Automating the detection process in acoustic-seismic landmine detection speeds up the detection process and eliminates the need for a human operator in the minefield. Previous automatic detection algorithms for acoustic landmine detection showed excellent results for detecting landmines in various environments. However, these algorithms use environment-specific noise-removal procedures that rely on training sets acquired over mine-free areas. In this work, we derive a new detection algorithm that adapts to varying conditions and employs environment-independent techniques. The algorithm is based on the generalized likelihood ratio (GLR) test and asymptotically achieves a constant false alarm rate (CFAR). The algorithm processes the magnitude and phase of the vibrational velocity and shows satisfying results of detecting landmines in gravel and dirt lanes.

Remote acoustic or seismic forms of excitation for laser Doppler vibration landmine detection are low false alarm rate detection strategies. A more recent approach now under investigation includes a direct mechanical excitation through a prodder or probe. In this research, we report on simple laboratory measurements of the VS-1.6 landmine undergoing direct mechanical excitation from a modified prodder while measuring the landmine's pressure plate vibrational response with a scanning laser Doppler vibrometer. The direct mechanical excitation mechanism, located near the prodding end of a rod, consists of a miniature piezoelectric stack actuator. We additionally compare direct excitation to both acoustic and seismic methods in a large sandbox filled with dry sand. We show that for the landmine buried almost flush, direct contact mechanical excitation compares favorably to both seismic and acoustic excitation responses for the (0,1) mode of the pressure plate. We also observe additional features not previously seen in either seismic or acoustic excitation.

In a series of previous papers, analytical results dealing with the effects of soil electromagnetic properties on the performance of induction metal detectors were reported. In this paper experimental data are provided to verify some previously reported results. The time-domain response of a magnetic soil half-space and a small metallic sphere situated in air as well as buried in the soil were measured using a purpose-designed system based on a modified Schiebel AN19/2 metal detector. As in the previous work, the sphere is chosen as a simple prototype for the small metal parts in low-metal landmines. The soil used was Cambodian "laterite" with dispersive magnetic susceptibility, which serves as a good model for soils that are known to adversely affect the performance of metal detectors. The metal object used was a sphere of diameter 0.0254 m made of 6061-T6 aluminum. Experimental data are in good agreement with theoretical predictions. Data also show that for the weakly magnetic soil used in the experiments, the total response of the buried sphere is the sum of the response of the soil and that of the sphere placed in air. This finding should simplify the prediction or measurement of response of buried targets as one can separately measure/compute the response of an object in air and that of the host media and simply add the two. This simplification may not be possible for soils that are more strongly magnetic.

Studies have showed that magnetically susceptible soils significantly affect on the EMI sensors performances, which in return reduce the sensors discrimination capabilities. In order to improve EMI sensors detection and discrimination performances first soil's magnetic susceptibility needs to be estimated, and then the soils EMI responses have to be taken into account during geophysical data inversion procedure. Until now the soil's magnetic susceptibility is determined using a tiny amount (up to 15 mg) of soil's probe. This approach in many cases does not represent effective magnetic susceptibility that affects on the EMI sensors performances. This paper presents an approach for estimating soil's magnetic susceptibility from low frequency electromagnetic induction data and it is designed namely for the GeoPhex frequency domain GEM-3 sensor. In addition, a numerical code called the method auxiliary sources (MAS) is employed for establishing relation between magnetically susceptible soil's surface statistics and EMI scattered field. Using the MAS code EMI scatterings are studied for magnetically susceptible soils with two types of surfaces: body of revolution (BOR) and 3D rough surface. To demonstrate applicability of the technique first the magnetic susceptibility is inverted from frequency domain data that were collected at Cold Regions Research and Engineering Laboratory's test-stand site. Then, several numerical results are presented to demonstrate the relation between surface roughness statistic and EMI scattered fields.

The identification of unexploded ordnance (UXO) using electromagnetic-induction (EMI) sensors involves two essentially independent steps: Each anomaly detected by the sensor has to be located fairly accurately, and its orientation determined, before one can try to find size/shape/composition properties that identify the object uniquely. The dependence on the latter parameters is linear, and can be solved for efficiently using for example the Normalized Surface Magnetic Charge model. The location and orientation, on the other hand, have a nonlinear effect on the measurable scattered field, making their determination much more time-consuming and thus hampering the ability to carry out discrimination in real time. In particular, it is difficult to resolve for depth when one has measurements taken at only one instrument elevation. In view of the difficulties posed by direct inversion, we propose using a Support Vector Machine (SVM) to infer the location and orientation of buried UXO. SVMs are a method of supervised machine learning: the user can train a computer program by feeding it features of representative examples, and the machine, in turn, can generalize this information by finding underlying patterns and using them to classify or regress unseen instances. In this work we train an SVM using measured-field information, for both synthetic and experimental data, and evaluate its ability to predict the location of different buried objects to reasonable accuracy. We explore various combinations of input data and learning parameters in search of an optimal predictive configuration.

Electromagnetic Induction sensor (Metal Detector) has wide application areas for buried metallic object searching, such as detection of buried pipes, mine and mine like-targets, etc. In this paper, identification of buried metallic objects was studied. The distinctive features of the signal were obtained, than classification process was performed. Identification process was realized by utilizing k-Nearest neighbor and Neural Network Classifiers.

This work explores possible performance enhancements for landmine detection algorithms using frequency domain wideband electromagnetic induction sensors. A pre-existing four parameter model for conducting objects based on empirically collected data for UXO is discussed, and its application for accurately modeling landmine signatures is also considered. Discrimination of mines versus clutter based on the extracted model parameters is considered. Furthermore, this work will compare the effectiveness of discrimination based on the four parameter model to a matched subspace detection algorithm. Experimental results using data from government run test sites will be presented.

This paper combines the normalized surface magnetic charge (NSMC) model and a pole series expansion method to determine the scattered field singularities directly from EMI measured data, i.e. to find a buried object location and orientation without solving a time consuming inverse-scattering problem. The NSMC is very simple to program and robust for predicting the EMI responses of various objects. The technique is applicable to any combination of magnetic or electromagnetic induction data for any arbitrary homogeneous or heterogeneous 3-D object or set of objects. In this proposed approach, first EMI responses are collected at a measurement surface. Then the NSMC approach, which distributes magnetic charge on a surface conformal, but does not coincide to the measurement surface, is used to extend the actual measured EMI magnetic field above the data collection surface for generating spatially distributed data. Then the pole series expansion approach is employed to localize the scattered fields singularities i.e. to determine the object's location and orientation. Once the object's location and orientations are found, then the total NSMC, which is characteristic of the object, is calculated and used for discriminating between UXO and non-UXO items. The algorithm is tested against actual EM-63 time domain EMI data collected at the ERDC test-stand site for actual UXO. Several numerical results are presented and discussed for demonstrating the applicability of the proposed method for determining buried objects location as well as for discriminating between objects on interested from non-hazardous items.

In this paper the normalized surface magnetic charge model (NSMC) is employed for discriminating objects of interest, such as unexploded ordnances (UXO), from innocuous items, in cases when UXO electromagnetic induction (EMI) responses are contaminated by signals from other objects or magnetically susceptible ground. The model is designed for genuine discrimination and it is a physically complete, fast, and accurate forward model for analyzing EMI scattering. In the NSMC the overall EMI inverse problem can be summarized as follows: first, for any primary magnetic field the scattered magnetic field at selected points outside the object is recorded; and second, using the scattered field information an object buried object location, orientation and the amplitude of the NSMC are estimated. Finally, the total NSMC is used as a discriminant for distinguishing between UXO and non-UXO items. To illustrate the applicability of the NSMC algorithm, blind test data, which are collected at Cold Regions Research and Engineering Laboratory facility for actually buried objects under different type soil, are processed and analyzed.

