This paper establishes the class separability of mines in various backgrounds for the low sun angle using Dec. 2000 and June 2001 airborne 808nm laser imagery. Specifically, this paper provides the polarization distribution of mines and background types based on four statistics: in-plane (P); cross-plane (S); P-S, and degree of polarization (DoP = (P-S)/(P+S)). This study provides a first look at which polarization can benefit the performance for airborne minefield detection and under which background conditions for a particular time of the year (i.e. low sun angle scenarios). This study presenting the polarization class distribution provides a good basis for the algorithm development effort for an automatic mine/minefield detection system using 808nm laser imagery. This study used two subsets from the December 2000 and June 2001 airborne data collections collected with the Sci-Tech breadboard 808nm laser. To accurately represent the distribution of the mines and background, there are 24,000 mine and 144,000 background pixels were manually to ensure the perfect registration between pixels located in P and S images for the same mine or background.
This paper quantifies the overall detection performance for landmines in various background and solar conditions in an attempt to provide the performance bounds for airborne mine detection systems. Specifically, for comparison purposes, this paper quantifies the detection performance based on the RX detection algorithm implemented as the baseline LAMD approach, RX implementation as a correlation operator, and intensity thresholding approach using airborne laser imagery. The generated receiver operating characteristic (ROC) curves, in turn, provide a good basis for system trade-off study in terms of computational time and complexity, and performance benefit for real-time systems. This paper includes the ROC curves with and without man-made objects to access the effect of the man-made objects based on these algorithms. The paper uses two subsets from the December 2000 and June 2001 airborne data collections using the SciTech Breadboard 808nm laser at a U.S. Army test site. The total mine opportunities and the area coverage are 1619 and 146,000 m2, respectively. The total number of man-made objects are 800 (approximately 137 images of which each image contains approximately 8 man-made objects). The man-made object list contains mine sized aluminum plates and wood, coke cans and others. Mine list contains M20, M19, TM62M, TM62P2, TM62P3, RAAM, VS1.b.
This paper focuses on demonstrating the complexity in the optimization process of the sequential algorithm, which is a multi-0stage algorithm with each stage using fewer bands than the previous stages. Specifically, this paper describes the process used to obtain the optimal confidence level and the class separation parameter to quantify the hyperspectral detection performance using the sequential algorithm with Chebyshev's inequality test. This paper also presents the computational complexity involved in reaching the optimum confidence level and the recommended methodology for lessening the computational burden. The detection performance for different spatial resolutions are presented and compared with the ARES baseline performance using all spectral bands. The Forest Radiance I database collected with the HYDICE hyperspectral sensor is utilized. Scenarios include targets in the open, with footprints of 1 m, 2 m and 4 m; and different times of day. The total area coverage and the number of targets used in this evaluation are approximately 10km2 and 108, respectively. The description of the database and sensor parameters can be found.
This paper discusses some preliminary results of the application of simple neural networks to the problem of landmine detection in IR imagery. A large data set of IR imagery (3-5)mum) collected as part of the U.S. Army's Lightweight Airborne Multispectral Minefield Detection (LAMD) system is used as the basis for the analysis. The data set is divided into training and testing subsets then used to train and evaluate the performance of some neural networks. A single neuron perceptron is trained and evaluated using two different types of input feature. The first type of input feature is based on the raw pixel values with typical maximum vale normalization. The second type is based on the unity vector of the inputs to take advantage of the angular displacement feature of the vector [1]. A more complex, multiple neuron network is also trained and evaluated. The results are compared to determine whether the increased computational complexity of the multiple neuron network is justified in terms of improved performance.
This is a follow-up work to analyze completely the detectability of the buried mines for the spectral regions extending from Visible/Near IR (VNIR) to Longwave IR (LWIR). Similar to previous work focusing on the VNIR region (1) this paper presents the quantitative detectability of the buried mines in the 3-5)mum and 8-12)mum regions. Specifically, this paper presents a statistical analysis for the buried mines in specified spectral regions for various soils and burial durations. As shown in the previous work (1) the performance based on the single hypothesis test using the distance measure was better than the intensity thresholding method. This paper focuses on only the distance measure method for statistical analysis of the data, and subsequently, classification to quantify the detectability of the buried mines in the 3 to 5 and 8 to 12 micron regions.
This paper includes analysis/assessment and development of detection algorithms: (1) the assessment of detectability of surface mines using the RX algorithm implementation, which in turn, provides a first look at the limitation of the algorithm for suitable real-time implementation; (2) the development of the adaptive real-time mine detection algorithm (ARMD) based on statistical analysis of the data. The statistical analysis includes the class distribution between mines and background, the underlying distribution for mines and background based on the quantile-quantile plot. The paper also compares the quantitative performance of probability of detection (Pd) and false alarm rates (FAR) for different detection techniques for different background and mine types. This paper also presents the minefield probability of detection versus minefield false alarm rate to gauge the minefield detection performance trade-off using: (1) only mine density; and (2) mine density with pattern. This paper also demonstrates the importance of the observables that offer the class separability between mine/target and background for automatic target detection/recognition applications. Detection algorithms with high computational capability are not the 'silver bullet' for automatic target detection/recognition as commonly believed. The art of ATR is the ability to be able to pinpoint the observables that distinguish mines and background. Once the observables offer the class separability between classes are established, any simple correlation method can deliver an acceptable performance (demonstrating that highly computational methods, indeed, are wasteful and unnecessary). This paper uses the multiband and broadband data collected with the AMBER (3.5-5)mum) camera in May 2000. This data set contains about 513 (approximately 1.1 in resolution) images covering three spectral regions: 3-5)mum, 3- 4.2)mua and 4.2-5)mua. The total number of mines and the area coverage for these three spectral regions are approximately 579, and 25200m2, respectively. Note that each spectral region contains 171 images (of which 53 images contain mines with 131 large mines, and 62 small mines) covering about 8400m2. Also note that to stimulate minefields, an image containing 3 mines with a straight line pattern is defined as a minefield opportunity.
