining improved cross-modal fusion features to enhance speech separation performance.To evaluate the effectiveness of speech separation, an audio-visual bilingual Cantonese corpus (AVCC) was established, and the actual performance of three speech separation methods—Conv-TasNet, VisualVoice, and VisualVoice with joint cross-attention—was tested using Perceptual Evaluation of Speech Quality (PESQ), Short-Time Objective Intelligibility (STOI), and Signal-to-Distortion Ratio (SDR).The study indicates that when separating mixed speech of two speakers with different genders, the recommended VisualVoice method with joint cross-attention achieves an increase in SDR of at least 0.87dB, with the highest improvement reaching up to 1.99dB.The Joint Cross-Attention Model for feature fusion can utilize the correlations between cross-modalities more effectively, thereby enhancing the separation results.

Implementing table-based question answering can fully leverage the rich information in tabular data, and the development of large language models has provided strong natural language processing and understanding capabilities. Current research on table-based question answering with large language models mainly focuses on how to improve text-to-SQL abilities and the development of table pre-training models, lacking in studies on feasible technical approaches for applying general large language models to engineering applications involving large-scale database table question answering. In response, combining techniques such as prompt learning and chain of thought，this paper constructs a database question answering chain of thought based on large language model, proposes a table filtering scheme based on vectorized matching of table structural information, specifically addressing poor table filtering effects, and constructs a database optimization chain of thought in the context of de-identification scenarios. Experimental verification shows that after adopting the chain of thought construction and vector matching filtering scheme, the database question answering capability of the large model has been effectively improved.

In social platform sentiment analysis tasks, existing methods only consider sentiment features from a single pole and scale, which cannot fully exploit and utilize sentiment feature information, and the model performance is not satisfactory. To address this problem, a Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) model with multi-level and multi-scale feature extraction is proposed. The model first uses the pre-trained Chinese word vector model and combines the embedding layer fine-tuning to obtain word-level features; then uses the multi-scale phrase-level feature representation module and the sentence-level feature representation module to obtain phrase-level and sentence level features, respectively, and in the multi-scale phrase-level feature representation module, uses convolutional networks with different convolutional kernel sizes to obtain phrase-level features at different scales; finally uses the multi-level feature The word-level features, phrase-level features at different scales and sentence-level features are fused to form multi-level joint features, which have more sentiment information compared with unipolar and single-scale features. Finally, the reliability of the proposed method is verified by simulation tests of sentiment classification on social platforms. The experimental results show that both multi-level and multi-scale features methods exhibit higher classification performance under different word vector size samples compared with commonly used sentiment classification algorithms

Breast ultrasound (BUS) images are pivotal for early tumor detection, yet accurate segmentation faces challenges like noise and indistinct boundaries. Existing methods, such as multi-task learning and adaptive attention networks, often overlook morphological features intrinsic to malignant tumors, resulting in suboptimal performance. In response, our study introduces EGUnetr, a novel segmentation model with Edge Guidance strategy designed to address these limitations. EGUnetr incorporates an improved Transformer block using the Local-Global shifted window (LGWT), enhancing the encoder's ability to learn robust representations of malignant tumor samples in breast ultrasound images. Additionally, we develop an Edge Guidance Module to extract and fuse target boundary features with semantically rich information. This guides the decoder to generate more precise prediction maps, offering a detailed and accurate portrayal of breast tumor boundaries. Experimental results demonstrate the superior performance of EGUnetr in malignant tumor segmentation over existing medical image segmentation models.

Clustering ensemble can effectively improve the accuracy and robustness of clustering. The quality and diversity of base clustering are crucial to the clustering ensemble effect. However, traditional clustering ensemble algorithms usually treat a base clustering result as a whole, ignoring the quality differences within the same base clustering result. Clusters of different quality have completely different contributions to clustering ensemble results, and ignoring the quality differences of clusters can seriously affect the effectiveness of clustering ensemble. To solve this problem, this paper proposes a clustering ensemble method based on multiscale cluster reliability. In this method, multiscale global structural information is first mined by performing random walks on the cluster similarity graph to obtain multiscale correlation between clusters. Then, the multiscale cluster reliability is measured by using the multiscale correlation of clusters, combined with the information entropy, and the clusters are weighted accordingly. Lastly, the clustering results are obtained by graph segmentation. Experimental results on numerous real datasets show that compared with 10 typical and state-of-the-art clustering ensemble methods, the clustering ensemble method based on multiscale cluster reliability not only performs better in clustering effect but also has higher stability.

A vehicle's lane-changing behavior is affected by the surrounding environment and driver factors, which makes it difficult to identify accurately. To solve this problem, a personalized lane-changing decision model based on a Long Short-term Memory (LSTM) network is proposed. First, an unsupervised clustering method is applied to recognize three distinct driving styles; Second, by considering the interactions among the target vehicle and surrounding vehicles, a benefit function is constructed to measure these interactions and generate the lane-changing benefit values. The lane-changing gain values and feature parameters are used as model inputs to construct a personalized lane-changing decision model using LSTM. Finally, the proposed method is validated with the NGSIM dataset: the overall accuracy of the model reaches 97.8% when considering different driving styles, which proves that the proposed method can achieve personalized lane-changing decisions based on different drivers' lane-changing behaviors.

In traditional Monte Carlo (MC) path-tracing denoising approaches, uniform processing across all pixels often overlooks the variable importance of different image regions as perceived by human observers. This study introduces a novel denoising method tailored to 1spp (one sample per pixel) MC renderings, leveraging human visual perception to prioritize computation in visually salient areas. By classifying pixels based on visual saliency, our method efficiently allocates computational resources, enhancing quality in high-saliency regions while reducing unnecessary processing in less noticeable areas. Experimental results validate the effectiveness of our approach, demonstrating improved denoising performance with reduced computational overhead. This saliency-based strategy not only achieves high-quality denoising but also paves the way for more perception-driven approaches in real-time rendering applications.

During the rescue process of mine accidents, deviations in the drilling position or the occurrence of new local collapses may lead to unsatisfactory results using traditional life detection methods such as audiovisual means. Ultra-wideband radar life detection technology can penetrate coal to detect trapped people. However, due to the unclear attenuation characteristics of radar signal propagating in coal and the neglect of the differences in signal propagation between coal and air in traditional localization methods, significant positioning errors may occur. Therefore, this manuscript proposes the life detection and localization method suitable for mine rescue scenarios through studying the attenuation characteristics of radar signal propagating in coal. Experimental results show that the attenuation of radar waves is proportional to the thickness of the coal. The proposed method can detect trapped human behind coal in real-time, with an average relative error of 3.8% in estimation of the thickness of coal and 5.6% in localization, significantly improving localization accuracy compared to traditional methods. This study provides technical support for the application of UWB radar in mine drilling rescue, which is of scientific and practical significance for ensuring the safety of coal miners.

In order to solve the problem of cylinder fitting, this paper proposes a method to fit the initial value more accurately. Using the characteristics of the cylinder data collected by laser displacement sensor, the point cloud data is first fitted in plane, then the original data points are projected, and the section where the axis is located is found. Then, a point passing through the axis and the axis direction vector are found out. In this way, the method is used to find out Find the initial value of the cylinder. Then, the particle swarm optimization algorithm is used to iterate with the initial value, and then the 7 parameters of the cylinder are calculated accurately, and the cylinder equation of point cloud data is obtained. The method can obtain accurate data when the cylinder fitting is used to find the initial value, which makes it difficult to find the optimal particle swarm optimization. Compared with commercial software SA, the radius error is less than 0.05mm. The experiment shows that the accuracy of the method meets the requirements of the field.

With the comprehensive construction of smart campuses in universities, various departments have built a large number of information systems, generating a large amount of data. The complexity and variety of student data make it an urgent issue to analyze student behavior data from a large and diverse campus student population, explore potential abnormal student behavior, and provide early warning for abnormal behavior. This article uses K-means clustering algorithm to analyze student behavior data, and uses Pearson coefficient to analyze correlation. The experimental results show that students who spend more than 4 hours online for entertainment have a higher risk of failing their grades. By analyzing data such as online entertainment duration, book borrowing volume, consumption, and access control of students, we can profile their behavior, identify abnormal behaviors through data analysis, and provide early warning for campus management.

In order to solve the problem that the traditional point of interest recommendation algorithm will miss user characteristic information, a geographical location point of interest recommendation algorithm based on deep learning was studied. By combining the user's personal historical preference data with spatio-temporal information, taking into account the differences in personal long-term and short-term interests, the user check-in information is classified according to working days and rest days, and the results are calculated using a Voronoi diagram to calculate geographic similarity. Location point of interest recommendation model. This model is compared with algorithms such as RankGeoFM, Caser and NARM, and accuracy, recall and F1 value are selected as evaluation indicators. The experimental results show that this algorithm has obvious advantages in the same test set, under different K values. , are better than other algorithm models and are suitable for recommendation of points of interest in geographical locations.

