An automated band selection algorithm suitable for real-time application with fixed filter multispectral cameras is presented for multispectral target detection. Fixed filter multispectral cameras collect all bands regardless of the background. Background adaptive band is the selection of a subset of the bands for target detection processing. Fixed filter systems typically include a small number of general-purpose bands. The bands are chosen to enhance target-background contrast but are not keyed to specific target features. In some situations it is unlikely that all bands contribute to target discrimination. Using only a subset of the available bands can decrease false alarms while maintaining target detection performance and reduced processing requirements. The advantages are demonstrated using six band multispectral data and two distinct background categories.

Mine detection systems have traditionally used close-range sensors designed to detect mines within a few feet of the sensor. It would be advantageous to be able to detect mines from a greater distance, especially if the sensor is on a vehicle-mounted platform. Forward-looking cameras are a possible way to achieve this and to provide a 24 hour capability thermal imagery would seem most suited to this application. As many mine targets have flat surfaces, radiation reflected by the target is likely to have some degree of polarization which can be differentiated from the surrounding area, even when the target is partially obscured. This paper, based on work carried out by the Defense Science and Technology Laboratory (Dstl), outlines how the polarization of thermal radiation in a scene can be used to detect surface lain mine targets at longer ranges than traditional sensors and discusses how partially obscured targets may be detected using this system.

Estimating the temperature of bulk explosives in landmines is necessary for an optimally designed Nuclear Quadrupole Resonance explosive detection system, a technology that holds promise for mine detection system false alarm reduction. Pursuant to this objective it is necessary to study the heat conduction and temperature profiles of buried mines over diurnal cycles. Finite element models are constructed and a thermal analysis is performed for buried SIM-25 (landmine simulant) and VS 1.6 anti-tank landmine. The Ansys finite element software is used to create, to solve and to post-process the thermal models. Transient thermal analyses with various boundary conditions and simple soil models are performed. We report on the bulk explosive temperature, thermal flux and thermal gradients for these mine models over diurnal cycles.

A numerical model for polarimetric signatures of surface-laid land mines is presented. The model simulates high-resolution images formed from Stokes parameters that describe thermal emission and reflection of sunlight and skylight. The temperature of the mine and its surroundings is computed as a function of time using a finite element solution of the heat transport equation. The effects of surface roughness are included via a two-scale model, in which the gross shape of the mine is represented by triangular facets (the surface facets of the finite element tetrahedra). Theoretical solutions for rough surface scattering are used to describe small-scale roughness on a facet. Multiply reflected contributions are neglected in the current implementation. An example is presented in which the role of each component is described and related to the observed image.

Soil surface temperatures not only exhibit daily and annual cycles but also are very variable in space and time. Without knowledge of the spatial and temporal variability of soil surface temperatures, it will be difficult to determine what times of day are most suitable for mine detection using Thermal Infra Red (TIR) technology. In this study we monitor the spatial and temporal variability of soil surface temperatures under a range of soil texture and soil moisture conditions on undisturbed plots and plots with a buried anti-tank mine in arid New Mexico. We also analyzed soil surface temperature measurements taken at the test facility for land mine detection systems at the TNO Physics and Electronics Laboratory under the temperate climatic conditions of The Netherlands. The measurements in both areas show a cyclic behavior of the thermal signatures of the mines during the day and night that can be predicted by physics of the mine-soil-sensor system. However, unexpected behavior of the thermal signatures in a silt loam demonstrated that prediction of thermal signatures of buried mines is not straightforward.

A specular model has been used to predict the passive polarimetric infrared (IR) signature of surface-laid landmines. The signature depends on the temperature of the landmine and the sky radiance. The temperature of the landmine is measured using a thermocouple. The signature itself is measured using a polarimetric IR camera setup. The predictions are fit to the measurements using the refractive index as an optimization parameter. The obtained refractive indices of each landmine type are consistent, but for the PMN landmine much lower than determined in a previous indoor experiment. Throughout the measurement day, the average landmine polarimetric signature was higher than the average background signature. Moreover the polarimetric signature appears to be a more robust indicator of the shape of the landmine's top surface than the normal IR signature. A simulator of passive polarimetric imagery is also being developed. That work is based on a physical model for both the thermal and radiometric processes, and it includes a finite-element solution for the heat transfer problem, ray tracing to describe the incident sunlight and the effects of shadowing, and analytical models for the Mueller matrices of rough dielectric surfaces. Preliminary results from that model show substantial qualitative agreement with measured images.

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 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.

We discuss the problem of detecting minelike shapes in data coming from mine detection sensors that can provide images, such as an imaging metal detector, an infrared camera or a ground-penetrating radar. Firstly, we show a way for selecting possibly dangerous regions that should be further analyzed, i.e. to which shape analysis methods should be applied. Then, two shape detection methods are presented, both based on the randomized Hough transform. Most of the mines are of a cylindrical shape, so, due to some burial angle, they appear elliptical in 2D images that are taken parallel to the ground. Thus, one of the two presented methods deals with the detection of elliptical shapes. The other method is developed for detecting the hyperbolic signatures of mines in B-scans (vertical data slices into the ground) of ground-penetrating radar data. Finally, pieces of information that can be extracted from detected ellipses and hyperbolas are discussed, and two ways are suggested for their further use towards determining whether a particular selected region contains a mine indeed or not. Both methods are illustrated on real data.

The current minefield detection approach is based on a sequential processing employing mine detection followed by minefield detection. In case of patterned minefield, minefield detection algorithms seek to exploit the minefield pattern (such as linearity) while in case of scattered minefield they utilize the spatial distribution of the mine targets. However, significant challenges remain in adequate modeling and detection of the minefield process especially in the presence of false alarms due to cultured as well as natural clutter. A short review of the literature on spatial point processes is included especially for the case of scattered minefields. It is further noted that, minefields are characterized by as a pattern (or spatial distribution) of similar looking mine-like objects. The sequential mine-detection followed by mine-field detection paradigm fails to exploit this critical aspect of similarity of targets for minefield detection. In this paper we propose a minefield detection scheme that incorporates similarity based clustering of targets in order to improve the performance of minefield detection. This approach can be interpreted as statistics of a marked point process. Some preliminary comparative ROC curves are evaluated for simulated minefield data in order to show the effectiveness of the minefield detection based on the marked point process. An autonomous self-organizing scheme for on-line clustering of mine-targets is also presented.