This paper identifies the optimal bands in the 3 to 5 micron region for surface mines for the lightweight airborne detection (LAMBD) sysem. Specifically, this paper focuses on the analysis to identify the optimum bands in the 3 to 5 micron region which can be used in a filter wheel implementation in an attempt to add the multi-band capabilities to enhance the detection performance for various background, times of day, while lessening the weight/size/power problem in a lightweight airborne mine detection (LAMBD) system. The analysis includes the hyperspectral signatures of various mines and backgrounds, the contrast between mines and backgrounds for various spectral regions, different times of day, and differential heights simulating sensor airborne scenarios. This paper uses data collected with a Design and Prototypes Fourier Transform Infrared (FTIR) Spectrometer (D&P) mounted 3m above the ground on the Mobile Sensor Platform (MSP) at a temperate test site. The total number of signatures used was 1000 for mines and 1000 for backgrounds.
KEYWORDS: Sensors, General packet radio service, Metals, Land mines, Palladium, Mining, Ground penetrating radar, Data fusion, Gallium, Infrared sensors
This paper quantifies the mine detection performance by fusing ground penetrating radars and a metal detector. Specifically, the fusion scheme used in this paper is ANDing different sensors with high probability of detection regardless of the false alarm rate. As the false alarms are random, and each sensor processes detected objects differently to produce high probability of detection, fusion by ANDing eliminates the majority of false alarms, and hopefully maintains the high probability of detection based on the mutually exclusive property of the sensor being fused. This paper uses data collected with different GPR's of Vehicular Mounted Mine Detection ATD systems and a handheld metal detector at Aberdeen Testing Center, Maryland and Socorro, New Mexico test sites. The total number of mines encountered and area coverage are approximately 450 and 13000m2, respectively.
This paper present the quantitative detection performance of buried mine reflectance signature sin various soils and burial durations. The spectral signatures including the distribution representing class separation between mines and background is performed. The quantified detection performance is based on single hypothesis test using the distance measure and the thresholding method. This paper uses a subset of the data collected under the Night Vision and Electronic Sensors Directorate sponsored ERIM Hyperspectral Mine Detection Phenomenology data collections. The dat set used was collected with an Analytical Spectral Devices Field Spectrometer at Ft. Carson, and contains about 700 mine, and 600 background signatures with hundreds of bands extending from .35 to 2.5 micrometers .
This paper describes an automatic target detection algorithm based on the sequential multi-stage approach. Each stage of the algorithm uses more spectral bands than the previous stage. To ensure high probability of detection and low false alarm rate, Chebyshev's inequality test is applied. The sequential approach enables a significant reduction in computational time of a hyperspectral detection system. The Forest Radiance I database collected with the HYDICE hyperspectral sensor at the U.S. Army Proving Ground in Aberdeen, Maryland is utilized. Scenarios include targets in the open, with footprints of 1 m and different times of day. The total area coverage and the number of targets used in this evaluation are approximately 6 km2 and 126, respectively.
This paper describes an automatic mine detection algorithm (AMDA) based on the template matching technique. Specifically, this paper demonstrates that regardless of sensor artifacts and other perplexities including environment effects such as terrain variation or weather conditions, there will be distinctive information between targets and clutter imbedded in the signatures for the discrimination. This paper also includes the data analysis of the ground penetration radar signatures and quantifies the AMDA performance. This paper use a subset of the DARPA clutter data collected with the Geo-Centers ground penetration radar at Fort A.P. Hill and Fort Carson. This subset contains anti-personnel, and anti-tank mines buried from 1 to 6 inches deep with the size of the mine ranging from 2 to 12 inches in diameter. The total number of mines and the area coverage of this subset are about 30 and 600m2, respectively.
This paper presents (1) trade-off studies of detection performance versus the number of bands using reflective hyperspectral imagery; (2) the quantitative detection performance of various approaches used in automatic target detection. The trade-off studies of detection performance versus the number of bands are based on the Adaptive Real-Time Endmember Selection and Clutter Suppression (ARES) algorithm. The ARES algorithm presents a new concept and approach for spectral-spatial aided/automatic target detection based on the unique characteristics of the spectral signatures produced by the hyperspectral imaging system for remote sensing surveillance and reconnaissance applications. This paper compares the quantitative detection performance based on the ARES algorithm with other automatic target detection approaches. This paper uses the Forest Radiance I database collected with the HYDICE hyperspectral sensor at Aberdeen U.S. Army Proving Ground in Maryland, including scenarios such as targets in the open, with footprint of 1 meter, and at different times of day.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.