In the field of autonomous driving, vision-based environmental perception methods usually use monocular or multi-ocular cameras to capture images, and use computer vision technology to analyze images to complete perceptual tasks such as object recognition, target detection and tracking, semantic segmentation, three-dimensional reconstruction, and distance measurement. Where there is a lack of prior knowledge or relational structure, the application of existing methods is limited. A method for calculating similarity matching between azimuth and polarity eigenvectors of components is proposed, and 5 defects of common components are accurately detected. The experimental results show that the proposed algorithm can quickly and accurately locate and identify PCB components, and effectively detect the defects of different components of the circuit board, because information is only passed between tightly connected nodes, this approach fails to capture novel interactions between distant nodes. The way to solve this limitation is to introduce an attention layer with induced sparsity to find. Therefore, based on specific projects in the laboratory, this paper conducts in-depth research on the visual environment perception for automatic driving in complex environments, aiming at the actual needs of intelligent unmanned units for automatic driving on complex off-road road conditions, and the problems such as low accuracy and insufficient robustness of existing algorithms.

With the rapid rise of online shopping platforms, the emergence of a large volume of complaints on these platforms poses a significant challenge in accurate text classification. However, existing research predominantly focuses on exploring superficial features of complaint information, with a primary emphasis on English language corpora, neglecting the research on the classification of Chinese complaint information. Moreover, early studies face limitations due to the scarcity of Chinese datasets for online platform complaints. To address these issues, in this paper, we first employed web scraping tools to dynamically collect complaints from online platforms, constructing a Chinese Complaint Text Dataset (CCD) specifically for online platform complaints, thereby facilitating subsequent research endeavors. Building upon this foundation, we propose an enhanced Chinese Complaint Text Classification Method (CTCD) based on pre-trained language models, which is designed to achieve effective model training with a minimal number of annotated samples. Specifically, we extracted keywords from the dataset and employ three distinct optimization strategies to filter the extracted vocabulary, enhancing the quality as the final Verbalizer extension. Experimental results demonstrate the superiority of our method over other baseline approaches on the Chinese online platform complaint dataset. The research aims to fill the research gap in the field of Chinese online platform complaint classification, presenting an effective approach to enhance classification performance.

By analyzing the interface model of SolidWorks software, the tolerance optimization method based on solidworks was studied. Through the optimization of the three-dimensional model size chain, a tolerance optimization allocation method that considers the quality loss function is formed. In the process of tolerance optimization, not only the price factors of the processed products are considered, but also their quality factors are integrated. By establishing a tolerance database, the experimental verification of the tolerance optimal allocation theory is achieved. Through comparison of experimental results, different parts processing costs will be obtained by using tolerance optimization methods of different models. Although the cost of the quality loss tolerance optimization method in this study is not the lowest, it is better than the cost-tolerance model in terms of product quality, making the product of higher quality and achieving higher-precision product production at a lower cost. Suitable for the development needs of enterprises.

Aiming at the problems of unsatisfactory cross-modal feature interaction in multimodal sentiment analysis and the influence of each modality on the final sentiment classification results. Through the research on multimodal sentiment analysis methods, a multimodal sentiment analysis model based on heterogeneous extracted feature network and bimodal interaction is constructed. The model firstly uses DPCNN network to capture the deep semantics of text, BiLSTM network to obtain the upper and lower dependencies of speech, and TCN network to extract the relevant features for facial images; in order to learn the fusion features between different modalities, the interaction information within and between modalities is captured by the bimodal interaction network layer; and then the attention mechanism is used to determine the attentional weights of the modalities in the final sentiment features, and to classify the influence of modalities on the final sentiment classification results. attentional weights and fuses the modalities; finally, the final sentiment classification is performed through the fully connected layer and multi-head attention. The model is experimented and evaluated on the datasets CMUMOSI and CMU-MOSEI, and the experimental results show that the model outperforms other models in the sentiment analysis problem compared to the existing models.

The study of FMCG recognition system based on deep learning is of great significance in improving the logistics efficiency of FMCG products and the speed of warehousing in and out of the warehouse. This paper analyzes the research and development status of image-based FMCG recognition system in China and other countries, and aims to design a FMCG recognition system based on deep learning, including fully connected neural network and convolutional neural network, which can meet the requirements of recognizing specific FMCG products after training datasets, and the recognition accuracy is in line with expectations. In this paper, after determining the overall structure of this FMCG recognition system, the fully connected neural network, convolutional neural network, pooling layer and pooling function were learned accordingly, and adopted the two optimization methods: Regularization and Dropout, to avoid the occurrence of overfitting as much as possible. For the test data, the images of five different types of FMCG products were crawled by crawler, and then the steps of data cleaning, sample expansion, resolution conversion, labeling, and storing as test.mat and train.mat files were carried out sequentially. Finally, the fully connected convolutional neural network was built, the datasets were read into it, and the training effect was tested after training to get the FMCG recognition accuracy. After that, it was necessary to compare the actual accuracy with the expected accuracy. If the expected accuracy was not reached, adjusted the parameters in the fully connected convolutional neural network and repeated the training and testing process until the final test accuracy met the requirements. In summary, this research and design work on deep learning based FMCG recognition system embodied in this paper has been successfully completed. The features of this FMCG recognition system: light weight, easy to maintain and port; the recognition accuracy for a given FMCG picture is high, and the performance requirements for the running system are low.

To achieve automatic grading of potting seedlings, improve performance, and reduce complexity, we collected and constructed a grading dataset using tomato potting seedlings as the test subject. We selected the YOLOv5n target detection model as the baseline model. We reduced the number of model parameters through lightweight improvements, integrated the similarity-based attention mechanism into the backbone network, enhanced the accuracy of potting seedling feature recognition by improving the CIOU loss function to EIOU, and performed global channel pruning on the improved model to further reduce its complexity. Experiments demonstrate the following results: the final model achieves a recall of 92.1%, an average precision mean of 94.9%, has 0.9×106M parameters, performs 2.1G floating-point operations, has a model weight size of 2.2M, and achieves a detection speed of 130 frames per second. Deployment and testing on edge devices confirm that the model achieves low computational requirements, has a small parameter count, maintains fast and accurate performance, and can be used for real-time classification of potting seedlings.

With the rapid advancement of digital technology, such as big data, artificial intelligence, cloud computing, etc., these technologies provide unprecedented possibilities for enterprises. Enterprises realize that in order to establish themselves in a fiercely competitive market, they must leverage digital technology to enhance the efficiency of cultivating skilled talents. The cultivation of employee presentation ability is an important guarantee for improving work efficiency. In this study, an Enterprise Technical Talent Training System, E3T, was developed to improve technical talents’ body and vocal expressions. Using the real-time feedback functions, E3T can detect undesirable states, provide instructions to presenters, and help them achieve the desired states. Among the basic operational skills, the practice support functions were implemented to improve face direction, body direction, and voice volume. By comparing the vocal feedback function with the visual feedback function, the results show that vocal feedback is more effective than visual feedback in improving face direction and voice volume.

In order to establish a unified data source for ship design, effectively integrate 3D design into each stage of the design process, and achieve lightweight models while systematically organizing modelling knowledge, this article primarily conducts research on automated construction of a knowledge graph for multi-resolution 3D ship design. Firstly, the concept and process of multi-resolution modelling are proposed, with detailed explanation provided through an example focusing on hull structure. Subsequently, the technology for automatic construction of a knowledge graph for multi-resolution 3D ship design is discussed. The vertical domain-specific knowledge graph is automatically constructed using a deep neural network based on BERT-LSTM-Softmax. Finally, the effectiveness of the proposed method is validated by utilizing the Neo4j graph database storage to represent and query the constructed knowledge graph. The resulting knowledge graph facilitates acquisition of 3D modelling expertise by design and construction personnel, enhances their capabilities, and improves overall design quality

With the rapid development of computer vision and machine learning technology, the extraction and classification of image features has become one of the core tasks in these fields. Traditionally, classification tasks primarily relay on convolution neural networks, which are very effective at capturing local features of images. However, there are limitations when dealing with the global and complex relationships of images. In this work, we propose a novel transformer model to dispose the features extraction and classification task. Initially, the model explores how to apply the transformer model to the extraction of image features, including methods for segmentation and encoding of images. Additionally, the model can identify and understand the different parts and attributes of the image, which is essential for subsequent classification tasks. Subsequently, the model uses the extracted features to distinguish between different classes of images. The study demonstrates the classification performance of the model in a variety of complex scenarios, including those that require a detailed understanding of image content. From our extensive experiments analysis, we can conclude that our proposed model can achieve the features extraction and outperform classification accuracy with existing models.