High detection performance is required for an operational system for the detection of landmines. Humanitarian de-mining scenarios, combined with inherent difficulties of detecting landmines on an operational (vibration, motion, atmosphere) as well as a scenario level (clutter, soil type, terrain), result in high levels of false alarms for most sensors. To distinguish a landmine from background clutter one or more discriminating object features have to be found. The research described here focuses on finding and evaluating one or more features to distinguish disk-shaped landmines from background clutter in infrared images. These images were taken under controlled conditions, with homogenous soil types. Two methods are considered to acquire shape-based features in the infrared imagery. The first method uses a variation of the Hough transformation to find circular shaped objects. The second method uses the tophat filter with a disk-shaped structuring element. Furthermore, Mahalanobis and Fisher based classifiers are used to combine these features.

Lateral migration radiography (LMR), a new form of Compton backscatter x-ray imaging, is applied to the detection and identification of buried land mines. A mobile LMR land mine detection system was developed and field tested. Weight for this initial system was about 175 kg; weight for a prototype should be about 100 kg. X-ray generator power level was 750 watts; the power level requirement for a prototype should be about 300 watts. An innovative rotating collimator for the x-ray source beam was developed to provide rapid side-to-side scanning of the beam without having to move the x-ray generator in this direction. Acquisition of images of a 40 cm by 40 cm area takes from 30 to 60 seconds, depending on the desired resolution. The imaging capabilities of LMR make it well suited for use as a land mine detection confirmation sensor. This system was employed on the vehicular test lanes at Fort A.P. Hill in October, 2001. High quality images were obtained for a variety of buried land mines. The system was also used to scan 30 locations on one of the test lanes where GPR consistently yielded false alarms. In only two cases did the LMR image sets yield a signature that could be considered to possibly indicate a mine.

Two novel solution methods for the inverse problem for the 2D Helmholtz equation are developed, tested and compared. The proposed approaches are based on a marching finite-difference scheme which requires the solution of an overdetermined system at each step. The preconditioned conjugate gradient method is used for rapid solution of these systems and an efficient preconditioner has been developed for this class of problems. The underlying target application is the imaging of land mines, which is formulated as an inverse problem for a 2D Helmholtz equation. The images represent the electromagnetic properties of the respective underground regions. Numerical results are presented.

Current research in minefield detection indicates that operationally no single sensor technology will likely be capable of detecting mines/minefields in a real-time manner and at a performance level suitable for a forward maneuver unit. Minefield detection involves a particularly wide range of operating scenarios and environmental conditions, which requires deployment of complementary sensor suites. Consequently, the NVESD sponsored Signal Processing and Algorithm Development for Robust Mine Detection (SPAD) Program is currently focusing on the development of computationally efficient and robust detection algorithms applicable to a variety of sensors and on the development of a robust decision level fusion algorithm that exploits these detectors. One SPAD detection technique, called the Ellipse Detector, has been previously reported in the open literature. We briefly report on the continued robust performance of this detector on some new sensor output. We also report on another robust detector developed for sensors that produce output not suitable for the Ellipse Detector. However, the focus of this paper is on the SPAD decision level fusion algorithm, called the Piecewise Level Fusion Algorithm (PLFA). We emphasize the robustness and flexibility of the PLFA architecture by describing its performance and results for both multisensor and multilook fusion.

Multi-band medium wave infrared (MWIR) image data collected from the Lightweight Airborne multispectral Minefield Detection-Interim (LAMD-I) program is examined for the detection of surface landmines. Because the orientation of the image acquisition from aircraft with respect to the mine and the minefield is unknown, there is a need to develop an orientation invariant-based approach for landmine and minefield detection. A rotation invariant circular harmonics transform (CHT)-based approach is presented for surface landmine detection. The magnitude information from the CHT is used for finding mine-like regions within the MWIR images. A three-tiered hierarchical thresholding technique provides the basis for highlighting potential surface landmines. Mine shape and size information are used for generating landmine confidence values. Surface landmine detection capability is presented for 82 MWIR broadband images with sand and short and long grass terrain conditions for daytime and nighttime acquired MWIR image data. Receiver operator characteristic (ROC) curves are used for comparing experimental results from this technique with an existing an adaptive multi-band CFAR detector (RX approach).

This paper presents preliminary analysis of the data from measurements on a minefield in Croatia done in the international cooperation project Airborne Minefield Area Reduction (ARC). Temperature differences above and around suspected mines and minefield indicators, were recorded with a long wave IR camera in 8-9 micrometers , over a time of several days, capturing data under different weather conditions. The data are compared to simulations of land mines, minefield indicators and other objects using a themodynamic FEM model, developed at FOI. Different detection methods are presented and applied to the data.

The Joint Mine Detection Technology (JMDT) project, following successful field-based testing of its new Tunable Filter Multispectral Camera (TFMC) has now completed initial Airborne Testing of the TFMC at both the Coastal Systems Station and Eglin Air Force Base sites. An overview of the testing is presented along with the investigations into the advantages of a system utilizing the TFMC in airborne operational scenarios. The TFMC-like tuning flexibility was flight-tested using optimized wavelength combinations, which were found using field test data, over a variety of backgrounds and altitudes. The data revealed the suitability of background tuning, polarization, and mechanically co-registered channels as benefits to multispectral target detection. The data were also compared to that collected with an IMC-201 camera, using the six filters of the Coastal Battlefield Reconnaissance and Analysis (COBRA) Advanced Technology Demonstration (ATD) system, in order to determine improvements over existing capabilities.

This paper describes the results of a pair of experiments to compare the performance of a collection of infrared sensors for imaging obstacles and mines in the surf zone regions. Results include a statistical and subjective analysis of the imagery. Emphasis is placed on sensor performance during thermal crossover, day/night transitions, and nighttime operations. Also the sensitivity required for imaging obstacles in water is measured.

QinetiQ is developing a hand held Multi-sensor mine detector prototype for humanitarian applications. The sensor consists of a GPR, a metal detector and ancillary sensors. This paper describes how data produced by ancillary sensors can be exploited in order to assist the GPR processing. The GPR consists of a 3x3 array of antennas, and focused images of the volume beneath the sensor are formed by post reception synthetic aperture processing. The mine detector is intended to detect sub surface targets, and an accurate knowledge of the ground surface position relative to the sensor is required. Also the high frequency dielectric constant of the ground medium is required in order to produce focused images. This paper analyses the requirements for good post reception synthetic aperture processing. The accuracy of the ground surface position data and the dielectric constant estimation are determined. A model for soil dielectric constant is used to derive the sensitivity of post reception synthetic aperture processing to unknown soil texture. It is show that for the GPR configuration considered, a wide range of texture variations is tolerable provided the soil moisture can be accurately estimated. Variations in soil composition are also tolerable.