GPR has achieved success against buried landmines in certain realisations such as handheld operation. There are however fundamental limitations in terms of propagation parameters, proximity to the ground surface, ground topography and bandwidth of operation. This paper discusses these limitations with reference to stand off landmine detection and with reference to published results establishes basic operating parameters within which GPR can operate successfully.

The U.S. Army Research Laboratory (ARL), as part of a mission and customer funded exploratory program, has developed a new low-frequency, ultra-wideband (UWB) synthetic aperture radar (SAR) for forward imaging to support the Army's vision of an autonomous navigation system for robotic ground vehicles. These unmanned vehicles, equipped with an array of imaging sensors, will be tasked to help detect man-made obstacles such as concealed targets, enemy minefields, and booby traps, as well as other natural obstacles such as ditches, and bodies of water. The ability of UWB radar technology to help detect concealed objects has been documented in the past and could provide an important obstacle avoidance capability for autonomous navigation systems, which would improve the speed and maneuverability of these vehicles and consequently increase the survivability of the U. S. forces on the battlefield. One of the primary features of the radar is the ability to collect and process data at combat pace in an affordable, compact, and lightweight package. To achieve this, the radar is based on the synchronous impulse reconstruction (SIRE) technique where several relatively slow and inexpensive analog-to-digital (A/D) converters are used to sample the wide bandwidth of the radar signals. We conducted an experiment this winter at Aberdeen Proving Ground (APG) to support the phenomenological studies of the backscatter from positive and negative obstacles for autonomous robotic vehicle navigation, as well as the detection of concealed targets of interest to the Army. In this paper, we briefly describe the UWB SIRE radar and the test setup in the experiment. We will also describe the signal processing and the forward imaging techniques used in the experiment. Finally, we will present imagery of man-made obstacles such as barriers, concertina wires, and mines.

For vehicle-mounted down-looking ground penetrating radar (DLGPR) systems, the largest response is typically due to the radar reflecting off the ground. Most DLGPR algorithms remove the ground bounce response as a first preprocessing step. The remaining subsurface response is then used to detect buried mines. It was observed that the ground bounce response over recently buried mines differs from the surrounding undisturbed soil. This suggests an approach in which the ground bounce response could be used to enhance detection performance. In this paper, we describe a technique for fusing the GPR ground bounce response with the GPS subsurface response to enhance mine detection performance. The technique is applied to data collected by a wide bandwidth impulse radar over buried mines in various soil conditions.

The purpose of this research is to jointly learn multiple classification tasks by appropriately sharing information between similar tasks. In this setting, examples of different tasks include the discrimination of targets from non-targets by different sonars or by the same sonar operating in sufficiently different environments. This is known as multi-task learning (MTL) and is accomplished via a Bayesian approach whereby the learned parameters for classifiers of similar tasks are drawn from a common prior. To learn which tasks are similar and the appropriate priors a Dirichlet process is employed and solved using mean field variational Bayesian inference. The result is that for many real-world instances where training data is limited MTL exhibits a significant improvement over both learning individual classifiers for each task as well as pooling all data and training one overall classifier. The performance of this method is demonstrated on simulated data and experimental data from multiple imaging sonars operating over multiple environments.

GE Security and the Naval Surface Warfare Center, Panama City (NSWC-PC) have collaborated to develop a magnetic gradiometer, called the Real-time Tracking Gradiometer or RTG that is mounted inside an unmanned underwater vehicle (UUV). The RTG is part of a buried mine hunting platform being developed by the United States Navy. The RTG has been successfully used to make test runs on mine-like targets buried off the coast of Florida. We will present a general description of the system and latest results describing system performance. This system can be also potentially used for other applications including those in the area of Homeland Security.

High-resolution sonar images of the sea floor contain rich spatial information that varies widely depending on survey location, sea state, and sensor platform-induced artifacts. Automatically segmenting sonar images into labeled regions can have several useful applications such as creating high-resolution bottom maps and adapting automatic target recognition schemes to perform optimally given the measured environment. This paper presents a method for sonar image segmentation using graphical models known as dynamic trees (DTs). A DT is a mixture of simply-connected tree-structured Bayesian networks (TSBNs), a hierarchical two-dimensional Bayesian network, where the leaf node states of each TSBN are the label of each image pixel. The DT segmentation task is to find the best TSBN mixture that represents the underlying data. A novel use of the K-distribution as a likelihood function for associating sonar image pixels with the appropriate bottom-type label is introduced. A simulated annealing stochastic search method is used to determine the maximum a posteriori (MAP) DT quadtree structure for each sonar image. Segmentation results from several images are presented and discussed.

In this paper, two different multi-aspect underwater target classification systems are evaluated based on their ability to correctly detect and classify mine-like objects. These methods are tested on a recently collected database that consists of sonar returns from various buried mine-like and non-mine-like objects in different operating and environmental conditions. In one approach, coherent features are extracted from the data using canonical correlation analysis (CCA) between two sonar pings. Classification is performed using a collaborative multi-aspect classifier (CMAC), which utilizes a group of collaborative decision-making agents capable of producing a high-confidence final decision based on these features. The second approach uses features generated by a multi-channel coherence analysis (MCA), which is an extension of CCA utilizing multiple sonar pings. The MCA features are then applied to a simple classifier. Results are presented in terms of correct classification rate and general detection and classification performance of each system in relation to the various operating and environmental conditions.

An improved computer-aided-detection / computer-aided-classification (CAD/CAC) processing string has been developed. The classified objects of 2 distinct strings are fused using the classification confidence values and their expansions as features, and using "summing" or log-likelihood-ratio-test (LLRT) based fusion rules. The utility of the overall processing strings and their fusion was demonstrated with new high-resolution dual frequency sonar imagery. Three significant fusion algorithm improvements were made. First, a nonlinear 2nd order (Volterra) feature LLRT fusion algorithm was developed. Second, a Box-Cox nonlinear feature LLRT fusion algorithm was developed. The Box-Cox transformation consists of raising the features to a to-be-determined power. Third, a repeated application of a subset feature selection / feature orthogonalization / Volterra feature LLRT fusion block was utilized. It was shown that cascaded Volterra feature LLRT fusion of the CAD/CAC processing strings outperforms summing, baseline single-stage Volterra and Box-Cox feature LLRT algorithms, yielding significant improvements over the best single CAD/CAC processing string results, and providing the capability to correctly call the majority of targets while maintaining a very low false alarm rate. Additionally, the robustness of cascaded Volterra feature fusion was demonstrated, by showing that the algorithm yields similar performance with the training and test sets.

Raytheon has extensively processed high-resolution sonar images with its CAD/CAC algorithms to provide real-time classification of mine-like bottom objects in a wide range of shallow-water environments. The algorithm performance is measured in terms of probability of correct classification (Pcc) as a function of false alarm rate, and is impacted by variables associated with both the physics of the problem and the signal processing design choices. Some examples of prominent variables pertaining to the choices of signal processing parameters are image resolution (i.e., pixel dimensions), image normalization scheme, and pixel intensity quantization level (i.e., number of bits used to represent the intensity of each image pixel). Improvements in image resolution associated with the technology transition from sidescan to synthetic aperture sonars have prompted the use of image decimation algorithms to reduce the number of pixels per image that are processed by the CAD/CAC algorithms, in order to meet real-time processor throughput requirements. Additional improvements in digital signal processing hardware have also facilitated the use of an increased quantization level in converting the image data from analog to digital format. This study evaluates modifications to the normalization algorithm and image pixel quantization level within the image processing prior to CAD/CAC processing, and examines their impact on the resulting CAD/CAC algorithm performance. The study utilizes a set of at-sea data from multiple test exercises in varying shallow water environments.