In order to reduce the impact of point cloud stitching errors in ground mounted 3D scanners, a unique scanning method was proposed to distinguish static and dynamic scanning objects from overlapping ones by optimizing the scanning method. In order to reduce stitching errors, dynamic objects were removed and only static objects were stitched. Correction experiments were conducted on the data results using several methods, A comparison was made between the concatenated error data before and after using the correction method and without using the correction. The results showed that the corrected point cloud concatenation was smoother and smoother, and the measurement error of the data was significantly reduced.

Space laser communication network plays an important role in the future deep space exploration, which has giant strategic value in global communications. Deep space exploration missions such as space gravitational wave detection, satellite gravity measurement extremely depend on ultra-long distance and ultra-high precision inter satellite laser interferometry. All of those are based on the establishment of satellite-to-ground and Geostationary Orbit (GEO)- Low Earth Orbit (LEO) – Middle Earth Orbit (MEO) high-speed laser communication link, which needs to maintain high speed, high precision and high sensitivity. However, that is an enormous challenge in low signal-to-noise ratio(SNR), high-speed movement, high-precision tracking, etc. Besides that accurate ephemeris prediction are required for establishing and maintaining intersatellite laser communication links, the use of interferometry to measure inter-satellite distances and angles is inevitable. However, that is also an enormous challenge in the aspects of cost and technical implementation. This paper presents a scenario design combined the accurate ephemeris prediction and compressed interferometry measurement to resolve the situation. Moreover, this paper proposes a method of compressed sensing interferometry technology to take advantage of high-throughput, sub-sampling, and super-resolution reconstruction, which can achieve low data rates to meet high-speed and high-precision tracking requirements at low cost. The effectiveness of this method is verified through the simulation experiment.

The automatic identification of grassland plants based on UAV remote sensing imagery is crucial for the management and conservation of grassland ecosystems. Grassland plants, characterized by small individual sizes, complex backgrounds, and dense distribution, pose a challenge for UAV image identification. An improved YOLOv8 algorithm is proposed in this study to enhance the detection effect of grassland plants from different angles and optimize the detection of small targets in complex backgrounds. Firstly, a rotation-invariant mechanism is introduced into the backbone to improve the model’s adaptation to grassland plant targets from various angles, thereby improving detection accuracy and robustness. Secondly, a self-attention mechanism is applied in the backbone to enhance the model’s understanding capability of correlations between different plants. Additionally, a multi-level feature extraction structure based on dilated convolution is employed in the neck to further enhance the model’s feature representation capability. Finally, the SlideLoss loss function is introduced to solve the problems of insufficient sample size and foreground-background detection impact. Experimental results demonstrate that the improved algorithm achieves remarkable results in the automatic identification of grassland plants in Inner Mongolia, validating the application prospects of UAV remote sensing technology in grassland plant species monitoring.

In order to realize the real-time monitoring and timely response control of Power Internet of Things (PIoT) equipment operating status, this paper uses Finite Blocklength Coding (FBC) to meet the requirements of time-sensitive business. However, the traditional delay performance is not a good representation of the timeliness of state update. Therefore, it is very necessary to study the Age of Information (AoI) of the state update system. Firstly, based on the moment generating function of stochastic network calculus, the approximate expression of AoI violation probability is derived. Then, to minimize the information age violation probability, an alternate iteration algorithm is used to jointly optimize the precoding matrix and channel block length. Simulation results show that the proposed algorithm can effectively reduce the information age violation probability.

Traditional book recommendation algorithms only consider external rating data when handling book recommendations and face issues such as cold start for items. This study proposes a book recommendation algorithm that integrates Knowledge Graphs with collaborative filtering. By constructing a book Knowledge Graph and introducing semantic information between books as a basis for recommendation, this approach uses the knowledge representation learning model TransE to map the book Knowledge Graph into a low-dimensional continuous vector space. This mapping captures the relationships between book entities and calculates the semantic similarity between books. Then, by combining this with user behavior similarity derived from collaborative filtering algorithms through linear fusion, a recommendation list is generated. Experimental results show that compared to baseline algorithms, this algorithm improves in terms of accuracy, recall, and F-measure.

Vision-based object detection in intelligent vehicles presents unique challenges distinct from regular object detection tasks. It requires identifying a critical attention region that significantly influences the vehicle's driving behavior, determined by the ego-vehicle's driving state and the behavior of surrounding traffic participants. Traditional approaches process the entire image uniformly, leading to unnecessary computational overhead. This study introduces a novel approach to precisely and efficiently estimate the attention region, enhancing the detection of crucial objects within it and overall detection performance. Firstly, the method proposes an attention region estimation technique. This technique treats the computation of the attention region as calculating a mass center, considering the ego-vehicle's driving state and nearby traffic participants. Subsequently, the study proposes a cascade region enhancement detection strategy. Utilizing the estimated attention region, this strategy adaptively alternates between lightweight and heavyweight models for the detection tasks, optimizing performance and resource usage. To validate the proposed method's effectiveness, a real-vehicle dataset is compiled, featuring critical objects annotated by drivers. Comparative results demonstrate that the proposed approach significantly boosts efficiency while maintaining high accuracy in detecting critical objects, surpassing the performance of baseline models.

Fine-grained classification in high-resolution remotely sensed ship images is a challenging problem. This is because the differences between their classes are very subtle and highly similar, but the intra-class differences are large and there are too few publicly available datasets in the field, in addition to being susceptible to factors such as low light intensity at sea and uneven illumination, which reduces the accuracy of the classification. In order to solve the above problems, the article proposes a remote sensing ship image classification method that combines the improved ConvNeXt network with the idea of contrast learning. Firstly, the ConvNeXt network incorporates the SimAM attention mechanism to strengthen the weight of the ship in the extracted features and suppress the background interference weight, so as to effectively strengthen the network’s ability to mine the discriminative features of the ship under different lighting conditions. Secondly, after the feature extraction backbone network, introduce the ”separation + aggregation” module, which separately forces all image features to be separated and aggregated into each of their own fine grain categories, thus making the final classification easier. The method is validated on the dataset FGSC-23 and achieves an average classification accuracy of 90.72% and an overall accuracy of 90.96% on the test set. The experimental results show that the proposed method is superior to other existing models, effectively improves the accuracy of ship fine-grained classification and has good generalization ability.

With the popularity of social media and online comments, people chat and comment more frequently on social media. Due to the sparse and high-dimensional characteristics of the massive Chinese text itself, its semantic interpretation is also diverse and strong context-dependent, so accurate analysis of user emotion has become very important.Aiming at the problems of poor performance of traditional language models in dealing with long-distance dependencies and inadequate capture of significant features of text by classification models, a sentiment analysis model based on BERT-BiLSTMDPCNN is proposed. The model uses BERT pre-trained language model for text representation, BiLSTM sequence modeling and DPCNN hierarchical feature extraction to get the emotional polarity of the comment text. The experiment was carried out on the data set of comment text crawled from platforms such as Bilibili and Weibo. The experimental results show that the BERT-BiLSTM-DPCNN model has improved significantly, and the classification accuracy is 96.12%, which is better than other baseline models. It shows the effectiveness of the BERT-BiLSTM-DPCNN text sentiment analysis model.

To improve the accuracy of medical information feedback, a 360 degree feedback evaluation model based on ontology mapping and information island elimination was designed to address the issues of large data volume and external feedback. Using ontology mapping to define global and local ontologies, establishing mapping relationships, eliminating information silos, and integrating and processing medical information. Integrate evaluation tools, design a 360 degree feedback evaluation model, and verify the reliability coefficient to meet the standard. The experiment shows that the model can complete feedback within 10ms, with an evaluation effectiveness of 7.4. The overall integration has been improved, achieving fast and effective feedback and evaluation of medical information.

With the continuous expansion of university campuses, campus navigation has become crucial for every student. Traditional campus navigation systems face challenges in simultaneously handling indoor and outdoor navigation. Moreover, these systems typically rely on 2D navigation maps, requiring users to integrate real-world surroundings with the map, leading to increased complexity. In response to these challenges, we propose an Augmented Reality (AR) campus navigation system based upon ARCore. This system leverages ARCore to present 3D information in real-world scenes, enhancing the user experience. Additionally, we incorporate indoor ranging algorithms for real-time positioning and map generation on mobile devices, specifically tailored for indoor navigation. The development is carried out using Unity3D scripts to enhance usability and adaptability. Experiments conducted at XXX University demonstrate that our designed campus navigation system outperforms outdoor positioning accuracy compared to navigation systems by Amap and Baidu. Furthermore, it excels in indoor navigation capabilities, seamlessly integrating 3D information with the real environment, resulting in an improved user experience. Through evaluations with multiple groups of students, our system demonstrates superior functionality, usability, and adaptability.