In previous papers, we reported on the high-resolution ground-penetrating radar system designed, built, and deployed by SRI under contract to the Night Vision and Electronic Sensors Directorate at Fort Belvoir. Here, we report on some of the latest test results from the field demonstrations performed at government test sites, carefully designed to produce statistically significant results by employing many samples of a few representative metal and plastic mine types, buried at several depths. Significant improvements in performance above the baseline have been realized by using a number of statistically optimal image processing algorithms based on sound mathematical principles and techniques, guided by electromagnetic models of both mines and clutter. Principal component analysis was employed to define empirical models of buried mines using the polarimetric, complex data. A detector based on the generalized likelihood ratio test was then used on each image. Finally, multilook processing combined the results from several independent views of the same mines (at various ranges). For buried metal mines, at a probability of detection of 94%, the probability of false alarm per m² decreased by an average factor of about four orders of magnitude over the baseline.

Spiral antennas are one of the common radiators used in ground penetrating radar (GPR). Mine detection is generally performed in a frequency band of interest between 500 MHz to 4 GHz. This paper discusses technical recommendations and R&D performed by Naval Air Warfare Center (NAWC), China Lake, CA , resulting in our best effort spiral design emphasizing highest low band gain while maintaining overall axial ratio purity. This design consisted of a spiral printed on a high dielectric substrate that allowed the antenna to be used at lower frequencies then conventional plastic substrate based two arm spirals of the same diameter. A graded dielectric overlay scheme was employed to facilitate matching to free space on one side, and absorber lined cavity on the other. Test data is given in terms of match and free space patterns using spin linear sources to obtain antenna axial ratios. The low-end gain was improved from -17 dBi to -5 dBi. Two Vivaldi slot antennas (star junction fed and an antipodal construction) are discussed as alternative antennas offering broadband high gain and economical construction. Both designs produced good patterns with a +5 dBi average gain over the band. Patterns for the log spiral and Archimedean spiral, together with recommendations for future improvements are provided.

For several years, ground-penetrating radar (GPR) has been used to search for buried landmines. Most of the evaluation effort on complete detection systems has focused on end-to-end performance metrics (e.g., Pd and Pfa). Here, we focus on the specific performance of one critical component of GPR systems-the antennas. This is the first in a series of papers that will compare the following parameters of many different antennas: (1) the most useful bandwidths, (2) the role of polarization and polarization diversity, (3) spurious coupling effects, and (4) phase-correction considerations. This paper compares four types of Planning Systems, Inc., antennas that were developed for current and past GPR systems. These are a 5.5-in. log-spiral antenna without balun or spiral-arm terminations; 5.75-in. log-spiral antenna with tapered balun and arm termination; 5.5-in. Archimedean-spiral antenna with tapered balun, but without arm terminations; and 5.75-in. Archimedean-spiral antenna with tapered balun and arm terminations. Three main tests were performed to compare the antennas: (1) S11, to show overall matching bandwidth and to reveal discontinuities in the balun-antenna-termination structure; (2) S21, to measure undesired direct-path coupling relative to intended target scattering; and (3) S21, to show direct coupling vs. antenna spacing.

PULSAR is an Ultra Wide Band (UWB) short pulse Radar developed by the CELAR (French Technical Center for Armament Electronics) and the IRCOM (Research Institute of Microwave and Optical Communications) in order to detect foliage and ground concealed mines. An instrumentation measurement system has been designed and implemented, in particular new 2D broad band antennas with a very low pulse distortion. The clutter suppression is based on background subtraction and wavelet transforms. These data are used to obtain SAR ultra wide band images by transient methods. The following discussion describes the device, the experimental results and the signal processing currently utilized. Future development efforts on this system (generator, acquisition means .) are detailed. At the same time a theoretical study is made to estimate target transient responses captured by the system. So a FDTD code is modified to simulate buried objects detection by the radar.

In order to detect buried land mines in clutter, Planning Systems Incorporated has developed a Ground Penetrating Synthetic Aperture Radar (GPSAR) system for the U.S. Army CECOM Night Vision and Electronic Sensors Directorate. The GPSAR system is a wide-band stepped-frequency radar operating over frequencies from 500 MHz to 4 GHz. Our GPSAR uses multiple transmit and receive antennas to acquire data at 58 across-track locations separated by 1.47 inches. Along-track data sampling is provided by the forward motion of the system. Multiple radar channels and high-speed radio frequency switching are used to accelerate the data acquisition process and increase the system's maximum speed of advance. Synthetic aperture, near-field beamforming techniques are used to reduce clutter and enhance the signature of buried objects. While the system is designed for mine detection it is capable of locating deeper objects such as buried utility pipes. Tests conducted in December 2001 at U.S. Army facilities indicate that the system can detect both metallic and plastic landmines at depths up to 6 inches. A description of the PSI GPSAR system and test results are presented.

Ground surface roughness is problematic when using a radar impulse to detect and locate land mines. Waves scatter from a random rough ground surface in unpredictable ways, contributing to clutter that is particularly hard to suppress. This clutter has proven experimentally and computationally to distort and obscure the desired scattered field from a buried target. To overcome this effect we have developed a lightweight, artificial dielectric that can be placed over a chosen area that will mimic flat ground and mitigate clutter effects. An artificial dielectric of close-packed array of small insulated metal-coated plastic spheres and lossless uniform plastic spheres can be formulated to match the dielectric properties soil. The ratio of these two spheres in the collection is adjusted to match a particular soil type and the moisture content. Placing them in a conformable bag and ensuring a flat upper interface with the air, ground reflections from an impulse radar can effectively be removed to reveal a target scattering signature. Furthermore, a matched filter can be used to distinguish between a landmine and a false alarm (such as a rock) The artificial dielectric was matched by running experiments in the frequency and time domains. A 1 GHz center frequency impulse ground penetrating radar was used to collect time signals and compare different cases: flat ground, rough ground and rough ground with artificial dielectric. Results indicate excellent rough surface reflection removal and target signal enhancement.