Detection and classification of underwater objects in sonar imagery are challenging problems. In this paper, a new coherent-based method for detecting potential targets in high-resolution sonar imagery is developed using canonical correlation analysis (CCA). Canonical coordinate decomposition allows us to quantify the changes between the returns from the bottom and any target activity in sonar images and at the same time extract useful features for subsequent classification without the need to perform separate detection and feature extraction. Moreover, in situations where any visual analysis or verification by human operators is required, the detected/classified objects can be reconstructed from the coherent features. In this paper, underwater target detection using the canonical correlations extracted from regions of interest within the sonar image is considered. Test results of the proposed method on underwater side-scan sonar images provided by the Naval Surface Warfare Center (NSWC) in Panama City, FL is presented. This database contains synthesized targets in real background varying in degree of difficulty and bottom clutter. Results illustrating the effectiveness of the CCA based detection method are presented in terms of probability of detection, and false alarm rates for various densities of background clutter.

Buried and surface laid landmines may be imaged by IR cameras. This paper considers some of the issues involved with processing images from trials and examines the pre processing and image recognitions algorithms for buried and surface laid mines.

To examine soil surface temperature evolution, a soil heat and water transfer model (HYDRUS-1d) is coupled to an atmospheric surface layer scheme. Idealized simulations are carried out for different meteorological conditions (wind speed and temperature). From the simulation results, the coupling between soil properties, surface temperature, and sensible heat flux is examined and implications for landmine thermal signatures are derived.

This paper presents the Swedish land mine and UXO detection project "Multi Optical Mine Detection System," MOMS, and the research carried out so far. The goal for MOMS is to provide knowledge and competence for fast detection of mines, especially surface laid mines, by the use of both active and passive optical sensors. A main activity was to collect information and gain knowledge about phenomenology; i.e. features or characteristics that can give a detectable signature or contrast between object and background, and to carry out a phenomenology assessment. A large effort has also been put into a scene description to support phenomenology assessment and provide a framework for further experimental campaigns. Also, some preliminary experimental results are presented and discussed.

In recent years, airborne minefield detection has increasingly been explored due to its capability for low-risk standoff detection and quick turnaround time. Significant research efforts have focused on the detection of surface mines and few techniques have been proposed specifically for buried mine detection. The detection performance of current detectors, like RX, for buried mines is not satisfactory. In this paper, we explore a methodology for buried mine detection in multi-spectral imagery, based on texture information of the target signature. A systematic approach for the selection of co-occurrence texture features is presented. Bhattacharya coefficient is used for the initial selection of discriminatory texture features, followed by principal feature analysis of the selected features, to identify minimum number of features with mutually uncorrelated information. Finally, a detection method based on unsupervised clustering of mine features in the reduced feature space, is employed for generating the test statistic for detection. Because the proposed method is based on co-occurrence matrix features, it is largely invariant to illumination changes in the images. Results for the proposed method are presented, which show improvement in the detection performance vis-a-vis multi-band RX anomaly detection, and validate the proposed clustering-based detection method.

The fundamental challenges of buried mine detection arise from the fact that the mean spectral signatures of disturbed soil areas that indicate mine presence are nearly always very similar to the signatures of mixed background pixels that naturally occur in heterogeneous scenes composed of various types of soil and vegetation. In our previous work, we demonstrated that MWIR images can be used to effectively detect the buried mines. In this work, we further improve our existing method by fusing multiple buried mine classifiers. For each target chip extracted from the MWIR image, we scan it in three directions: vertical, horizontal, and diagonal to construct three feature vectors. Since each cluster center represents all pixels in its cluster, the feature vector essentially captures the most significant thermal variations of the same target chip in three directions. In order to detect the buried mines using our variable length feature vectors, we have applied Kolmogorov-Smirnov (KS) test to discriminate buried mines from background clutters. Since we design one KS-based classifier for each directional scan, for the same target chip, there will be a total of three classifiers associated with vertical, horizontal, and diagonal scans. In our system, these three classifiers are applied to the same target chip, resulting in three independent detection results, which are further fused for the refined detection. Test results using actual MWIR images have shown that our system can effectively detect the buried mines in MWIR images with low false alarm rate.

A great development of technologies for the detection of buried landmines took place worldwide in the last years. In Argentina, a project for the development of an autonomous robot with sensors for landmines detection was recently approved by the Science and Technology National Agency. Within this project we are studying the detection of landmines by infrared radiation. Metallic and plastic objects with landmines shape and dimension were buried at different depths from 1 to 4 cm in soil and sand. Periodic natural warming by solar radiation or artificial warming by means of electric resistances or flash lamps were applied. Infrared images were obtained in the 8-12 micrometers spectral band with a microbolometer camera. The IR images were processed by different methods to obtain a definition as good as possible of the buried objects. After this a B-Spline method was applied to detect the targets contours and determine shape and dimensions of them so as to distinguish landmines from other objects. We are looking for a landmine detection method as simple and fast possible, with detection capability of metallic and plastic landmines and an acceptable false alarm rate which would be reduced when applied with other detection methods as GPR and electromagnetic induction. We present obtained and processed images and results obtained to distinguish buried landmines from other buried objects.

Presently, the number of landmines planted around the world totalizes more than 110 million and, far from slowing down, the landmine production planting rate is, at least, one order of magnitude higher than the rate at which they are removed. In this work a technique to detect buried landmines using boundary detection in IR images, is presented. The buried objects have different temperature than the surrounding soil. We find the object contours by means of an algorithm of B-Spline deformable curves. Under a statistical model, regions with different temperatures can be characterized by the values of the statistical parameters of these distributions. Therefore, this information can be used to find boundaries among different regions in the image. The B-Spline approach has been widely used in curve representation for boundary detection, shape approximation, object tracking and contour detection. Contours formulated by means of B-Splines allow local control, require few parameters and are intrinsically smooth. The algorithm consists in estimating the parameters along lines strategically disposed on the image. The true boundary is found when the values of these parameters vary abruptly on both sides. A likelihood function is maximized to determine the position of such boundaries. We present the experimental results, which show the behavior of the detection method, according to the buried object depth and the elapsed time from the cooling initial time. The obtained results exhibit that it is possible to recognize the shape of the objects, buried at different depths, with a low computational effort.

A typical minefield detection approach is based on a sequential processing employing mine detection and false alarm rejection followed by minefield detection. The current approach does not work robustly under different backgrounds and environment conditions because target signature changes with time and its performance degrades in the presence of high density of false alarms. The aim of this research will be to advance the state of the art in detection of both patterned and unpatterned minefield in high clutter environments. The proposed method seeks to combine false alarm rejection module and the minefield detection module of the current architecture by spatial-spectral clustering and inference module using a Markov Marked Point Process formulation. The approach simultaneously exploits the feature characteristics of the target signature and spatial distribution of the targets in the interrogation region. The method is based on the premise that most minefields can be characterized by some type of distinctive spatial distribution of "similar" looking mine targets. The minefield detection problem is formulated as a Markov Marked Point Process (MMPP) where the set of possible mine targets is divided into a possibly overlapping mixture of targets. The likelihood of the minefield depends simultaneously on feature characteristics of the target and their spatial distribution. A framework using "Belief Propagation" is developed to solve the minefield inference problem based on MMPP. Preliminary investigation using simulated data shows the efficacy of the approach.