Autonomous collision avoidance is vital for intelligent ship navigation. To improve the adaptability and effectiveness of collision avoidance policies, we propose a method that integrates meta-learning with reinforcement learning. Drawing inspiration from meta-learning, we developed a two-layered recurrent model to boost the adaptability and effectiveness of these policies. We then introduced a ship motion model considering wind, waves, and currents to better adapt to environmental changes. Subsequently, we designed a noise-enhanced autonomous ship collision avoidance reinforcement learning model to augment the exploration capabilities of the vessel agents. Lastly, simulation experiments were conducted to validate the feasibility of our approach. The results demonstrate that our collision avoidance policies surpass various comparative methods, showing superior effectiveness and adaptability. Overall, our innovative method offers a safer solution for enhancing intelligent ship navigation.

In order to solve the problem of low recall rate of conventional heterogeneous information aggregation processing methods and improve the effect of heterogeneous information processing, this paper proposes a heterogeneous information aggregation processing method based on cloud computing. Build a cloud computing platform architecture, on this basis, build a fuzzy membership function between heterogeneous information of scientific research topics, and identify multisource target features of heterogeneous information of scientific research topics; Transform heterogeneous information into paths of different types of nodes, construct an objective function, and obtain the results of heterogeneous information aggregation of scientific research topics. The experimental results show that the recall rate of this method is the best on most information sample sets, and the aggregation error rate is the lowest when the heterogeneous information of scientific research topics is aggregated, which has good aggregation performance.

Recently, Vehicle to Everything (V2X) technique has drawn great attention in various industries. Unlike traditional wireless communication systems, malicious attacks against V2X network may not only cause user privacy disclosure, but also may pose a direct threat to the personal safety of passengers. Radio frequency (RF) fingerprinting is a physical-layer based technique that has proved effective in device identification and has been widely used in many communication systems. However, it is still tough work to conduct RF fingerprinting in a V2X scenario due to its bad channel condition. This paper has proposed a channel insensitive radio frequency fingerprint identification (RFFI) scheme achieved by eliminating channel interference while retaining channel insensitive RF features. Signal pre-processing is firstly carried out including precise synchronization and carrier frequency offset (CFO) estimation. Then a channel insensitive feature extraction algorithm (CIFEA) which can effectively eliminate channel interference without channel estimation has been designed to obtain stable RF fingerprints under time-varying channels. Additionally, stacking algorithm has been adopted to further enhance performance of our system. Experimental results with practical Long Term Evolution (LTE)-V2X dataset demonstrate th13at the proposed scheme can achieve high identification accuracy in practical V2X systems.

Recent advancements in large-scale language models, as exemplified by ChatGPT, have undergone rapid and substantial development. Their efficacy in natural language processing (NLP) and related domains has notably surpassed that of traditional models. The pivotal role of cloud computing technology in supporting these advancements cannot be overstated. In this paper, we propose an effective and scalable 3D parallel optimization method for large-scale models, leveraging cloud computing capabilities to combine data, pipelines, and tensor slice-based parallelism, where the amalgamation of tensor slicing and pipeline parallelism operates with optimal efficiency in targeted areas. Specially, we curate a highquality natural language training corpus comprising hundreds of billions of tags and collaboratively develop training methodologies to enhance optimization efficiency and stability. Furthermore, a data parallelization strategy is proposed to expedite the training of large models, utilizing specialized hardware and software tailored for deep learning to augment training speed. To further reduce training time delays, we advocate for the use of more efficient optimization algorithms. This comprehensive approach addresses the intricate technical requisites of large language models in cloud computing, adapting to the burgeoning resource demands anticipated in future AIGC applications, thereby reinforcing the advancement of artificial intelligence. Experimentation has underscored the effectiveness and potential of our system.

In recent years, due to the influence of the post-epidemic era, some young people have mild anxiety, emotional depression and other problems^[1]. Many experts and scholars at home and abroad have proposed multi-modal emotion analysis methods. Based on this, we propose an OpencV-based emotion regulation method. We use Python language as a programming language, based on the related technologies of OpenCV module, PyQt UI interface technology, face recognition and speech recognition technology, and the use of multi-threading technology in the entire program, to avoid the occurrence of program no response phenomenon, to ensure the smooth operation of the program to the greatest extent. The facial expression data is collected through the camera, and the audio data is collected through the microphone and uploaded to the program. When the bad mood is detected, the user's mood is adjusted through language comfort, appropriate music, appropriate lighting and other ways to maintain the user's emotional stability. Or detect that the user is currently happy, but also through music and other means to enhance the user's interest in life. The product contains core devices such as cameras, microphones, and Raspberry PI development boards, so that the current flows in each component along the expected route to complete the modulation, demodulation and coding functions. The method proposed in this paper can effectively improve the emotion recognition accuracy, and the accuracy rate reaches 87.49% in the multi-class emotion recognition tasks in the open data set, which has a stable recognition effect

Image segmentation is an important branch of computer vision, which has been widely used in many research fields. In order to understand the research status in the field of image segmentation, this paper systematically analyses and compares the traditional image segmentation method and the image segmentation method based on deep learning. Firstly, the development, advantages, and disadvantages of traditional image segmentation methods are summarized according to the three types of segmentation methods based on threshold, edge, and region. Then, the segmentation network models of three commonly used image segmentation methods based on deep learning, namely, fully convolutional neural network, pyramid scene analysis network, and Mask R-CNN, are analyzed. Based on the previous experimental results and the current research progress based on deep learning models, the problems of contour edge optimization and multi-level pixel information fusion in the field of image segmentation are summarized. At the same time, improvement ideas are put forward for the existing problems. Finally, the improvement and development of future image segmentation methods are prospected, which provides a reference for the subsequent research of segmentation methods.

With the continuous development of intelligent mine construction, researchers are applying knowledge graph technology to further realize the intellectualization of coal mine. However, in the research field of coal mine safety equipment, there are little organization and construction of relevant knowledge, resulting in fragmentation, redundancy and limited localization of knowledge in this field. Based on this, this paper constructs knowledge graph of coal mine safety equipment to expand knowledge construction in this field. Firstly, self-built CMSE-EntityRel dataset. Secondly, entity relationship joint extraction is carried out based on CasREL-Mine framework. Finally, we use the graph database Neo4j to store and visually display entity and relational knowledge triples. At this stage, we construct the knowledge graph of coal mine safety equipment field. The method adopted in this paper can effectively realize the knowledge construction and organization in the field of coal mine safety equipment, and provides data support for intelligent management level of coal mine safety production.

Pharmacovigilance is an important aspect of healthcare aimed at reducing adverse events and drug interactions in patients due to drug use. With the continuous progress of medical technology and the wide application of drug therapy, the research of pharmacovigilance has attracted much attention. This paper utilizes Kleinberg algorithm and LLR algorithm to explore the development trend and research hotspots in the field of pharmacovigilance in China. Using CNKI as the data source and "pharmacovigilance" as the theme, we analyzed the relevant literature on pharmacovigilance from 1997 to 2023, revealing the hotspots and directions of current research. The upward trend in the number of articles can strengthen the communication and resource sharing among research institutions, and the hot topics in the field of pharmacovigilance, such as clinical trials, rational use of medication, risk management, adverse reactions, and traditional Chinese medicine, have emerged. At the same time, this study puts forward relevant suggestions for China's pharmacovigilance field, suggesting optimizing the monitoring of adverse reactions, strengthening pharmacovigilance risk management and promoting the development of the pharmacovigilance research field.

Under the background of digital transformation, all kinds of personnel often carry out data analysis and data product development through remote terminals. Once the end user has obtained the data access rights, the middle platform cannot limit the end user's data viewing and copying behavior. However, the current protection measures can only achieve static access control and internal circulation protection, and cannot extend the control and protection of user access behavior. In this paper, a dynamic data access control model supporting multi-link transfer and multi-level inheritance is proposed, and the technical characteristics and model framework of the model are introduced in detail. At the same time, the flow of access control, the dynamic change of permissions, the transfer and inheritance of permissions under tree rules are given in a specific scenario. Finally, the experimental analysis of the model proves that the model can realize the dynamic security protection and fine-grained access control of terminal data.