Soil water content, relative permittivity, electrical conductivity, thermal conductivity and heat capacity, directly or indirectly affect the detection capabilities of sensors used for land mine detection. The most important of these is water content since it also influences the other properties. Therefore an experiment was set up where water was applied to a test area and the water content was monitored over time. Simultaneously, measurements with a ground penetrating radar (GPR) were carried out. Subsequently the measurements of both the water content reflectometers (WCR) and GPR were compared against the outcome of a soil water content model and a model relating soil water content with medium relative permittivity. We find that the introduction of water in a dry sand soil, increases the impedance contrast of the land mine with respect to its surrounding (i.e. stronger electromagnetic signatures) which may result in better detection. Alternative effects also seem play a role in finding and identifying features of potential targets.

The contrast in relative dielectric constant between landmines and the surrounding soil is one of the most important elements for radar detection purposes. For most geologic materials the relative dielectric constant lies within the range of 3-30, with dry sand at the lower end of this range at about 3-5. Nonmetallic landmines have a dielectric constant range of 3.2-9.8 whereas metallic landmines have a much higher relative dielectric constant. In previous work, literature data were used to compose a MATLAB model that determines whether or not field conditions are appropriate for use of GPR instruments. This model has been verified for dry and moist sand, silt, and clay soils in New Mexico. The objective of this paper is to validate this model over a wider range of soil texture and soil moisture conditions. Therefore, GPR measurements will be taken on experimental test facilities for landmine detection at Yuma Proving Grounds in Arizona and at the TNO Physics and Electronics Laboratory in The Netherlands. These facilities cover a wide range of soil textures from ferruginous sand to clay and peat as well as many levels of soil moisture.

The utility of ultra-wideband (UWB) synthetic aperture radar (SAR) for detecting surface and flush buried unexploded ordnance (UXO) is examined using a layered-medium moment method analysis. Clutter models of a subsurface root system have been created using a set of discrete clutter objects. Given the size of the targets and the wideband frequencies of interest, it is shown that the problem size quickly grows beyond the capabilities of even supercomputers. As a result, an approximate linear superposition technique is developed to model the response from a large number of targets (UXO plus clutter objects). The root system clutter model is used in conjunction with the buried UXO targets. Results show that sufficient signal to clutter ratios are achieved to make such a scenario amenable to target detection. The scattering from multiple, randomly oriented, surface-laid UXO is examined next. Results show that targets oriented broadside to the radar aperture have the largest signatures in the SAR image. This suggests a multi-pass strategy over the potential UXO test area for airborne SAR systems.

Previous modeling studies have indicated that a multi-frequency radiometer could prove advantageous for humanitarian demining due to the oscillatory patterns in brightness temperature versus frequency that would be observed in the presence of a sub-surface target. Initial experimental results are reported in this paper from a multi-frequency radiometer (MFRAD) system operating at 19 frequencies in the 2.1-6.5 GHz band. The basic design of MFRAD is reviewed, and the calibration and noise background removal procedures discussed. Experimental results with sub-surface metallic and styrofoam targets are then provided that demonstrate the predicted oscillatory behavior. An FFT-based detection algorithm is also described and applied to measured data. Further plans for experiments and tests with this system are also detailed.

Here we consider the use of model-based methods for the detection of buried objects from a sequence of synthetic aperture images obtained by a radar sensor moving linearly down a track. The scattering physics of the underlying sensing modality cause the relevant target signatures to change in a complex yet predictable manner as new images are obtained. To arrive at a tractable processing scheme which exploits these motion-induced changes, we develop a flexible parametric model capable of capturing the full variation of these signatures. A detection method based on a principal components analysis of estimated model vectors is then derived. Results are demonstrated using field data from a forward-looking sensor designed for landmine remediation.

In realistic landmine fields, the anti-personnel plastic mine is often buried nearby a clutter-object under the ground. The presence of a second object buried near the mine under a two-dimensional (2-D) rough ground can easily obscure the target and/or cause a false alarm. The separation distance between the AP mine and clutter-object plays a significant role on the probability of true or false alarm in this situation. A rigorous electromagnetic model has been developed to analyze the scattering mechanism between the target and the clutter-object, between the target and the rough ground, between clutter-object and the rough ground and the multiple scattering between different spots on the rough ground itself. The new rigorous model is based on the classical electromagnetic equivalence theorem leading to producing six new integral equations. Using the Method of Moment (MoM), the new integral equations are transformed into a linear system of equations to be solved for the unknown electric and magnetic currents on the surface of three scatterers; 1) rough ground, 2)target and 3)clutter-object. The MoM impedance matrix completely represents every interaction between these three scatterers. The superior Steepest Descent Fast Multipole Method (SDFMM) is used to tremendously speed up the computations of the unknown MoM surface currents.

The use of ground penetrating radar (GPR) is one of the most popular techniques for the detection of anti-personnel mines and therefore it is desirable to accurately model such systems. For many GPR applications, FDTD models used to simulate the system are two-dimensional, because they are simple to implement and computationally inexpensive. However, a three-dimensional model is more accurate and allows complete freedom for the location of the object relative to the receivers. Instead of fully modeling the transmitter and receiver elements, and adding significant complexity, the transmitted field in this study is experimentally measured and used as the model's excitation. The model developed simulates a GPR system consisting of a parabolic reflector transmitter and a multi-static receiver array. The model is tested for both flat and rough ground with a Gaussian variation. The results are compared with experimental data and are found to be very accurate. The validation of this approach makes the model a powerful tool that can be used in different applications, where the exciting field is computationally or experimentally specified.

This paper investigates the feasibility of detecting plastic antipersonnel land mines buried in lossy, dispersive, rough soils using a stepped-frequency ultra wideband (WB) ground-penetrating radar (GPR). Realistic land mine scenarios were modeled using a two-dimensional (2D) finite difference frequency domain (FDFD) technique. Assuming normal incidence plane wave excitation, the scattered fields were generated over a large frequency bandwidth (.5 to 5 GHz) for a variety of mine-like shapes, different soil types, and multiple receiver locations. The simulation results showed that for a ground penetration sensor located just above the soil surface, the strong reflection signals received from the rough ground surface obscured the buried target's frequency response signal. The simulated GPR WB frequency response data at each receiver location was transformed to the time domain using the fast Fourier transform. Time domain processing permits high resolution measurement of target features that are invariant to the ground roughness and also that are dependent on the soil characteristics as well as the burial depth and size of the mine. Specifically, two or more characteristic timing peaks are observed in the simulation results suggesting that the ultra-wideband spectral radar response may yield particular advantages not exploited by currently employed detection systems. It is also shown that by using time-gating to remove the strong ground reflection signals, the target signals are selectively enhanced (as expected), but more surprisingly, the target frequency response signature is almost completely recovered.