A collaborative program has been undertaken by the UK and US Governments to develop Countermine Capabilities for Medium/Future Forces. The program is conducting research into a ground-based system for the detection and countering of land mines on military routes. The overall objective of the program is to jointly develop and then evaluate a demonstration system prototype. This project was established as a three stage program. The first stage established a common UK/US military requirement and conducted operational analysis based on generic sensors. Once the requirement and analysis were established, candidate technologies appropriate to the timeframe of the program were assessed according to their Technology Readiness Level (TRL). The program is currently in the second stage which is taking technologies identified from the first stage and performing trials in both the UK and US aimed at a more detailed understanding of their baseline performance. A trial in the UK was completed in 2005 where two US vehicle mounted sensor systems and one UK vehicle mounted sensor system were trialled. The UK sensor system is described herein and consisted of three Electro- Optic (EO) sensors that covered the visible, medium wave infra-red (IR) and long wave IR bands. The set-up of the UK trial site and the assembly of the UK EO sensor system are discussed. Analysis of the trial data and preliminary research on the feasibility of fusing data from the EO sensors are discussed.

DRDC Suffeld and Itres Research have jointly investigated the use of visible and infrared hyperspectral imaging (HSI) for surface and buried land mine detection since 1989. These studies have demonstrated reliable passive HSI detection of surface-laid mines, based on their reflectance spectra, from airborne and ground-based platforms. Commercial HSI instruments collect and store image data at aircraft speeds, but the data are analysed off- line. This is useful for humanitarian demining, but unacceptable for military countermine operations. We have developed a hardware and software system with algorithms that can process the raw hyperspectral data in real time to detect mines. The custom algorithms perform radiometric correction of the raw data, then classify pixels of the corrected data, referencing a spectral signature library. The classification results are stored and displayed in real time, that is, within a few frame times of the data acquisition. Such real-time mine detection was demonstrated for the first time from a slowly moving land vehicle in March 2000. This paper describes an improved system which can achieve real-time detection of mines from an airborne platform, with its commensurately higher data rates. The system is presently compatible with the Itres family of visible/near infrared, short wave infrared and thermal infrared pushbroom hyperspectral imagers and its broadband thermal infrared pushbroom imager. Experiments to detect mines from an airborne platform in real time were conducted at DRDC Suffield in November 2006. Surface-laid land mines were detected in real time from a slowly moving helicopter with generally good detection rates and low false alarm rates. To the authors' knowledge, this is the first time that land mines have been detected from an airborne platform in real time using hyperspectral imaging.

DRDC Suffeld and Itres Research have collaborated to investigate the use of hyperspectral imaging (HSI) for surface and buried landmine detection since 1989. Visible/near infrared (casi) and short wave infrared (sasi) families of imagers have been developed which have demonstrated reliable HSI detection of surface-laid mines, based on their reflectance spectra, from airborne and ground-based platforms. However, they have limited ability to detect buried mines. Thermal infrared (TIR) HSI may have the capability to detect buried mines. Disturbance of quartz-bearing soils has been shown to measurably change their TIR emissivity spectra due to mixing of surface/subsurface soil (restrahlen band intensities vary with particle size). Some evidence suggests that the effect can persist months after the visible disturbance has disappeared. Carbonates and other materials exhibit similar TIR spectral features and heat flow anomalies caused by buried mines can also be measured in the TIR band. There are no commercially available TIR hyperspectral imagers that are suitable for mine detection. The very few possibly suitable imagers are one-of-a-kind research instruments, dedicated to internal programs and not available for the general mine detection community. A TIR hyperspectral imager (tasi) based on a novel optical design and a cooled MCT focal plane array has been developed. The instrument has been designed with landmine detection in mind. First light images from the prototype were obtained in summer 2006 and initial test flights were completed in fall 2006. The design of the instrument and a comparison with design alternatives in the context of mine detection requirements is discussed. Preliminary images are presented.

In this paper we discuss our efforts to develop a digital hyperspectral signature model for terrain features. These efforts focused on developing models for the optical and infrared properties of soils and vegetation features. A detailed geometric model of the particulate soil surface is combined with radiative transport algorithms to compute emissive and reflective properties of terrain surfaces. These models support analysis of environmental effects on terrain signatures, signature generation to support object detection algorithm development, and terrain clutter characterization. Environmental effects modeled and studied include solar insolation, wind, soil particle size distributions, and soil constituents. The particulate nature of the soil surface is shown to significantly impact the optical properties. A discussion of the terrain modeling approach, the underlying analytical basis, the methodology for modeling solar absorptivity, and results from the model computations will be presented.

Over the years several methods have been used to determining the best bands for a visible near IR multi-spectral sensor. The most popular method, the committee method, places scientists with differing opinions on the phenomena and the sensor mission in one room, and a compromise set is developed. To avoid this, there have been several methods to automate this selection process. We have developed a method to examine hyperspectral data to find the best multi-spectral band set (whether 3, 4, 5 or 6 bands) based on the background, on the premise that, with the target unknown, the band set that best separates the background materials is the best. We start with a hyperspectral data set of a background area without any targets. We then run a program for determining the spectral endmembers. Any endmembers that look like they are due to sensor artifacts or an anomalous point on the ground (junk) are discarded from the list. The resulting hyperspectral endmembers are then input to an exhaustive search program. The goal of the exhaustive search is to find a set of N (say 4) multi-spectral bands that maximizes the spectral angles between all of the endmembers. Thus, at each trial the multi-spectral bands are made by binning the hyperspectral (to four bands in this case) and the spectral angles calculated between endmembers 1 and 2, 1 and 3, 1 and 4, 2 and 3, 2 and 4 etc. The endmembers in each case have been binned to four multi-spectral bands. We save the average of these spectral angle calculations. After examining often millions of combinations, the multi-spectral band set that maximizes the spectral separation is judged to be the best. We have applied this method to the selection of multi-spectral bands sets for several sensors.

An extension of the Iterated Constrained Endmembers (ICE) algorithm that incorporates sparsity promoting priors to find the correct number of endmembers is presented. In addition to solving for endmembers and endmember fractional maps, this algorithm attempts to autonomously determine the number of endmembers required for a particular scene. The number of endmembers is found by adding a sparsity-promoting term to ICE's objective function. This method is applied to long wave infrared, LWIR, hyperspectral data to seek out vegetation endmembers and define a vegetation mask for the reduction of false alarms in landmine data.

The UK Department for International Development (DfID), in collaboration with the German Foreign Ministry (Auswärtiges Amt), contracted ERA Technology to carry out extensive field trials in Cambodia, Bosnia and Angola of an advanced technology, dual sensor, and hand-held landmine detector system called MINEHOUND^TM. This detector combines a metal detector with a Ground Penetrating Radar (GPR). As a result of extremely successful trials MINEHOUND^TM was developed as a product by ERA Technology and Vallon GmbH and has been available for sale since late 2006. This paper describes the transition to production of the detector.

Since 2002, we have developed a new hand-held land mine detection dual-sensor ALIS. ALIS is equipped with a metal detector and a GPR, and it has a sensor tracking system, which can record the GPR and Metal detector signal with its location. It makes possible to process the data afterwards, including migration. The migration processing drastically increases the quality of the image of the buried objects. The new system, we do not need any standard mark on the ground. Also, ALIS uses two different GPOR systems, including VNA (Vector Network Analyzer) based GPR and an Impulse GPR. VNA based GPR can provide better quality GPR images, although the impulse GPR is faster and light weight. ALIS evaluation tests were held in mine affected courtiers including Afghanistan, Croatia, Egypt and Cambodia. In the two-month evaluation test in Cambodia, ALIS worked without any problem. After some demonstrations and evaluation, we got many useful suggestions. Using these advises, we have modified the ALIS and it is now more easy to use. ALIS will be commercialized in 2007.