In highly sparse reward composite tasks, agents often face a lack of reward feedback within fixed time steps, leading to getting trapped in local optima and compromising their ability to effectively explore superior strategies. Skill learning is one approach to increase the density of reward signals, enabling adaptation to multi-stage tasks and expediting the learning process. However, contemporary methods for skill acquisition heavily rely on online asynchronous training. Although certain intrinsic motivation approaches excel at addressing sparse reward challenges, they suffer from issues of low sampling efficiency and limited interpretability of skills. These challenges hinder the speed of model learning and severely impede the reusability of skill policies. In this study, we employ expert demonstration data to facilitate the learning of skill policies, aiming to accelerate the convergence of the model while increasing the utilization of sample data. Subsequently, we engage in interactive learning with the environment. Additionally, we define an evaluation criterion for skill redundancy to encourage the selection of the most cost-effective skill strategy among similar skill policies that manifest between initial and final states. This process helps the agent efficiently and effectively accomplish complex tasks. Our objective is to minimize ineffective and redundant exploration and learning during the skill acquisition process. We evaluate our approach in the simulated UGV-Pyramid and simulated UGV-Hallway tasks, both implemented using Unity3D modeling. The results demonstrate the superiority of our algorithm compared to previous skill learning methods.

The detection and estimation of gas hydrates have always posed challenges for geophysicists due to the multiplicity of seismic inversion solutions. In this abstract, we propose a combination of rock-physics modelling, structure-guided multiparameter full waveform inversion (FWI), and seismic inversion techniques to detect and estimate gas hydrates. We develop a rock physical model specifically for gas hydrates in three different occurrence states, tailored to the target area in the Qiongdongnan Basin. And we employ a structure-guided velocity-Q multi-parameter FWI approach, which takes into account the geological structure of the area, to obtain high-precision and high-resolution velocity and Q. Furthermore, we use poststack and amplitude versus offset (AVO) inversion techniques to estimate additional elastic parameters, providing further insights into the subsurface properties. By applying these methods to the detection and estimation of gas hydrates in the Qiongdongnan Basin, we are able to obtain relatively reliable results. These results can be used to determine potential targets for gas hydrate production. Our integrated approach combining rock physics modelling, structure-guided multi-parameter FWI, and seismic inversion techniques proves to be effective in detecting and estimating gas hydrates. This has significant implications for the exploration and exploitation of gas hydrate resources.

The aim of this study is to construct a network literature acceptance evaluation model based on data mining technology, deeply explore the reading behavior characteristics, preferences, and interaction patterns of readers, and effectively evaluate the acceptance effect of literary works and optimize content creation and promotion strategies. The core methodology includes utilizing the FP Growth algorithm to mine frequent data items and improving the processing efficiency and accuracy of the evaluation model through the PFP parallel computing system. This study not only provides a new methodology for the quantitative analysis of online literature, but also provides a scientific basis for the quality evaluation and cultural value dissemination of literary works. Through algorithm design and simulation testing, this article deeply explores the potential and challenges of data mining technology in the field of online literature evaluation, providing theoretical support and practical guidance for the healthy development and innovation of online literature.

With the rapid development of Internet technology, traditional industries such as university have made remarkable achievements in the process of informatization transformation of their own business, and informatization has been widely applied in the teaching and educational affairs, teaching resources, student information and other business fields of university. The management and operation of data plays an important role in university information system. When the database table data changes, the university information system may need to synchronize the cache according to the changed data to ensure the consistency between the database and the cache; There may be a need for real-time acquisition of change data from different systems for processing among various information systems in university. However, the traditional industries such as university usually lags behind the industry in the use of information technology, and there is often no good way for various business processing of data changed in database tables. Aiming at the above problems, this paper designs and implements a database table change data processing system based on configurable templates. The system can capture the change data of the subscribed database table in real time, and perform the automatic setting operation of the change data according to the template, or send the change data to the message queue for the required information system to consume. This template-based change data processing method effectively reduces the coupling of system functions, and the automatic operation and visual configuration also improve the efficiency of university information system developers.

Data-to-text task aims to convert structured data into natural language text. However, the lack of large parallel data is a major practical problem for many domains in data-to-text generation. To address these issues, this paper proposes a method based on dual learning. Dual learning simultaneously learns two mutually dual generator and extractor, where the generator is responsible for text generation and the extractor for information extraction. Through dual learning, we can effectively utilize the interrelationships between unaligned data, thereby improving the performance of the generation model. We conduct experiments on an advertising datasets, and compare with traditional generation models. Experimental results demonstrate that the dual learning-based method achieves nearly the same performance as fully supervised approaches for the data-to-text generation task validating its effectiveness and feasibility in data-to-text generation tasks.

Multi-sensor data fusion technology plays a crucial role in practical applications. Dempster-Schafer's evidence theory is widely employed in the field of information fusion due to its flexibility and effectiveness in modeling and processing uncertain information. To better quantify the differences among different pieces of evidence, this white paper presents illustrative examples within the framework of Dempster-Shafer's proof theory, utilizing fractal entropy as an intuitive indicator of changes in information. We propose a technique for measuring sine distance that utilizes maximum belief entropy to partition the mass function, allowing for one-time preprocessing using fractal entropy division. Subsequently, we utilize the cosine function to analyze disparities in the mass function post-preprocessing. Furthermore, this paper elucidates several characteristics of our novel distance measurement method and analyzes its advantages through numerical examples. Finally, we apply the proposed measurement approach to information fusion and evaluate its efficacy.

Signal acquisition system is widely used in various fields of reality, due to the data acquisition rate and processor performance enhancement as well as the development of sensor arrays and other technologies, the data acquisition system acquisition and transmission of the signal is exponential growth, in order to solve the high-speed signal acquisition process of the large amount of data, real-time, transmission rate and other issues, the study of high-speed acquisition of signals as well as the transmission program has an important practical significance. FPGA is very suitable to be applied in high-speed signal acquisition system because of its characteristics. In this paper, we firstly introduce the composition of a typical highspeed signal acquisition system based on FPGA, and then briefly introduce the function and development process of each module of the system and introduce the mainstream programs used in several key modules. In engineering applications, different programs should be used in different modules in conjunction with the actual.

In this paper, based on the multi-objective optimization framework, we use genetic algorithms to solve the multi-objective GCICA model to explore the commonality of group functional connectivity among group subjects (multi-subjects), and propose a group analysis method named GAGCICA, which is very effective in solving the deficiencies of the previous group analysis methods while maintaining the superior group functional network detection capability. The experimental results show the superior performance of brain network retesting reproduction at the group level obtained from the modal resting state data. It is demonstrated that the genetic algorithm has better signal recovery performance in solving the multi-objective group ICA problem and better reflects the commonality of subjects in the group.

This paper aims to address the issue of personalized content recommendation in remote education platforms, with the goal of enhancing the learning experience through precise management of the educational database. The article begins by introducing web personalization techniques, including content, rule-based, and collaborative filtering algorithms for recommendation. Subsequently, the paper expands on personalized recommendation algorithms tailored to specific needs in remote education, providing a detailed description of how these algorithms utilize user behavior analysis to accurately capture user interests and deliver educational content accordingly. The paper thoroughly describes the design and implementation process of the data collection system, encompassing the technical framework for collecting crucial user learning behavior data, as well as the development environment implemented using embedded Linux and SQLite ODBC interface. Furthermore, comprehensive testing is conducted to ensure the performance and stability of the data management system in high concurrent learning environments. Overall, the research findings not only contribute to the design of efficient database management systems to support remote education recommendation algorithms, but also offer valuable insights for enhancing learner interactivity and personalized experiences.

SOPC(System-on-a-Programmable-Chip) is a System-on-Chip solution based on FPGA (Field-Programmable Gate Array). SOPC integrates various digital logic modules such as microprocessors, controllers, memory, and IO interfaces onto a single FPGA chip. Leveraging the programmable nature of FPGA chips, SOPC achieves programmability of the entire system. High-power laser aging, also known as laser reliability testing, involves assessing the longevity of a laser through continuous monitoring of parameters such as optical power, operating current, operating voltage, surface temperature, etc., under constant current conditions over a specified duration. This process helps to determine whether the laser meets specified longevity standards. This paper presents a data acquisition card SOPC system utilizing domestically manufactured FPGA as the main control chip. This system is capable of controlling and organizing raw data collected from lasers, performing calculations, and then transmitting processed data to a host computer. It features coordinated operation between software and hardware, fast processing speed, and high reliability.