In this paper we describe a possible solution to the rigid Gaussian mixture problem of a continuous hidden Markov model (CHMM) in the context of landmine detection. The main idea of the solution is replacing the Gaussian representation of the feature distribution by a function that uses knowledge about real data distribution (sigmoidal in our case). The main advantage of this approach is that it is faster than the CHMM while maintaining the same performance, fact that can be critical in real-time systems. We use the CHMM as a benchmark for the performance of the newly developed algorithm.

Landmine detection using hand-held GPR unit has two modes of operation: search mode and locate mode. Search mode generates an initial detection at a suspected location and locate mode affirms that the suspected location contains a land mine. Search mode requires causal processing and has a high false alarm rate. Locate mode allows the operator to interrogate the suspected location to eliminate false alarms and does not require causal processing. This paper proposes an algorithm for locate mode processing. The proposed method uses the consistency of detection as well as the shape of detection peaks over several sweeps to improve the detection accuracy. The proposed algorithm was examined through a data set collected at a test site, and was compared with the performance when taking the maximum of detection values over several sweeps. Experimental results showed that the proposed method reduces the probability of false alarm by more than 70% at 100% correct detection rate.

Unlike Vehicle-mounted ground penetrating radar (GPR), the hand-held GPR data is highly variable. In this paper we propose an independent component analysis (ICA) based approach for processing hand held stepped frequency GPR data for mine detection. ICA is a linear transformation, which seeks prominent features in high-dimensional data. Compared to principal component analysis (PCA), which searches for basis vectors in the direction of maximum variance, ICA finds more interesting features in the direction of maximum non-gaussianity. In our current implementation, ICA is used to find a set of basis vectors corresponding to the background clutter. Residual error for this GPR with respect to ICA clutter basis shows the presence or absence of landmine. The performance of the ICA based detection is compared with the correlation detector for GPR only data for hand held mine detection. Comparative receiver operating characteristics (ROC) curves representing probability of detection verses false alarm rate are shown for both scan and investigative mode for ICA based detection and correlation detection.

Statistical signal processing approaches based on Independent Component Analysis (ICA) algorithms for clutter reduction in Stepped-Frequency Ground Penetrating Radar (SF-GPR) data are presented. The purpose of the clutter reduction is indirectly to decompose the GPR data into clutter reduced GPR data and clutter. The experiments indicate that ICA algorithms can decompose GPR data into suitable subspace components, which makes it possible to select a subset of components containing primarily target information (like anti-personal landmines) and others which contain mainly clutter information. The paper compares spatial and temporal ICA approaches on field-test data from shallow buried iron and plastic anti-personal landmines. The data are acquired using a monostatic bow-tie antenna operating in the frequency range from 500 MHz to 2.5 GHz.

The present paper proposes image analysis methods for the detection and classification of landmine targets in images acquired using a ground penetrating radar sensor. The detection methodology initially employs a preprocessing step based on principal component analysis principles. The preprocessed image is further subjected to a multilevel density slicing operation to generate a map of iso-intensity contours in the image. Salient regions, that correspond to true targets as well as false-alarms in the image, are then segmented by establishing hierarchical intensity links within the framework of iso-intensity contours based on parent-to-child nodal relations. Features are proposed to classify mines and FAs based on size, shape, contrast, and texture of the segmented regions.

The high-resolution sonar is one of the principal sensors used by the Navy to detect and classify sea mines in minehunting operations. For such sonar systems, substantial effort has been devoted to the development of automated detection and classification (D/C) algorithms. These have been spurred by several factors including (1) aids for operators to reduce work overload, (2) more optimal use of all available data, and (3) the introduction of unmanned minehunting systems. The environments where sea mines are typically laid (harbor areas, shipping lanes, and the littorals) give rise to many false alarms caused by natural, biologic, and man-made clutter. The objective of the automated D/C algorithms is to eliminate most of these false alarms while still maintaining a very high probability of mine detection and classification (PdPc). In recent years, the benefits of fusing the outputs of multiple D/C algorithms have been studied. We refer to this as Algorithm Fusion. The results have been remarkable, including reliable robustness to new environments. The Quadratic Penalty Function Support Vector Machine (QPFSVM) algorithm to aid in the automated detection and classification of sea mines is introduced in this paper. The QPFSVM algorithm is easy to train, simple to implement, and robust to feature space dimension. Outputs of successive SVM algorithms are cascaded in stages (fused) to improve the Probability of Classification (Pc) and reduce the number of false alarms. Even though our experience has been gained in the area of sea mine detection and classification, the principles described herein are general and can be applied to fusion of any D/C problem (e.g., automated medical diagnosis or automatic target recognition for ballistic missile defense).

The fusion of multiple Computer Aided Detection/Computer Aided Classification (CAD/CAC) algorithms has been shown to be effective in reducing the false alarm rate associated with the automated classification of bottom mine-like objects when applied to side-scan sonar images taken in Very Shallow Water (VSW) environments. This paper reports on the application of such CAD/CAC Fusion algorithms to the shallow water environment, using sidescan sonar data taken in the Gulf of Mexico during April 2000. The fusion algorithm accepts the classification confidence levels and associated contact locations from two different CAD/CAC algorithms, clusters the contacts based on the distance between their locations, and then declares a valid target when a clustered contact passes a prescribed fusion criterion. Two different fusion criteria are evaluated: the first based on the Fisher Discriminant, and the second based on a constrained optimization approach, which minimizes the total number of false alarms over the clustering distance and cluster confidence factor thresholds for a given probability of correct classification. The Fisher-based fusion provided an 82% probability of correct classification at a false alarm rate of 0.034 false alarms per image per side (port or starboard). This performance represented a 2:1 reduction in false alarms over a single CAD/CAC algorithm at this same probability of correct classification. The cluster confidence fusion algorithm performed nearly as well, yielding the 82% correct classification probability at a false alarm rate of 0.039 false alarms per image per side.

A novel sea mine computer-aided-detection / computer-aided-classification (CAD/CAC) processing string has been developed. The overall CAD/CAC processing string consists of pre-processing, adaptive clutter filtering (ACF), normalization, detection, feature extraction, feature orthogonalization, optimal subset feature selection, classification and fusion processing blocks. The range-dimension ACF is matched both to average highlight and shadow information, while also adaptively suppressing background clutter. For each detected object, features are extracted and processed through an orthogonalization transformation, enabling an efficient application of the optimal log-likelihood-ratio-test (LLRT) classification rule, in the orthogonal feature space domain. The classified objects of 3 distinct processing strings are fused using the classification confidence values as features and logic-based, M-out-of-N, or LLRT-based fusion rules. The utility of the overall processing strings and their fusion was demonstrated with very shallow water high-resolution sonar imagery data. The processing string detection and classification parameters were tuned and the string classification performance was optimized, by appropriately selecting a subset of the original feature set. A discussion is presented illustrating the statistical independence of the CAD/CAC processing string outputs and providing insights as to the processing gains to be expected with fusion. It was shown that LLRT-based fusion algorithms outperform the logic based or the M-out-of-N ones. The LLRT-based fusion of the CAD/CAC processing strings resulted up to a four-fold false alarm rate reduction, compared to the best single CAD/CAC processing string results, while maintaining a constant correct mine classification probability.