We are developing a new hand-held land mine detection dual-sensor (ALIS) which is equipped with a metal detector and a GPR. ALIS is equipped with a sensor tracking system, which can record the GPR and Metal detector signal with its location. It makes possible to process the data after the data was acquired, including migration. The migration processing drastically increases the quality of the images of the buried objects. Evaluation test of ALIS has been conducted in several test sites. In February 2006, a one-month evaluation test was conducted in Croatia, and in October- December 2006, a two-month evaluation test was conducted in Croatia. Since the dual-sensor is a new landmine detection sensor, and the conventional evaluation procedure developed for metal detectors cannot directly be applied for the dual sensor. In Croatia, the detection probability was comparable to that by a metal detector operated by local deminers. In addition, we showed that ALIS provides image of buried objects by GPR, which can be used for identification. Therefore, their performances were sufficiently high. Then the test was also conducted in Cambodia. The test was carried out by 2 local deminers independently, which allows studying the influence of different operators and increases the statistical value of the results.

This paper presents an experimental design and the evaluation result of a trial that were carried out from 1 February to 9 March 2006 using real PMA-1A and PMA-2 landmines at the Benkovac test site in Croatia. The objective of the Croatia- Japan joint trial is to evaluate dual sensor systems, which use both ground penetrating radar (GPR) and electromagnetic inductive (EMI) sensors. A comparative trial was also carried out by Croatian deminers using an existing EMI sensor, i.e., a metal detector (MD). The trial aims at evaluating differences in performance between dual sensors and MDs, especially in terms of discrimination of landmines from metal fragments and extension of detectable range in the depth direction. Devices evaluated here are 4 prototypes of anti-personnel landmine detection systems developed under a project of the Japan Science and Technology Agency (JST), the supervising authority of which is the Ministry of Education, Culture, Sports, Science and Technology (MEXT). The prototypes provide operators with subsurface images, and final decision whether a shadow in the image is a real landmine or not is left to the operator. This is similar to the way that medical doctors find cancer by reading CT images. Since operators' pre-knowledge of locations of buried targets significantly influences the test result, three test lanes, which have 3 different kinds of soils, have been designed to be suitable for blind tests. The result showed that the dual sensor systems have a potential to discriminate landmines from metal fragments and that probability of detection for small targets in mineralized soils can be improved by using GPR.

Both force protection and humanitarian demining missions require efficient and reliable detection and discrimination of buried anti-tank and anti-personnel landmines. Widely varying surface and subsurface conditions, mine types and placement, as well as environmental regimes challenge the robustness of the automatic target recognition process. In this paper we present applications created for the U.S. Army Nemesis detection platform. Nemesis is an unmanned rubber-tracked vehicle-based system designed to eradicate a wide variety of anti-tank and anti-personnel landmines for humanitarian demining missions. The detection system integrates advanced ground penetrating synthetic aperture radar (GPSAR) and electromagnetic induction (EMI) arrays, highly accurate global and local positioning, and on-board target detection/classification software on the front loader of a semi-autonomous UGV. An automated procedure is developed to estimate the soil's dielectric constant using surface reflections from the ground penetrating radar. The results have implications not only for calibration of system data acquisition parameters, but also for user awareness and tuning of automatic target recognition detection and discrimination algorithms.

Multimodal detection of subsurface targets such as tunnels, pipes, reinforcement bars, and structures has been investigated using both ground-penetrating radar (GPR) and seismic sensors with signal processing techniques to enhance localization capabilities. Both systems have been tested in bi-static configurations but the GPR has been expanded to a multi-static configuration for improved performance. The use of two compatible sensors that sense different phenomena (GPR detects changes in electrical properties while the seismic system measures mechanical properties) increases the overall system's effectiveness in a wider range of soils and conditions. Two experimental scenarios have been investigated in a laboratory model with nearly homogeneous sand. Images formed from the raw data have been enhanced using beamforming inversion techniques and Hough Transform techniques to specifically address the detection of linear targets. The processed data clearly indicate the locations of the buried targets of various sizes at a range of depths.

Can human vision supplement the information that handheld landmine detection equipment provides its operators to increase detection rates and reduce the hazard of the task? Contradictory viewpoints exist regarding the viability of visual detection of landmines. Assuming both positions are credible, this work aims to reconcile them by exploring the visual information produced by landmine burial and how any visible signatures change as a function of time in a natural environment. Its objective is to acquire objective, foundational knowledge on which training could be based and subsequently evaluated. A representative set of demilitarized landmines were buried at a field site with bare soil and vegetated surfaces using doctrinal procedures. High resolution photographs of the ground surface were taken for approximately one month starting in April 2006. Photos taken immediately after burial show clearly visible surface signatures. Their features change with time and weather exposure, but the patterns they define persist, as photos taken a month later show. An analysis exploiting the perceptual sensitivity of expert observers showed signature photos to domain experts with instructions to identify the cues and patterns that defined the signatures. Analysis of experts' verbal descriptions identified a small set of easily communicable cues that characterize signatures and their changes over the duration of observation. Findings suggest that visual detection training is viable and has potential to enhance detection capabilities. The photos and descriptions generated offer materials for designing such training and testing its utility. Plans for investigating the generality of the findings, especially potential limiting conditions, are discussed.

NATO Task group SCI-133 on "Countermine Technologies" under the NATO RTO (Systems Concepts and Integration) panel, has the goal to identify technologies for mine detection (close-in detection and remote detection) and mine neutralization (breaching, route clearing, area clearance) which provide the best short, medium and long-term potential in countermines operations. A brief overview of the activities will be given. The main focus of the paper is on the work of the NATO RTO SCI-133 task group on test and evaluation procedures for landmine-detection/neutralization and mechanical clearance equipment. Interpreting test results in test reports, is often difficult due to incomplete descriptions of test procedures and lack of clear definitions of test parameters. Task group SCI-133 prepared a list of issues that should be addressed in test reports of landmine-detection equipment. These guidelines also give references to documents containing relevant definitions. The presentation is in the form of a checklist of questions to be answered when designing, conducting, and reporting on test and evaluation efforts.

This paper reviews landmine neutralisation and marking systems and assesses how they can be down-selected for incorporation into a technology demonstrator system. The aim will be to illustrate detection, marking and route clearance capabilities against various anti-tank mines. A technology comparison matrix has been constructed to allow the down selection of technologies according to defined criteria. The matrix captures information from a top-level review of a broad range of neutralisation and marking techniques both currently in use and in development. The methodology allows filtering of technologies using the matrix and is flexible enough to take into account a range of operational scenarios and requirements. The results of an initial sub-system down-selection are shown. The requirements for the final technology down selection with respect to the overall system concepts are discussed. The application of this technique for down selecting technologies/methods in a broader system context is highlighted.

To understand the fate and transport mechanisms of TNT from buried landmines is it essential to determine the adsorption process of TNT on soil and clay minerals. In this research, soil samples from horizons Ap and A from Jobos Series at Isabela, Puerto Rico were studied. The clay fractions were separated from the other soil components by centrifugation. Using the hydrometer method the particle size distribution for the soil horizons was obtained. Physical and chemical characterization studies such as cation exchange capacity (CEC), surface area, percent of organic matter and pH were performed for the soil and clay samples. A complete mineralogical characterization of clay fractions using X-ray diffraction analysis reveals the presence of kaolinite, goethite, hematite, gibbsite and quartz. In order to obtain adsorption coefficients (K_d values) for the TNT-soil and TNT-clay interactions high performance liquid chromatography (HPLC) was used. The adsorption process for TNT-soil was described by the Langmuir model. A higher adsorption was observed in the Ap horizon. The Freundlich model described the adsorption process for TNT-clay interactions. The affinity and relative adsorption capacity of the clay for TNT were higher in the A horizon. These results suggest that adsorption by soil organic matter predominates over adsorption on clay minerals when significant soil organic matter content is present. It was found that, properties like cation exchange capacity and surface area are important factors in the adsorption of clayey soils.