The paper focuses on applying deep learning methods to analyze rock mechanical parameters in the underground caverns of the Kala Hydropower Station, aiming to develop an efficient methodology using the scikit-learn MLP regression algorithm to estimate rock physical parameters based on observed displacements or deformations, with the evaluation of its performance compared to traditional methods contributing to advancing parameter estimation techniques in geotechnical engineering. This novel method underscores the significance of understanding rock mechanical parameters for designing and constructing underground caverns, focusing on rock strength, deformation characteristics, and stress distribution, exploring the use of deep learning techniques, particularly the MLP regressor, to estimate these rock mechanical parameters, aiming to develop a methodology for accurate parameter estimation and back analysis, emphasizing the potential of deep learning in advancing geotechnical engineering practices and solving complex problems in rock mechanics. The comparison between the calculated displacement increments and the measured values from the 3DEC warning prediction model indicates some level of discrepancy between the predicted and actual values. Despite this, the predicted warning values demonstrate satisfactory control standards and carry important guiding implications. The derived rock mass parameters can serve as the fundamental basis for future warning predictions.

With the promulgation of the equal protection 2.0 standard, there are more and more application systems involved in the evaluation of network security level protection, and the items and data of on-site evaluation are more and more complicated. However, in the traditional evaluation mode, it is difficult for manual operation to effectively use the accumulated data information, and it is easy to interfere with the final evaluation results, which is difficult to meet the actual needs of the current network security level protection evaluation work. In this regard, this paper will carry out in-depth research on the above problems, and put forward a new evaluation scheme of grade protection based on data mining technology, which will provide necessary data support for evaluation decision guidance. Practice has proved that using Apriori algorithm to analyze the relevance of non-conforming evaluation items can effectively improve the execution efficiency of evaluation work and reduce human error. By clustering the historical evaluation data with K-means algorithm, the evaluation results can be classified, and then it is convenient for evaluators to put forward corresponding security strategies to ensure the network security of each information system

Conventional talent selection algorithms for large enterprises mainly use induction and deduction to obtain selection competency characteristics, which is vulnerable to the influence of orientation consciousness, resulting in low matching degree of talent selection. Therefore, a new talent selection algorithm for large enterprises needs to be designed based on entropy weighted clustering mining. That is to say, based on entropy weighted clustering, a model for selecting talents in large enterprises is constructed, and an optimization algorithm for selecting talents in large enterprises is generated. The case analysis results show that the selection algorithm designed for the selection of outstanding talents in large enterprises has good selection effect, high selection matching degree, reliability and certain application value, and has made certain contributions to improving the comprehensive strength of Chinese enterprises and promoting the sustainable development of China.

This paper explores the significance and utility of optimization strategy modeling within the context of operational research, emphasizing its value in improving decision-making processes across various domains. By integrating data analysis and linear programming, we present a novel approach to optimizing operational efficiencies, focusing on generic methodologies. Our research contributes to the field by introducing a comprehensive framework that leverages correlation analysis, polynomial regression, decision tree modeling, and ARIMA time series forecasting. The model's versatility and effectiveness in addressing complex optimization problems are highlighted, offering insights into its potential for wider application and the importance of predictive modeling in strategic planning. This study not only enriches the academic discourse on optimization strategies but also provides a practical tool for enhancing operational performance, paving the way for future innovations in the field.

In order to improve the efficiency of bridge inspection as well as the realization of information technology, a number of domestic companies have developed a bridge inspection software based on standard specifications, which can realize data standardization, management standardization, one-click report, etc., the inspection efficiency is greatly improved, and the inspection data also has a continuity. However, the current inspection software still has the problems of poor guidance, low degree of intelligence, etc. In the face of a large amount of inspection data, it can not be deeply mined and applied. Therefore, this paper is based on the vector method and collaborative filtering method, in-depth analysis of a large number of actual engineering inspection data, the construction of highway bridge review guideline algorithm model, access to the intelligent detection system, which can solve the problem of leakage and under-inspection. The following conclusions can be drawn: the adoption rate of the review guideline of the relevant diseases reaches more than 70%, which can better guide the inspectors to review the missed diseases; verified by the actual engineering data, the difference rate of the technical condition is within 10%, which indicates that the review guideline of the technical condition is more effective.

This work investigates the multicast routing problem for single-source networks, proposing a novel solution that specifically considers multicast routing with delay constraints. A new algorithm is introduced to address this issue, and its performance is evaluated through experimental simulations. Additionally, the presented algorithm is compared concerning performance with the classic heuristic algorithms, namely the KPP algorithm and the BSMA algorithm. The core of this research lies in handling the multicast routing problem within single-source networks. This issue is critical in computer networks as it involves how to transmit data from one source to multiple destinations in the most efficient way within a network. Specifically, the multicast routing problem with delay constraints, which is the focus of this study, is of great importance in real-world networks. The paper innovatively proposes a new algorithm to solve this problem. Initially, by employing Lagrange relaxation, the constrained multicast routing problem is transformed into an unconstrained multicast routing problem, which is then balanced between delay and cost weights using the Bilad algorithm. Subsequently, the unconstrained optimization problem is solved using an exact dynamic programming algorithm. To ascertain the effectiveness of the algorithm, extensive experimental simulations are executed to simulate and evaluate its performance under different network environments and scenarios.

The rise of the Internet of Vehicles has brought revolutionary changes to the transportation system and also presented new challenges for data privacy protection. This study is dedicated to exploring how advanced machine learning techniques can enhance the data processing capabilities of the Internet of Vehicles while ensuring the security and privacy of communication data. The research emphasizes the collaborative role of federated learning and deep reinforcement learning. The introduction of federated learning facilitates collaborative learning of distributed data without the need to share raw information, a significant advantage in the decentralized environment of the Internet of Vehicles. Additionally, the application of deep reinforcement learning in the decision-making process provides efficient adaptability to various complex scenarios in the Internet of Vehicles environment. In terms of differential privacy protection, the study discusses protective strategies based on a layered correlated propagation mechanism aimed at reducing the risk of individual information leakage. Furthermore, reinforcement learning plays a crucial role in constructing defense mechanisms, enhancing the system's resistance against potential privacy attacks. Through comprehensive experiments, the proposed methods in the paper can be effectively implemented in practical Internet of Vehicles environments, demonstrating significant effects. The experiments show that the differential privacy protection strategy combining federated learning and reinforcement learning successfully reduces the risk of privacy leakage without sacrificing data utility. This discovery not only holds important significance for academic research but also provides practical solutions for ensuring the data security of Internet of Vehicles in practice.

In a complex electromagnetic environment, accurately extracting target radar signal pulses is crucial for radar signal sorting. With the increasing pulse density in the electromagnetic environment, even though the pulse flow is diluted through presorting, when performing sequence extraction, there are still situations where multiple pulses exist within the search range. This results in the current pulse search methods struggling to extract the correct pulses. To solve this problem, a multiparameter pulse sequence search method is proposed. Firstly, the main sorting algorithm is employed to determine the PRI modulation type of radar signals in the electromagnetic environment. Subsequently, a pre-search is conducted on the pulse flow. Based on the clustering results of the pre-search, the potential carrier frequency, pulse width, and arrival angle information of the current radar signal can be inferred. Consequently, using this parameter information, pulses that do not belong to the current radar signal can be excluded from the search range. The simulation experiments prove that the method can make the accuracy of the sequence search of both slip and jitter radar signals reach more than 95% under the premise that the sequence search of ordinary radar signals is maintained at a high accuracy. Simulation results indicate that in scenarios with high pulse density, especially for jitter and slippage radar signals, this method can improve the accuracy of pulse sequence searches, reducing the probability of wrong selection and the probability of missing selection.

Federated Learning enables collaboration among parties without the need for external data sharing, achieving knowledge sharing without data sharing. However, existing federated learning algorithms still encounter issues such as non-IID (nonindependent and identically distributed) data and catastrophic forgetting, such as data transfer, statistical differences, and local update overfitting. In this paper, we propose a new federated learning algorithm, FCLW, which integrates continuous learning to address catastrophic forgetting and introduces a weight auto-allocation mechanism to solve the heterogeneity of federated learning data. Experimental results demonstrate that, compared to baseline algorithms, FCLW achieves faster convergence and obtains a more accurate global model in both IID and non-IID data settings.

Driving safety has always been a core concern in the field of Autonomous driving vehicles, and the safety issue of interaction between the ego-vehicle and the surrounding vehicles is particularly important in lane-changing scenarios. This paper proposes a deep deterministic policy gradient algorithm (DDPG) based on driving risk for obstacle avoidance decision-making model (RISK-DDPG), which aims to learn risk-aware driving decision-making strategies at a minimized cost. The DDPG algorithm is used to learn the minimally risky driving strategy by using Gaussian risk field to assess the magnitude of the risk of obstacles and the value of the driving risk quantum as a reward function. The performance of the proposed algorithm in this paper is evaluated against traditional obstacle avoidance algorithms by evaluating the performance in the CARLA simulation environment. The results show that the proposed model achieves safer obstacle avoidance decisions while speeding up the model convergence.