We describe here the current form of Alphatech's image processing and neural network based algorithms for detection and classification of mines in side-scan sonar imagery, and results obtained from their application. In particular, drawing on the Machine Learning literature, we contrast here results obtained from employing the bagging and boosting methods for classifier fusion, in the attempt to obtain more desirable performance characteristics than that achieved with single classifiers.

Modern sidescan sonars produce echographs that look rather like aerial photographs of the seafloor, with a few important differences. Not least is their obvious sonar speckle---random pixel-to-pixel variations of the image intensity across flat surfaces (smooth sand, mud, clay, or fine gravel for instance) that cannot be attributed to system noise alone. Speckle is in fact sediment dependent, suggesting that it might be used for sediment classification, but the wide variation of speckle typically encountered within each traditional sediment class must first be overcome. Following a different approach, we use speckle to automatically subdivide the image into just two, much broader classes that are relevant in a search for objects: 1) regions of interest (ROI) where attention is warranted because their pixel-to-pixel variations cannot be attributed to speckle alone---i.e., resolvable seafloor features or distinct objects must be present; and 2) empty regions whose pixel-to-pixel variations are speckle-like and therefore of no interest. The distinction is posed as a hypothesis test based on physical and statistical theory. The test is suited for detecting small targets comprised of few pixels whose intensity and uniformity are unlikely deviations from local speckle statistics.

This paper presents a new sequential decision feedback approach for pattern classification in a changing environment. An adaptive classification system is developed that uses the decisions of multiple aspects that may not be separated uniformly. A tap delay mechanism is used to impact the final decision at the current aspect of the object. This system minimizes the error of the classifier while it maps the new feature vector to a familiar feature space for the classifier. The test results on an acoustic backscattered data set collected from six different objects: two mine-like and four non-mine-like at 72 aspect angles with 5 degrees of separation and with varying signal-to-reverberation ratio (SRR) from 4 to 16 dB are presented.

The problem of classification of underwater targets involves discrimination between mine-like and non-mine-like objects as well as the characterization of background clutter. In this work this problem is addressed using a newly collected wideband data set. The developed system consists of a pre- processing scheme, which includes removing multi-path effects and other artifacts from the acquired data. Features are then extracted and fed to a back-propagation neural network (BPNN). Test results will be given for the classification between various types of mine-like and non- mine-like objects and for different bottom conditions and depression/elevation angles of the sonar to test the robustness and generalization of the classification scheme.

Sandia National Laboratories (SNL) has been collaborating with the University of Montana's (UM) engineered honeybee colony research under DARPA's Controlled Biological and Biomimetric Systems (CBBS) program. Prior work has shown that the monitoring of contaminants that are returned to a hive by honeybees (Apis mellifera) provides a rapid, inexpensive method to assess chemical distributions and environmental impacts. Members from a single colony make many tens of thousands of foraging trips per day over areas as large as 2 km². During these foraging trips, the insects are in direct contact with most environmental media (air, water, plants, and soil) and, in the process, encounter contaminants in gaseous, liquid and particulate form. These contaminants are carried back to the hive where analysis can be conveniently conducted. Three decades of work by UM and other investigators has demonstrated that honeybees can effectively and rapidly screen large areas for the presence of a wide array of chemical contaminants and for the effects of exposures to these chemicals. Recently, UM has been exploring how bee-based environmental measurements can be used to quantify risks to humans or ecosystems. The current DARPA program extends this work to the training of honeybees to actively search for contaminants such as the explosive residue being released by buried landmines. UM developed the methods to train bees to detect explosives and chemical agent surrogates. Sandia provided the explosives expertise, test facilities, electronics support, and state-of-the-art analytical instrumentation. We will present an overview of the training procedures, test parameters employed, and a summary of the results of field trials that were performed in Montana and at DARPA field trials in San Antonio, TX. Data showing the detection limits of the insects will be included.

Nuclear quadrupole resonance (NQR) is being researched in order to confirm the presence of explosives as part of landmine sensor suites for the UK MOD hand held and vehicle mounted detection applied research programs. A low power NQR system has been developed as a non-contacting, but short range, detection method for explosives typically found in landmines. The results of stand-off detection of buried anti-personnel and anti-tank quantities of RDX and TNT by this system are presented and the differences in the detection of these explosives by NQR are discussed. Signal processing and radio frequency interference rejection methods to improve the performance of NQR explosive detection have been investigated.

Neutron albedo land mine detection involves irradiating the ground with fast neutrons and subsequently detecting the thermalized neutrons which return. This technique has been studied since the 1950's, but only using non-imaging detectors. Without imaging, natural variations in hydrogen content in the soil, chiefly due to moisture, and surface irregularities, produce enough false alarms to render the method impractical in all but the driest conditions. This paper describes research to design and build a prototype landmine detector based on neutron albedo imaging. Realistic Monte Carlo simulations were performed to assess the signal-to-noise ratio for various soil types and moisture contents, assuming a perfect two dimensional neutron imaging system. The study showed that a neutron albedo imager was feasible for mine detection and that image quality could be good enough to significantly improve detector performance and reduce false alarm rates compared to non-imaging albedo detection, particularly in moist soils and where surface irregularities exist. After reviewing various neutron detector technologies, a design concept was developed. It consisted of a novel thermal neutron imaging system, a unique neutron source to uniformly irradiate the underlying ground and hardware and software for image generation and enhancement. Performance capability, including spatial resolution and detection times, were estimated by modeling. A proof-of-principle imager is now being constructed with an expected completion date of Spring 2002. The detector design is described and preliminary results are discussed.