Detection of explosive-related chemicals (ERCs) derived from landmines sources is influenced by fate and transport processes. Characterization and quantification of the effects of environmental factors on the fate and transport behavior ERCs near soil surface environments requires the development of physical models that can mimic the conditions found in the field. The development of the scalable systems and methods involves proper reproduction of soil composition, lithology and structures, appropriate placement of boundary conditions, and suitable simulation of representative environmental conditions. This paper evaluates the ability of different packing methods for clayey soils to attain physical and transport properties representative of field conditions, and which can yield reproducible results across different scales and dimensions. Characteristics and reproducibility of packing properties is evaluated in terms of soil bulk density, porosity, flow capacity and particle size distribution. The packing methods were tested under different water content conditions and they are described as infiltration packing, saturation packing, plastic limit packing, inverse infiltration packing, induced settling packing, and vibration packing. The systems were evaluated for consistent bulk density, porosity, flow capacity and particle size distribution with depth. Preliminary results exhibit satisfactory bulk density and porosity values for the vibration packing method under field water content conditions, ranging from 1.15 to 1.31 g cm-3 and from 42 to 44%, respectively. This method also shows acceptable flow capacity and the particle size distribution that is found in the field.

As part of a large research program aiming to the development of chemical sensor for detecting landmines, we have studied the fate and transport of TNT subject to different ambient parameters. The space and temporal concentration profiles of TNT, and its degradation compounds have been measured using soil tanks. The following ambient parameters were controlled to emulate environmental factors: water content, temperature, relative humidity, and UV-VIS radiation. A series of soil tanks were kept under controlled conditions for longer than a year and sampled periodically at the surface. After several months, all tanks were sampled vertically and disposed of. Chromatography (GC-&mgr;ECD) with direct injection was used for the analysis of the samples. Of particular interest is the presence of several degradation compounds, as time evolves, responding to the ambient parameters imposed. The vertical concentration profiles of the several chemicals found, gives an interesting view of the degradation process as well as of the transport mechanisms. The results agreed with our computer simulations, and are used to validate previous numerical analyses.

Chemical, biological, and canine detection of buried explosive devices (BEDs) rely on the presence of explosive related compounds (ERCs) near the soil-atmospheric surface. ERC distribution near this surface and its relation to the location of BEDs is controlled by fate and transport processes. Experimental work was conducted in a 3D laboratory-scale SoilBed system to determine the effect of cyclic rainfall, evaporation, temperature, and solar radiation on on the fate, transport and detection of ERCs near soil surfaces. Experiments were conducted by burying a TNT/DNT source under the soil surface, and applying different rainfall and light radiation cycles while monitoring salt tracers and TNT solute concentrations temporally and spatially within the SoilBed. Transport of non-reactive solutes was highly influenced by the cyclic variations on water flux, water content, evaporation, and influx concentrations. Concentrations of TNT and other ERCs were further affected by vapor transport and sorptive and degradation processes.

The existence of explosive related chemicals (ERCs) near the soil-atmospheric and other surfaces depend on their fate and transport characteristics within the environmental settings. Consequently, detection of ERC in environmental matrices is influenced by conditions that affect their fate and transport. Experimental work to study the fate and transport behavior of ERCs relies on proper temporal and spatial sampling techniques. Because the low vapor pressure of these chemicals and their susceptibility to adsorption and degradation, vapor concentrations in environmental matrices are very low. Depending on the environmental conditions, the amount of samples that can be withdrawn for analysis is also limited. It is, therefore, necessary to develop sampling technologies that can provide quantitative measures of ERC concentrations in limited sampling environments. This paper presents experimental work conducted to develop a sampling technique to quantify DNT and TNT vapor concentrations of low vapor-pressure ERCs in environmental setting having limited sampling volumes and large sample numbers. Two potential vapor sampling techniques, Solid phase Microextraction (SPME) and Solid Phase Extraction (SPE), were developed and evaluated. SPME sampling techniques are excellent to quantify for DNT and TNT at very low concentrations. Its passive sampling capabilities meet the requirement for low-volume environmental sampling, but measured concentrations may be lagged in time. SPMEs' requirements for immediate analysis after sampling limit the technique for continuous vapor sampling. SPE showed to be a sensitive and reproducible technique to determine vapor concentrations of TNT and DNT in atmospheric and soil setting having limited sampling volumes and large sample numbers. Smallvolume (600&mgr;L) air samples provide measurements in the &mgr;gL^-1 concentration range using isoamyl acetate and acetonitrile as the solvents. Small extraction volumes make this technique cost efficient and attractive. Issues with extraction inefficiencies, however, were observed and are being investigated.

Chemical detection of buried explosives devices (BEDs) through chemical sensing is influenced by factors affecting the transport of chemical components associated with the devices. Explosive-related chemicals, such as 2,4-dinitrotolune (DNT), are somewhat volatile and their overall transport is influenced by vapor-phase diffusion. Gaseous diffusion depends on environmental and soil conditions. The significance of this mechanism is greater for unsaturated soil, and increases as water content decreases. Other mechanisms, such as sorption and degradation, which affect the overall fate and transport, may be more significant under diffusion transport due to the higher residence time of ERCs in the soil system. Gaseous diffusion in soil was measured using a one-dimensional physical model (1-D column) to simulate the diffusion flux through soil under various environmental conditions. Samples are obtained from the column using solid phase microextraction (SPME) and analyzed with a gas chromatography. Results suggest that DNT overall diffusion is influenced by diffusive and retention processes, water content, source characteristics, and temperature. DNT effective gas phase diffusion in the soil decreases with increasing soil water content. Vapor transport retardation was more dominant at low water contents. Most of the retardation is associated to the partition of the vapor to the soil-water. DNT vapor flux is higher near the explosive source (mine) than at the soil surface. This flux also increases with higher soil water content and temperature. Results also suggest non-equilibrium transport attributed to mass transfer limitations and non-linear sorption.

Quadrupole Resonance sensors have the unique capability of detecting explosives from buried, plastic-cased antipersonnel and antitank landmines. The chemical specificity of this radio-frequency technique provides the potential to deliver remarkably low false alarm rates during landmine detection. This is of particular importance to deminers, who frequently come across numerous clutter items before uncovering a mine. Quadrupole Resonance is typically utilized in a confirmation mode; preceded by rapid primary scans carried out by, for example, metal detectors, ground penetrating radars or a fusion of these. Significant technical and scientific advances have resulted in the fabrication of handheld and vehicle mounted Quadrupole Resonance landmine detectors in compact, power-efficient configurations. The development work is focused on baseline sensitivity increase, as well as the achievement of high detection performance under field conditions. The mine detection capability of Quadrupole Resonance detectors has been evaluated during various blind tests. A modular handheld unit, combining primary and confirmation sensors, was designed to be operated by a single person. A series of field tests demonstrate the unique capability of Quadrupole Resonance for significant false alarm reduction.

Current results are described on the research and development of the advanced humanitarian landmine detection system by using a compact discharge-type fusion neutron source called IECF (Inertial-Electrostatic Confinement fusion) devices. With a 50 mm-thick water-jacketed IEC device (IEC20C) of 200 mm inner diameter can have produced 10⁷ neutrons/s stably in CW mode for 80 kV and 80 mA. Ample 10.8 MeV γ-rays produced through (n, γ) reaction with nitrogen atoms in the melamine (C₃H₆N₆) powder (explosive simulant) are clearly measured by a BGO-NaI-combined scintillation sensor with distinct difference in case of with/without melamine, indicating identification of the buried landmines feasible.