In recent years, the rise of distance education has transformed the traditional learning paradigm, with English learning being particularly emphasized as a crucial tool for global communication. However, effective management of online educational resources has become a bottleneck in enhancing the quality of remote English teaching. Leveraging big data mining technology, this study proposes a novel system for managing remote English education resources. Initially, a deep directed graph knowledge tracking model is introduced to analyze and predict learners' grasp of knowledge; this model classifies learners scientifically based on acquired knowledge and recommends corresponding learning paths. Subsequently, the study constructs a model centered around reinforcement learning neural networks, which dynamically adjusts educational content through real-time interaction with learners. This model integrates academic data, learning behaviors, and personal feedback to optimize the learning experience and effectiveness through training. Moreover, the designed system combines resource collection, analysis, and personalized recommendations to ensure the smoothness and efficiency of the management process. Quantitative experiments demonstrate that the proposed system significantly enhances learning outcomes and satisfaction, validating the effectiveness and potential of applying big data mining technology in the management of remote English education resources. The outcomes of this study foreshadow the immense potential and expansion room for data-driven management systems in the field of educational technology, particularly in remote English teaching.

At present, the Quick Access Recorder (QAR) is the most widely used device for storing aircraft recording data. QARdata contains a wealth of valuable and unexplored information, detailing the true state of the aircraft. Therefore, the mining and development of QAR data is particularly crucial. The decoded QAR data contains a large number of outliers, and is always doped with nonlinear and non-stationary noise signals, and the wavelet denoising method is widely used in the signal field because of its simple algorithm and small amount of calculation, which improves the accuracy and efficiency of data processing. The extreme learning machine has strong learning ability and good generalization performance. In order to further improve the noise reduction effect of the sensor signal, a hybrid noise reduction method based on Kernel Extreme Learning Machine (KELM) and wavelet algorithm is proposed, which designs a filter based on the KELM algorithm to reduce the noise of the sensor signal on the one hand, and further smooths the KELM noise reduction signal by using the classical wavelet transform on the other hand, so as to further improve the noise reduction effect under the premise of ensuring that the signal is not distorted. Simulation results show that the noise reduction effect of the proposed hybrid noise reduction method is better than that of the single noise reduction method.

LOAM stands as a quintessential 3D Lidar SLAM algorithm capable of real-time robot positioning and mapping; however, it may succumb to positioning drift and mapping inaccuracies during prolonged operation. Addressing LOAM's limitations, LeGO-LOAM enhances robustness by integrating loop closure detection via ICP, yet it remains susceptible to false positives or omissions in expansive environments. In light of these challenges, this study introduces a Lidar SLAM algorithm that leverages a bag-of-words approach for loop closure detection. Adopting LOAM as the preliminary odometry, the method incorporates LinK3D alongside a bag-of-words model to devise a novel loop detection module. The process unfolds as follows: initially, the LinK3D algorithm is employed to extract and characterize point cloud features; subsequently, a hash data structure is utilized to construct the bag-of-words model for these descriptors. Thereafter, drawing inspiration from TF-IDF, the method expedites loop closure detection by computing the 6-DoF pose transformation between valid loop frames and the current frame. Ultimately, pose adjustments are refined using the graph optimization tool GTSAM. To enhance the feature representation and robustness of the LinK3D algorithm, this paper introduces an augmented LinK3D feature extraction technique, which integrates plane feature data. The algorithm's efficacy was ascertained through a series of tests on six Lidar point cloud sequences from the KITTI public dataset, including sequences 00, 05, and 09, benchmarked against classical SLAM algorithms such as A-LOAM and LeGO-LOAM. Evaluation across two dimensions—pose precision and mapping quality—confirmed the proposed algorithm's significant reduction in cumulative errors, elevated positioning accuracy, and enhanced mapping fidelity, all while meeting the real-time operational criteria.

In recent years, visual intelligence-assisted driving technology has been widely applied and promoted, but with it comes challenges to its safety, such as poor recognition performance of algorithms for slender objects like wires, ground seams, tree branches, etc. To address this pain point, this paper proposes a slender object detection model based on YOLOv8. Firstly, the D-C2f module is employed in the backbone feature extraction network. This deformable convolution module enables the network to better fit the unique shapes of slender objects. Secondly, the integration of the Biformer attention mechanism and the dysample upsampling network enhances the focus on the target objects in high-resolution feature maps. Finally, an occlusion-aware attention module is adopted at the detection end to improve the model's ability to detect occlusion issues on the road surface. Through experiments conducted on a wire dataset, the model achieves a detection accuracy of 93.2%, a recall rate of 90.2%, and a mean Average Precision (mAP) of 94.0%. The model has 11,181,427 parameters. Compared to the original YOLOv8 model, these metrics represent improvements of 3.8%, 1%, and 2.3%, respectively. The results demonstrate the superior detection accuracy of the proposed algorithm for slender objects, making it a viable solution for related challenges in autonomous driving.

The triangle mesh model with huge volume is not conducive to computer storage, analysis and rendering, and the mesh simplification method can effectively reduce the complexity of the triangle mesh model, but there are problems such as the loss of visual features and the degradation of mesh quality. For this reason, a triangle folding simplification method that can maintain visual features is proposed. First, the geometric error value is defined by combining the quadratic error metric and the feature constraint factor, which in turn determines the position of the new vertex. Then, Gaussian curvature and visual feature factor are used to define the visual importance of the triangle. Next, the error metric function for triangle folding is defined by combining the geometric error value and visual importance to characterize the folding cost. Finally, the triangles are sorted according to the folding cost and the folding simplification is accomplished according to the specified simplification rate. The experimental results show that the method can reduce the simplification error of the mesh model and generate high-quality simplified models while maintaining the visual characteristics of the model.

With the continuous development of water transportation in China, ships entering and leaving the ports have become increasingly significantly. The Automatic Identification System (AIS) provides powerful data support for port management by transmitting critical information such as ship location, heading, and speed in real-time. Based on data warehouse technology, a port ship management platform is constructed, and it uses AIS data analysis to achieve dynamic monitoring for ship trajectories and operations. The platform employs a refined layered design that encompasses key aspects such as data acquisition, processing, storage, and visualization, optimizing the data processing workflow and significantly enhancing the stability and flexibility of the platform. It utilizes message queues for real-time data acquisition and performs immediate processing through stream computing. The processed data is intelligently stored in corresponding data storage services based on its granularity. With a front-end and back-end separation design approach, the platform presents data to users in an intuitive and clear graphical format. The analysis of AIS data not only facilitates the efficient management of ship arrivals and operations in ports, but also contributes to the optimization of berth allocation, channel management, and personnel scheduling, ultimately enhancing the overall efficiency of port management.

This article first introduces the basic characteristics of Harmony OS. As an emerging operating system, the core advantage of Harmony OS lies in the unified architecture standard and distributed soft bus capability, which effectively solves the problems of inconsistent equipment and interface standards and complex experimental wiring in traditional wind tunnel data acquisition systems. This article further designs a wind tunnel data acquisition system based on Harmony OS, which can synchronously collect temperature, pressure and other information from different measurement points in the wind tunnel. The hot air tunnel was tested through technologies such as embedded development, hardware communication and interface conversion, and distributed synchronous data collection. Successfully completed the requirement for distributed data synchronization collection, with synchronization deviation maintained between 2.5 microseconds. The results indicate that Harmony OS has broad application prospects and research value in wind tunnel data acquisition systems.

Cracks are one of the most common damages in highway bridges, and timely detection of existing cracks is crucial for the safety of highway bridges. Targeting the issues of low crack detection accuracy and many missed detections in complex backgrounds, a new network ES-YOLOX (Embed-Shuffle-YOLOX) is proposed based on the YOLOX object detection algorithm. This method enhances the backbone network's capability for feature extraction by embedding additional CSPLayer, introducing the reconfigured ShuffleNetV2 structure, introducing depth-separable convolution to decrease both the parameter count and computational load of the network, and finally introducing the ECA attention module to make the network more concerned with crack features. Experimental results show that the proposed method increases the mAP50 value by 2.11 % and the F1 score by 2.17 %, and reduces the number of parameters and calculations by 55.8% and 62.7%.

