In this paper, we proposed a multi-scale convolutional neural network for hyperspectral image classification task. Firstly, compared with conventional convolution, we utilize multi-scale convolutions, which possess larger respective fields, to extract spectral features of hyperspectral image. We design a deep neural network with a multi-scale convolution layer which contains 3 different convolution kernel sizes. Secondly, to avoid overfitting of deep neural network, dropout is utilized, which randomly sleeps neurons, contributing to improve the classification accuracy a bit. In addition, new skills like ReLU in deep learning is utilized in this paper. We conduct experiments on University of Pavia and Salinas datasets, and obtained better classification accuracy compared with other methods.
Proc. SPIE. 9142, Selected Papers from Conferences of the Photoelectronic Technology Committee of the Chinese Society of Astronautics: Optical Imaging, Remote Sensing, and Laser-Matter Interaction 2013
KEYWORDS: Fluctuations and noise, Detection and tracking algorithms, Aerospace engineering, Cameras, Image processing, Bismuth, Spatial resolution, Defense technologies, Field emission displays, Astronomical engineering
Target location is a fundamental application in aerial image process. In this work, a fast normalized cross correlation algorithm is proposed for the application of target location in aerial image. Firstly, normalized cross correlation has been proved equivalent to Euclidean distance. In the search step, the target template and the corresponding window of base image are projected to a set of mutually orthonormal vectors for calculating the lower bound of the distance, where the windows with too large distance relative to the target template will be rejected in this step. Finally, the directly normalized cross correlation calculating is applied to the rest windows of base image to achieve the final correct location of target. The experimental results show that compared with traditional method, the proposed method significantly improved the computational complexity without sacrificing the spatial resolution or the accuracy of the match result.
A new synthetic aperture radar interferometry (InSAR) data processing method comprehensively based on three parts of complex images is introduced. The method includes image pair registration and interferogram creation. These methods are improved and extended, and a set of integrated technology, which is named contoured correlation interferometry (CCI), is formed for the InSAR data processing. The CCI method needs only an arbitrary three parts of InSAR complex image pair to generate an interferogram without noise and blurring effect. The formulae of the CCI method are deduced and proved in a different way, the details and steps of the whole method are explained systemically, the algorithms for the calculation of fringe orientations are improved, and the discussions about selecting of parameters and comparing results of different methods are shown.
As a rising navigation technology, vision navigation has many advantages, such as passive measurement, antiinterference,
no accumulation of error and comprehensive parameters, etc. It shows a promising application prospects in
autonomous navigation for UAV. Based on an efficient, reliable and accurate scene matching, a vision altimeter and 3-D
position estimation method are proposed. By matching multiple points between aerial image and reference image, it
estimates UAV's position and height according to photogrammetry. To measure UAV's velocity, a mapless speed
measurement method which tracks ground features between image sequences is introduced. Flight tests had shown the
effectiveness and accuracy of our methods.
Coherence of interferometric synthetic aperture radar (InSAR) complex image pair is a fundamental observable in interferometric radar measurements, which is usually measured by comparing the radar return across several nearby radar image pixels and has found diverse applications. This paper proposes a coherence estimation method which requires three arbitrary parts of the two complex images to implement interference. The proposed method can also be used to co-register InSAR image pair, which means that the imaginary part of the master image can be left away. In addition, an improved quality-guided phase unwrapping method is forwarded with the quality map generated by the coherence estimation method we have recommended above. A look-up table is adopted to reduce the processing time. The experimental results show that both methods are effective and greatly reduce the phase unwrapping time compared with the existing quality-guided methods.
Speckle fringe patterns of ESPI are full of high-spatial-frequency and high contrast speckle noise which defies normal
process methods. Filtering with contoured windows has been proven to be an efficient approach to filter off the speckle
noise while reserving the fringe patterns obtained by subtraction of two original speckle patterns. Furthermore, with
contoured windows, the contoured correlation fringe pattern (CCFP) method proposed by the authors can derive highquality
fringe patterns of ESPI with speckle-free, smooth, normalized and consistent fringes from two original speckle
patterns. CCFP method can also extract the phase field with a single-step phase-shifting. Determination of contoured
windows is a key step in CCFP method. The contoured windows used to be determined by fringe orientations only and
this process would generate accumulated errors. In this paper, two new algorithms to determine the contoured windows
according to the fringe intensity slope and the distance ratio to neighboring skeletons with the help of the local fringe
direction are proposed. These new techniques can determine contoured windows more precisely and more robustly with
no accumulated errors. Some applications of our new contoured windows are also presented.
Silica coated magnetic nanoparticles have been prepared by sign water-in-oil micro emulsion technique, and have been characterized by Transmission Electron Microscopy and e-ray diffractometer. Concanavalin A was covalent immobilized on the surface of silica coated magnetic nanoparticles. Dextran-Rhodamine B could be easily separated from aqueous solutions by the functionalized nanoparticles. The result implied that the preparation route is easy to yield silica coated magnetic nanoparticles which can be modified for the separation of other target biomolecules, cells, and etc.
A refractometric fiber-optic chemical sensor bas don mode- filtered light detection (MFLD) was designed. This sensor was fabricated by inserting a non-chem- or bio-modified fiber core into a silica capillary with a CCD detector as the multichannel-signal capture. An interesting observation was obtained that there is an increase in signal with a decrease in sample refractive index in this unclad fiber sensor, just contrary to the polymer-clad fiber instead sensor reported previously. Academic deduction of the light propagation in detail, which was demonstrated by our experimental results, further interprets the reason of the difference between the unclad sensor and the polymer-clad sensor, and provides theoretical foundation on development of a MFLD-derived analyzer with synchronization of separation and assay in future work. This sensor was also demonstrated to successfully apply to detection of acetic acid with linear response of 0-90 percent and relative coefficient of 0.9972. The sensor's advantages in high S/N ratio and instant response show its potentials in food and chemical industries as a general sensor.
A novel method of synchronization of separation and analysis is described, which use the mode-filtered light detection as an on line detector of capillary electrophoresis, furthermore, corresponding to this method an instrument has been established. Capillary electrophoresis runs in an annular column, which is constructed by a naked optical fiber inserted into a fused silica capillary, instead of a round capillary, instead of a round capillary. Moreover, the annular column is not only a vessel for electrophoresis but also a sensor to the mode-filtered light. Along the annular column side several detection channels are set up to gather the mode-filtered light and transmit it to CCD, so every channel provides the information of the sample form where it is located. Using the capillary isotachophoresis in this annular column, a sample containing alanine and glycin is separated and analyzed simultaneously by the multi-channel mode-filtered light detection.
Novel biotinylated molecular beacon (MB) probe was developed to prepare DNA biosensor using sol-gel chemistry. Biotinylated bovine serum albumin was encapsulated in the sol-gel matrix to form a biotinylated silica gel. Streptavidin acted as a bridge between biotinylated MB and silica gel. The MB based DNA biosensor can detect nonlabeled DNA target in real time with high sensitivity and one-base- mismatch selectivity, which indicated the potential applications in single point mutation detection. Compared with the previous report, the sol-gel based DNA biosensor has three advantages: (1). The response time was significantly shortened to 10 minutes when the cDNA concentration was over 10 nM. (2). The silica gel prolonged the lifetime of the MB based biosensor, which could keep its activity for thee weeks if stored in a buffer. (3) The gel formed by sol-gel method has a larger surface area than the common glass cover at the same scope. The number of binding sites for biotinylated increased significantly. Hence the detection limit was down to subnanomolar/L.