We investigated the sensitivity of NQR for explosives such as trinitrotoluene (TNT) and cyclotrimethylenetrinitramine (RDX), the main constituent of explosives of landmines. We succeeded in the remote detection of RDX from 8 cm away using NQR. We have developed a prototype of an NQR landmine detector.

Solid phase microextraction (SPME) has been coupled with liquid chromatography to widen its range of application to nonvolatile and thermally unstable compounds, generally limited for SPME-GC. A method for analysis of nitroaromatic explosives and its degradations products was developed by coupling SPME and high performance liquid chromatography with ultraviolet detection (HPLC/UV), introducing a modified interface that ensure accuracy, precision, repeatability, high efficiency, unique selectivity and high sensitive to detection and quantification of explosives from surface soil samples and increased chromatographic efficiency. A pretreatment step was introduced for the soil samples which extracted the target compounds into an aqueous phase. Several parameters that affect the microextraction were evaluated, such as: fiber coating, adsorption and desorption time and stirring rate. The effect of salting out (NaCl) on analyte extraction and the role of various solvents on SPME fiber were also evaluated. Carbowax-templated resin (CW/TPR) and Polydimethilsiloxane-divinilbenzene (PDMS-DVB) fibers were used to extract the analytes from the aqueous samples. Explosives were detected at low &mgr;g/mL concentrations. This study demonstrates that SPME-HPLC is a very promising method of analysis of explosives from aqueous samples and has been successfully applied to the determination of nitroaromatic compounds, such as TNT.

The anglar dependence of emitted NQR signal intensity from a polycrystalline hexamethylenetetramine has been investigated. Measurement from the radial direction reveals that the NQR signal from a long column sample showed a very inhomogeneous radiation pattern which has strong signal along the direction of the excitation and few along the perpendicular direction from the excitation axis. A series of measurement by a receiver set face to face to the sample at every 10° from 0° to 350° from the excitation direction revealed that the signal intensity measured has a trigonometric divergence. This is useful to design an antenna coil of a landmine detector to get strong NQR signal remotely.

Previous work by the authors using information-based sensor management for static target detection has utilized a probability of error performance metric that assumes knowledge of the number of targets present in a grid of cells. Using this probability of error performance metric, target locations are estimated as the N cells with the largest posterior state probabilities of containing a target. In a realistic application, however, the number of targets is not known a priori. The sequential probability ratio test (SPRT) developed by Wald is therefore implemented within the previously developed sensor management framework to allow cell-level decisions of "target" or "no target" to be made based on the observed sensor data. Using these cell-level decisions, more traditional performance metrics such as probability of detection and probability of false alarm may then be calculated for the entire region of interest. The resulting sensor management framework is implemented on a large set of data from the U.S. Army's autonomous mine detection sensors (AMDS) program that has been collected using both ground penetrating radar (GPR) and electromagnetic induction (EMI) sensors. The performance of the sensor manager is compared to two different direct search techniques, and the sensor manager is found to achieve the same P_d performance at a lower cost than either of the direct search techniques. Furthermore, uncertainty in the sensor performance characteristics is also modeled, and the use of uncertainty modeling allows a higher P_d to be obtained than is possible when uncertainty is not modeled within the sensor management framework.

Many automatic target detection (ATD) algorithm suites include a prescreener as an initial link in their processing chains. It assists the downstream algorithms by eliminating many potential false alarms while still retaining a large percentage of the objects of interest, thereby allowing for greater specialization by the downstream algorithms. Many such prescreeners have been implemented for individual sensors-for example the constant false alarm rate detectors of radar systems or the RX (Reed-Xiaoli) detection algorithms of hyperspectral imaging (HSI) systems. In this paper we examine straightforward methods for fusing the outputs from synthetic aperture radar (SAR) and HSI prescreeners to create a multi-sensor prescreening algorithm. We begin by examining the sensor phenomenology for a specific operational scenario, and we incorporate this phenomenological information into both the individual sensor prescreener designs and the final fusion algorithm design. We describe how the SAR and HSI prescreener detects are associated with one another prior to fusion. Finally, we describe multiple fusion methodologies-namely a method based on the Dempster-Shafer theory of evidence and a method based on a Bayesian approach under an assumption of independence. We compare results from each fusion algorithm with those obtained using a single sensor.

This paper examines the confidence level fusion of several promising algorithms for the vehicle-mounted ground penetrating radar landmine detection system. The detection algorithms considered here include Edge Histogram Descriptor (EHD), Hidden Markov Model (HMM), Spectral Correlation Feature (SCF) and NUKEv6. We first form a confidence vector by collecting the confidence values from the four individual detectors. The fused confidence is assigned to be the difference in the square of the Mahalanobis distance to the non-mine class and the square of the Mahalanobis distance to the mine class. Experimental results on a data collection that contains over 1500 mine encounters indicate that the proposed fusion technique can reduce the false alarm rate by a factor of two at 90% probability of detection when compared to the best individual detector.

We present a novel method for fusing the results of multiple landmine detection algorithms that use different types of features and different classification methods. The proposed fusion method, called Context-Dependent Fusion (CDF) is motivated by the fact that the relative performance of different detectors can vary significantly depending on the mine type, geographical site, soil and weather conditions, and burial depth. The training part of CDF has two components: context extraction and algorithm fusion. In context extraction, the features used by the different algorithms are combined and used to partition the feature space into groups of similar signatures, or contexts. The algorithm fusion component assigns an aggregation weight to each detector in each context based on its relative performance within the context. Results on large and diverse Ground Penetrating Radar data collections show that the proposed method can identify meaningful and coherent clusters and that different expert algorithms can be identified for the different contexts. Our initial experiments have also indicated that the context-dependent fusion outperforms all individual detectors.

The Borda Count was proposed as a method of ranking candidates by combining the rankings assigned by multiple voters. It has been studied extensively in the context of its original use in political elections and social choice-making. It has recently seen use in machine learning and in ranking web searches, but few of its formal properties have been extensively investigated. In this paper, we describe unsupervised, and (barely) supervised learning systems that employ the Borda Count as their underlying bases. We analyze the strengths and weaknesses of the technique in the context of landmine discrimination. We discuss and evaluate methods for algorithm fusion using several weighted Borda Count approaches and show how they affect algorithm fusion performance.

Provided is a summary of Holographic Neural Technology (HNeT) and its application in detecting land mines using airborne Synthetic Aperture Radar (SAR) imagery. Tests were performed for three surface mine classes (small metallic, large metallic, and medium-sized plastic) located within variable indigenous background clutter (bare dirt, short/tall grass). This work has been performed as part of the Wide Area Airborne Minefield Detection (WAAMD) Program at the U. S. Army Night Vision Labs and Electronic Sensors Directorate in Fort Belvoir, VA. The ATR algorithm applied was Holographic Neural Technology (HNeT); a neuromorphic model based upon non-linear phase coherence/de-coherence principles. The HNeT technology provides rapid learning capabilities and an advanced capability in learning and generalization of non-linear relationships. Described is a summary of the underlying HNeT technology and the methodologies applied in the training of the neuromorphic system for mine detection using target images (land mines) and back ground clutter images. Provided also is a summary description of the software tools applied in the development of the mine detection capability. Performance testing of the mine detection algorithm separated training and testing sensor image sets by airborne sensor depression angle and surface ground condition indigenous to site location (Countermine Alpha, Yellow Sands). Detection performance was compared in the analysis of complex versus magnitude sensor data. Performance results from independent test imagery indicated a reasonable level of clutter rejection, providing > 50% probability of detection at a false detection rate < 10^-3/m². A description of the test scenarios applied and performance results for these scenarios are summarized in this report.