Aging is a problem faced by many countries. China's aging is becoming more and more serious, and the physical and mental health of the elderly is a problem that can not be ignored. As an entertainment activity, square dance APP can realize the interaction between people and mobile phones. When the elderly participate in square dance interactive activities, on the one hand, they can eliminate most negative emotions by interacting with machines; On the other hand, participating in APP activities can improve hand-brain collocation and reduce the occurrence of Alzheimer's disease. Based on the analysis of the psychological and behavioral characteristics of the elderly when participating in square dance, the recognition of body motion itself has great flexibility and diversity, and the analysis of movements also involves the processing of background and noise. Combined with the research of somatosensory game development elements and design methods, noise elimination and frame rate adjustment are realized from the process of image acquisition and image preprocessing based on image depth learning. 2DCNN, Time Displacement Module (TSM), RNN modeling timedependent feature fusion extraction, Adam optimizer optimization, Euclidean distance to judge the similarity of images, discriminate square dance action recognition. Experimental results show that this method can effectively recognize somatosensory gestures, hoping that this result can provide a new idea for the development of somatosensory games.

The continuity risk and integrity risk in the traditional advanced receiver autonomous integrity monitoring (ARAIM) algorithm are equally distributed to all visible satellites, which leads to a relatively conservative vertical protection level (VPL) calculation, and a method to optimize the allocation of risk probability based on the improved particle swarm algorithm is proposed. The improved particle swarm algorithm takes the risk probability allocation strategy as the particle of the algorithm, takes the weighted sum of the vertical protection level VPL as the adaptive function of the algorithm, improves the search capability of the algorithm through nonlinear dynamic adjustment of the inertia weights, and improves the convergence speed of the algorithm through adaptive adjustment of the learning factor, which enables the algorithm to globally perform a stochastic search within the given range to ensure that the algorithm can be more precisely track down the risk probability assignment method that optimizes the VPL value, thus achieving the purpose of improving the availability of ARAIM system. The experimental results show that the proposed method is able to optimize the calculation of VPL and improve the availability of the ARAIM algorithm in both single and dual systems for integrity monitoring.

As a combination of information technology and manufacturing technology, advanced CNC machining technology is the key of modern manufacturing technology, which is of great significance to realize the automation of CNC machining. In this paper, the bibliometric analysis methods are adopted to investigate the development direction of CNC technology in terms of patent information. Through the analysis of patent application trend and patent distribution area, the development status of CNC technology is summarized. The market competition status of CNC technology is identified through the analysis of major patent applicant. Through the analysis of technical topic, the research trend and core technology of CNC technology are illuminated in this paper.

With the rapid development of cloud storage technology, ensuring the integrity of data on cloud remains a significant challenge. This paper introduces an blockchain-based verification scheme, DVOD, to ensure transparent, efficient, and secure verification of the integrity of cloud storage data, in which data blocks are salted to generate tags with challenge numbers, the chameleon hash algorithm is then used to transform these tags into a series of generate leaf nodes for constructing a Merkle tree. Updating the salt in challenge-response while preserving the immutability of the Merkle tree. addresses the real-time data issue and the challenge-response solution's limited number of challenges, significantly enhancing efficiency compared to traditional schemes that reset salt after each verification. Comprehensive security analysis demonstrates the effectiveness of the scheme in defending against malicious entities targeting the system. Conducting experiments with other schemes show the DVOD has the substantial enhancements in computation and communication.

In response to the demand for data transmission efficiency and accuracy for the diversity of power line carrier services in China, an adaptive data compression scheme is introduced in the paper, which is trained to compress and decompress the data after dividing it into data blocks, and selects an appropriate compression procedure by predicting the data entropy through deep neural network inference. By dividing the data blocks and compressing them according to the characteristics, this compression scheme avoids the worst compression rate and ensures that the compression rate of various types of business data is close to the optimal value. This scheme has some reference value for power line carrier multi-class service data transmission

The effective enhancement of braking energy recovery and safety in electric vehicles can be achieved by accurately identifying the driver's braking intention and developing a corresponding regenerative braking control strategy based on different braking intentions. This paper presents a categorization of braking conditions into mild, moderate, and emergency levels, followed by the construction of a test system for recognizing braking intentions. Multiple sets of braking conditions are tested under various initial speeds to obtain parameters for recognizing the driver's intended action during braking. Through feature selection using random forest, acceleration, brake pedal displacement, and brake pedal force are identified as key parameters for recognizing the driver's intended action during braking. Subsequently, an AdaBoost-based model is established for recognizing the driver's intended action during braking. Experimental data is used to validate this model offline and compare it with various other models for recognition of driving intentions during brakes. The results show that the braking intention recognition model based on AdaBoost has a high recognition accuracy.

This study aims to dig deeper into the evolution of consumers' willingness to purchase electric vehicles (EVs) between 2010 and 2023, along with the research hotspots and cutting-edge trends within this domain. We employed big data analytics and text mining techniques to achieve this objective, utilizing two bibliometric analysis software tools: CiteSpace and VOSviewer. These tools enabled us to thoroughly analyze and visualize the extensive dataset indexed by the Web of Science Core Collection within the specified timeframe. Through big data analytics, we excavated how various factors, including technological advancements, policy environments, infrastructure developments, and consumer psychology, have collectively influenced consumers' acceptance of EVs. Simultaneously, text-mining techniques facilitated the extraction of critical insights from the vast literature, further elucidating the underlying motivations behind consumers' willingness to purchase EVs. This study transcends the analytical framework of traditional bibliometrics and provides a fresh perspective and profound understanding of EV purchase intention research through integrating big data and text mining techniques.

With the expansion of power system scale and the increase of complexity, electromagnetic transient simulation plays a more and more important role in power system analysis. However, traditional simulation methods face the problems of large amount of calculation and long-time consumption, which is difficult to meet the needs of rapid analysis and decisionmaking of modern power systems. At the same time, the working mode of many people in different places requires that the platform can realize real-time data sharing and exchange. In order to solve this problem, this paper proposes and implements a distributed platform for cooperative calculation and simulation results management of electromagnetic transients. The platform uses multi-client collaborative computing technology to achieve efficient calculation of largescale electromagnetic transient simulation. At the same time, through the collaborative sharing and management mechanism, the cloud browsing, displaying and publishing functions of electromagnetic transient models, graphs and results for multi-users are realized, which can promote efficient communication of projects. This paper first introduces the overall architecture and key technologies of the platform, and then describes the implementation process of the platform in detail, including task division, data synchronization, collaborative computing and result management. Finally, the feasibility and superiority of the platform are verified by experiments.

In order to solve the problem that the current distribution network project cost influencing factors are numerous and complex, and it is difficult to accurately predict the project cost, a combined prediction model of construction project cost. Firstly, the important cost influencing factors are selected from the big data of construction cost based on gray correlation analysis, secondly, the key parameters of the least squares vector machine algorithm are optimized based on cross-validation and grid search algorithm.

This study introduces a technique for analyzing experimental data from Newton's ring experiments to measure lens curvature, utilizing Origin software. This technique allows for precise determination of the lens curvature radius and supports comprehensive analysis of the experimental data. The method is distinguished by its straightforwardness, high efficiency, and precision, making it a valuable resource for analyzing similar experimental data.

The examination of the spatial and temporal patterns of evacuees' movements in the community following an earthquake offers practical assistance to government emergency response and management. It also provides a more accurate foundation for earthquake relief efforts and the timely restoration of social order. On December 22, 2021, a series of earthquakes, ranging from 4.2 magnitude (M4.2), struck the Tianning District of Changzhou City, Jiangsu Province, China, leaving a profound impact on the local community. Through the utilization of 81.69 million mobile phone location records collected after the earthquake, this research conducted a quantitative analysis of people's locations post-earthquake at different scales. Furthermore, it identified the optimal bandwidth for kernel density estimation (KDE) in each study area. This study's findings revealed the flow pattern of evacuated individuals at the community level following a non-disastrous earthquake, offering a quantitative basis for the planning and optimization of disaster shelters. It holds significant research and application value in enhancing earthquake preparedness and disaster mitigation capabilities.

In recent years, informationization and digitalization have become more and more closely integrated with the the shipbuilding industry. The software which is being studied in this study, is a exclusive cruise ship design and ship survey software, by establishing a model according to the position coordinates and parameters or import a model created in SolidWorks, using of lumen method and the point-by-point method, could quickly and conveniently get the calculation result. Due to the different rules of different classification societies, which are built-in the software system, and the software would compares the results and rules, check whether the illumination figures of each position of cabins conforms to the rules of classification society according to the lighting layout. Lumen method and point-by-point method are two of three main methods for illumination calculation presently. The ranges of application of the two approaches are complementary. So the software can satisfy demands for both general cases and special shapes of cabins.