This paper presents a radioisotope-based concept for determining whether an anomaly in the ground contains an explosive material or not. Nitrogen-based explosives, commonly used in landmines, are rich in carbon, nitrogen and oxygen. Hydrogen is also present in the explosive material, but can be found as well in the mine casing (if plastic or wood). Moreover, explosive materials have a density that is higher than that of most organic materials, but lower than that of metals. All these features are employed in a detection concept that relies on the use of an isotopic source of neutrons. The amount of slowing-down of the source's fast neutrons, as they scatter to the thermal energy, is indicative of the hydrogen content. The enhancement of the intensity of fast neutrons, due to resonance scattering by carbon, nitrogen and oxygen, is indicative of their combined presence. Moreover, the photons that accompany neutron production, are Compton-scattered, providing an indication of the electron-density of the anomaly. These three measurements, contrasted against those obtained from the surrounding soil, are indicative of the presence or absence of a mine, or a mine-like, target.

We report on progress toward development of mid-infrared spectroscopic chemical sensors for explosive related chemicals (ERCs) associated with land mines, particularly 1,3-dinitrobenzene (DNB) and 2,4-dinitrotoluene (DNT). Our goal is to develop a fast, reversible and sensitive device for mobile, real-time detection of ERCs. Fiber waveguide sensors are fabricated by coating thin polymer films onto a mid-IR transmissive chalcogenide fiber. ERCs partition into the polymer film where they absorb IR radiation at characteristic wavelengths in the evanescent wave region near the fiber-polymer interface. A wide variety of polymers possessing potentially desirable physical and chemical properties for ERC detection have been characterized. We have demonstrated reversible detection of DNB and DNT at low concentrations in air using a dynamic gas flow system.

The vapor classification performance of arrays of conducting polymer composite vapor detectors has been evaluated as a function of the number and type of detectors in an array. Quantitative performance comparisons were facilitated by challenging a collection of detector arrays with vapor discrimination tasks that were sufficiently difficult that at least some of the arrays did not exhibit perfect classification ability for all of the tasks of interest. For nearly all of the discrimination tasks investigated in this work, classification performance either increased or did not significantly decrease as the number of chemically different detectors in the array increased. Any given subset of the full array of detectors, selected because it yielded the best classification performance at a given array size for one particular task, was invariably outperformed by a different subset of detectors, and by the entire array, when used in at least one other vapor discrimination task. Arrays of detectors were nevertheless identified that yielded robust discrimination performance between compositionally close mixtures of 1-propanol and 2-propanol, n-hexane and n-heptane, and meta-xylene and para-xylene, attesting to the excellent analyte classification performance that can be obtained through the use of such semi-selective vapor detector arrays.

Mine detection dogs have been found to be an effective method to locate buried landmines. The capabilities of the canine olfaction method are from a complex combination of training and inherent capacity of the dog for odor detection. The purpose of this effort was to explore the detection thresholds of a limited group of dogs that were trained specifically for landmine detection. Soils were contaminated with TNT and 2,4-DNT to develop chemical vapor standards to present to the dogs. Soils contained ultra trace levels of TNT and DNT, which produce extremely low vapor levels. Three groups of dogs were presented the headspace vapors from the contaminated soils in work environments for each dog group. One positive sample was placed among several that contained clean soils and, the location and vapor source (strength, type) was frequently changed. The detection thresholds for the dogs were determined from measured and extrapolated dilution of soil chemical residues and, estimated soil vapor values using phase partitioning relationships. The results showed significant variances in dog sensing thresholds, where some dogs could sense the lowest levels and others had trouble with even the highest source. The remarkable ultra-trace levels detectable by the dogs are consistent with the ultra-trace chemical residues derived from buried landmines; however, poor performance may go unnoticed without periodic challenge tests at levels consistent with performance requirements.

Night Vision Electronic Sensors Directorate (NVESD) has initiated a program for land mine sensor development based upon explosives-related chemical (ERC) detection. As part of the NVESD ERC sensor program, we have sampled soils surrounding buried land mines at the experimental mine lanes, U.S. Army Test Site, with the assistance of the Army Corps of Engineers, Cold Regions Research and Engineering Laboratory. Our goal has been to quantify concentrations of explosive related chemicals found in surface soils above and around nine types of buried land mines and to study their fate over time. This paper will describe the field tests conducted post-DARPA Dog Nose Program, the sampling and analysis protocol and the conclusions drawn from these tests.

An overview of the progress on the UK MOD Applied Research Program for Land Mine Detection. The Defense Science and Technology Laboratory (Dstl) carries out and manages the whole of the UK MOD's Mined Area Detection Applied Research Program both within its own laboratories and in partnership with industrial and academic research organizations. This paper will address two specific areas of Applied Research: hand held mine detection and vehicle mounted mine detection in support of the Mine Detection Neutralization and Route Marking System which started in April 1997. Both are multi-sensor systems, incorporating between them metal detection, ground penetrating radar, nuclear quadrupole resonance, ultra-wideband radar, and polarized thermal imaging.

The Johns Hopkins University Applied Physics Laboratory (JHU/APL) has developed a prototype backpack-portable robot system for mine and unexploded ordnance (UXO) detection and identification. The robot system is compact, lightweight and is estimated to be inexpensive to construct. The robot has been designed with an inexpensive, highly accurate, wide bandwidth time-domain electromagnetic induction (EMI) sensor for the detection and identification of metal components in mines and UXO. The robot can be configured for autonomous or person-in-the-loop control. The robot system can be configured with additional light-weight and low-cost mine and UXO sensors such as ground penetrating radar (GPR) and chemical explosive detectors.

We have been developing path planning techniques to look for paths that balance the utility and risk associated with different routes through a minefield. Such methods will allow a battlegroup commander to evaluate alternative route options while searching for low risk paths. A risk management framework can be used to describe the relative values of different factors such as risk versus time to objective, giving the commander the capability to balance path safety against other mission objectives. We will describe our recent investigations of two related path planning problems in this framework. We have developed a stochastic search technique to identify low risk paths that satisfy a constraint on the transit time. The objective is to generate low risk paths quickly so that the user can interactively explore the time-risk tradeoff. We will compare this with the related problem of finding the fastest bounded-risk path, and the potential use of dynamic graph algorithms to quickly find new paths as the risk bound is varied.