A Sequential Monte Carlo (SMC) method is proposed to locate the ground bounce (GB) positions in 3D data collected by ground penetrating radar (GPR) system. The algorithm is verified utilizing real data and improved landmine detection performance is achieved compared with three other GB trackers.

Magnetic soils are a major source of false positives when searching for landmines or unexploded ordnance (UXO) with electromagnetic induction sensors. In adverse areas up to 30% of identified electromagnetic (EM) anomalies are attributed to geology. The main source of the electromagnetic response is the magnetic viscosity of the ferrimagnetic minerals magnetite and maghaemite. The EM phenomena that give rise to the response of magnetically viscous soil and metal are fundamentally different. The viscosity effects of magnetic soil can be accurately modelled by assuming a ferrite relaxation with a log-uniform distribution of time constants. The EM response of a metallic target is due to eddy currents induced in the target and is a function of the target's size, shape, conductivity and magnetic susceptibility. In this presentation, we consider different soil compensation techniques for time domain and frequency domain EM data. For both types of data we exploit the EM characteristics of viscous remnantly magnetized soil. These techniques will be demonstrated with time domain and frequency domain data collected on Kaho'olawe Island, Hawaii. A frequency domain technique based on modeling a negative log-linear in-phase and constant quadrature component was found to be very effective at suppressing false-alarms due to magnetic soils.

Hidden Markov Models (HMMs) are useful tools for landmine detection and discrimination using Ground Penetrating Radar (GPR). The performance of HMMs, as well as other feature-based methods, depends not only on the design of the classifier but on the features. Traditionally, algorithms for learning the parameters of classifiers have been intensely investigated while algorithms for learning parameters of the feature extraction process have been much less intensely investigated. In this paper, we describe experiments for learning feature extraction and classification parameters simultaneously in the context of using hidden Markov models for landmine detection.

In this paper, we have addressed the problem of visual inspection, recognition, and discrimination of UXO based on computer vision techniques and introduced three complimentary color, texture, and shape classifiers. The proposed technique initially enhances an image taken from an UXO site and removes terrain background. Next, it applies a blob detector to detect the salient objects of the environment. The UXO classification begins with a perceptive color classifier that classifies the found salient objects based on their color hues. The color classifier attempts to differentiate and classify the color of salient objects based on the color hue information of some known UXO objects in the database. A color ranking scheme is applied for ranking color hue likelihood of the salient objects in the environment. Next, an intuitive texture classifier is applied to characterize the surface texture of the salient objects. The texture signature is used to disjointedly discriminate objects whose surface texture properties matching the priori known UXO textures. Lasting, an intuitive Object Shape Classifier is applied to independently arbitrate the classification of the UXO. Three soft computing methods were developed for robust decision fusion of three UXO feature classifiers. These soft computing techniques include: a statistical-based genetic algorithm, a hamming neural network, and a fuzzy logic algorithm. In this paper, we present details of the UXO feature classifiers and discuss the performance of three decision fusion methods for fusion of results from the three UXO feature classifiers. The main contributing factor of this work is toward designing an ultimate fully-automated tele-robotic system for UXO classification and decontamination.

As ground penetrating radar sensor phenomenology improves, more advanced statistical processing approaches become applicable to the problem of landmine detection in GPR data. Most previous studies on landmine detection in GPR data have focused on the application of statistics and physics based prescreening algorithms, new feature extraction approaches, and improved feature classification techniques. In the typical framework, prescreening algorithms provide spatial location information of anomalous responses in down-track / cross-track coordinates, and feature extraction algorithms are then tasked with generating low-dimensional information-bearing feature sets from these spatial locations. However in time-domain GPR, a significant portion of the data collected at prescreener flagged locations may be unrelated to the true anomaly responses - e.g. ground bounce response, responses either temporally "before" or "after" the anomalous response, etc. The ability to segment the information-bearing region of the GPR image from the background of the image may thus provide improved performance for feature-based processing of anomaly responses. In this work we will explore the application of Markov random fields (MRFs) to the problem of anomaly/background segmentation in GPR data. Preliminary results suggest the potential for improved feature extraction and overall performance gains via application of image segmentation approaches prior to feature extraction.

We propose a general method for detecting landmine signatures in vehicle mounted ground penetrating radar (GPR) using discrete hidden Markov models and Gabor wavelet features. Observation vectors are constructed based on the expansion of the signature's B-scan using a bank of scale and orientation selective Gabor filters. This expansion provides localized frequency description that gets encoded in the observation sequence. These observations do not impose an explicit structure on the mine model, and are used to naturally model the time-varying signatures produced by the interaction of the GPR and the landmines as the vehicle moves. The proposed method is evaluated on real data collected by a GPR mounted on a moving vehicle at three different geographical locations that include several lanes. The model parameters are optimized using the BaumWelch algorithm, and lane-based cross-validation, in which each mine lane is in turn treated as a test set with the rest of the lanes used for training, is used to train and test the model. Preliminary results show that observations encoded with Gabor wavelet features perform better than observation encoded with gradient-based edge features.

This study looks at application of the Kullback-Leibler distance to classification in landmine discrimination. The paper explores the relationship between the information theoretic concepts of the Kullback-Leibler divergence and mutual information with special attention to the asymmetry of the typical formulation of the Kullback-Leibler distance. It shows how to use the Kullback-Leibler distance to discriminate between mines and nonmines by performing Fuzzy C-Means clustering of the density distribution of landmine electromagnetic induction signatures. Finally, it reports cross-validation results of applying these techniques to a large collection of actual landmine data and compares the results to those yielded using moments and Euclidean distance.

In this paper, an effective numerical method for the solution of Helmholtz equation with radiation boundary conditions is considered. This approach is based on the combination of the Krylov subspace type of iterative technique and FFT based preconditioner. The main novel element presented in this paper is the use of the modified FFT type preconditioning that allows us to keep the discretized Sommerfeld-like boundary conditions in preconditioning matrices and still have the numerical efficiency similar to the FFT method. The results of numerical experiments are compared to the standard application of GMRES method and FFT type preconditioner obtained by the replacing radiation boundary conditions with Neumann boundary conditions on the preconditioning step. The convergence of proposed algorithm was investigated on two test problems. Numerical results for realistic ranges of parameters in soil and mine-like targets are presented.

The Region Processing Algorithm (RPA) has been developed by the Office of the Army Humanitarian Demining Research and Development (HD R&D) Program as part of improvements for the AN/PSS-14. The effort was a collaboration between the HD R&D Program, L-3 Communication CyTerra Corporation, University of Florida, Duke University and University of Missouri. RPA has been integrated into and implemented in a real-time AN/PSS-14. The subject unit was used to collect data and tested for its performance at three Army test sites within the United States of America. This paper describes the status of the technology and its recent test results.

In this study buried object detection on the GPR data is examined using CA-CFAR detector. In the first part of the study the background signals of B-scan frames from a pulse GPR is statistically inspected. The results revealed that the background signals residual from a removing process of the dominant GPR signals due to air-to-ground interface have shown K-Distributed statistics. The form and scale parameters of K-Distribution are estimated using the fractional moments. The background or the clutter signals from three different soils have resulted in distinctive shape parameters. The shape parameter of the distribution could generally discriminate three soils. In the second part of the study the receiver loss of CA-CFAR detector is estimated using a numerical method and the Monte-Carlo simulation. The receiver loss is also associated to the K-Distribution and CA-CFAR detector parameters in the simulation. Time series with statistical properties similar to those of the real measurements are obtained using SIRV and employed in the Monte-Carlo simulation. In the third part of the study effectiveness of CA-CFAR detector on B-scan frames is analyzed by measuring the ROC of the detector. High detection probabilities of buried objects at relatively low SNR data are obtained by CA-CFAR detector.