When searching for land mines using vehicular mounted sensors, it is important that the ground locations of the detected mines be accurately determined. This is useful for data association when one has multiple looks at a mine by a single sensor or if one uses multiple sensors. It is of ultimate importance for the primary mission, which is to neutralize the detected mines or at least to mark them for avoidance. Factors that contribute to errors in geo-location include inaccurate knowledge of the vehicle position and/or attitude, and also incomplete knowledge about the terrain being searched. This paper addresses the problem of incomplete terrain knowledge and presents relationships between terrain unevenness and the resulting geo-location errors. The results of this analysis indicate that there may be significant geo-location errors for situations where the terrain is not so smooth, e.g., off road searches. The problem can be alleviated via better knowledge of the terrain. Such knowledge could be acquired via scanning the field of view with a ranging device, recording range as a function of azimuth and elevation. A variety of uneven surfaces have been simulated. Two types of sensors are considered, Linear-Array Radar and Camera Type Sensors. Geo-location is then computed based on: a) no range measurements, b) four range measurements (to the four vertices of the sensor field of view), and c) nine range measurements (to the four vertices and intermediate points at the top and bottom row, as well as three measurements across the center row). The geo-location errors are much worse for the Camera Type Sensor, but they can be significant for the Linear-Array Radar also. If the field of view is planar or almost planar, even coarse range scanning can improve geo-location accuracy. For more complex surfaces fine scanning may be required. The computed geo-location errors, and the conclusions drawn as to the effectiveness of the different models are presented in the paper.

A system is under development that uses seismic surface waves to detect and image buried landmines. The system, which has been previously reported in the literature, requires a sensor that does not contact the soil surface. Thus, the seismic signal can be evaluated directly above a candidate mine location. The system can then utilize small amplitude and non-propagating components of the seismic wave field to form an image. Currently, a radar-based sensor is being used in this system. A less expensive alternative to this is an ultrasonic sensor that works on similar principles to the radar but exploits a much slower acoustic wave speed to achieve comparable performance at an operating frequency 5 to 6 decades below the radar frequency. The prototype ultrasonic sensor interrogates the soil with a 50 kHz acoustic signal. This signal is reflected from the soil surface and phase modulated by the surface motion. The displacement can be extracted from this modulation using either analog or digital electronics. The analog scheme appears to offer both the lowest cost and the best performance in initial testing. The sensor has been tested using damp compacted sand as a soil surrogate and has demonstrated a spatial resolution and signal-to-noise ratio comparable to those that have been achieved with the radar sensor. In addition to being low-cost, the ultrasonic sensor also offers the potential advantage of penetrating different forms of ground cover than those that are permeable to the radar signal. This is because density and stiffness contrasts mediate ultrasonic reflections whereas electromagnetic reflection is governed by dielectric contrast.

An array of radars is developed as a stand off sensor for use in elastic/seismic mine detection systems. The array consists of N radar sensors which operate independently to sense the displacement of the surface of the earth due to elastic waves propagating in the earth. Each of the sensors consists of a lens-focused, conical, corrugated, horn antenna and a homodyne radar. The focused antenna allows the sensor to have greater standoff than with the previous unfocused antenna while maintaining the spatial resolution required for a mine detection system. By using an array of N sensors instead of a single sensor, the scan rate of the array is improved by a factor of N. A theoretical model for the focused antenna is developed and an array of two radars is developed and used to validate the theoretical model. This array is tested in both the experimental and the field models for the elastic mine detection system. Results from both systems are presented.

To optimize a landmine detection system currently being developed at Georgia Tech that uses both electromagnetic and elastic waves, wave propagation in soils has been studied to evaluate propagation characteristics and to identify nonlinear mechanisms. The system under development generates elastic waves in the soil using a surface-contacting transducer designed to preferentially excite Rayleigh waves, thus interrogating the surface layers of the soil. These waves propagate through the region of interest and interact with buried landmines and typical clutter objects (i.e., rocks, sticks, and man-made objects). Surface displacements are measured using a non-contact radar sensor that is scanned over the region of interest. To characterize the wave propagation effects as a function of drive amplitude and as a function of input signal type, a series of experiments was conducted using the radar sensor, accelerometers, and geophones at two test sites, the experimental model at Georgia Tech and a field test site at the Georgia Tech Research Institute's Cobb County Research Facility in suburban Atlanta. The two test sites presented different soils as the experimental model uses damp, compacted sand as a soil surrogate while the field test site has a well-weathered mixture of sand, silt, and clay. Surface displacement measurements were made using the radar sensor while both surface and subsurface measurements were made using triaxial accelerometers and geophones. Linear and nonlinear dispersion, wave speed changes, and nonlinear saturation were observed in the measured data.

This paper discusses the development and performance of a multi-beam laser Doppler vibrometer specifically designed to locate buried landmines with a laser-acoustic technique. The device aims at increasing the speed of landmine detection with this technique by at least one order of magnitude. The present system is capable of simultaneously probing sixteen positions on the ground over a span of one meter, and of measuring the ground velocity at each of these positions with a velocity resolution of about 1 micrometers /s. This architecture could also be scaled to a larger number of beams or into two dimensions. The present system uses a low (100 kHz) carrier frequency, which enables digital signal processing in a simple architecture. This paper also discusses a numerical model to simulate and predict the performance of the multi-beam vibrometer. In particular, the model attempts to address issues associated with speckle dropout, signal/noise, and maximum scanning velocity.

Land mine detection research demonstrates that sending acoustic to seismic waves in the ground produces a unique vibrational response in hollow objects such as land mine casings. Even when they are buried, damped vibrations of mines can be measured at the surface above them. These vibrations can be distinguished from the ground or other natural buried objects. Field tests utilizing acoustic technology performed under static (stand still) conditions have yielded high probabilities of detection coupled with low false alarm rates. Army requirements necessitate a forward moving system; therefore we have been investigating the application of acoustic technology for land mine detection under non-static, forward moving conditions. This paper will present the results of a series of field tests in which two laser doppler vibrometers are passed over buried land mine targets that are excited by an acoustic source. The paper will discuss the experiment protocol, the results and the interpretation of these results. This paper will also discuss our future efforts at acoustic land mine detection.

Methods for processing continuously acquired data using an Acoustic/Seismic system are described. Data were acquired from 80-300 Hz. Two independent chains of processing were pursued. In one chain, pre-processing and normalization were followed by shape feature extraction using eccentricity and minor axis length. In another chain, Independent Component Analysis was used to generate image templates. The results were combined using piecewise linear discriminants. Probabilities of detection of 97.5% and 100% with false alarm rates of 0.01 and 0.03 were achieved on training and validation sets, respectively.

A model has been developed to allow the scanned data obtained using a laser Doppler vibrometer-based acoustic-to- seismic landmine detection system to be analyzed without operator interaction. The ground vibration data from the LDV are pre-processed to form images in a 2-D data format. A parametric model was established to describe the amplified magnitude velocity phenomena induced by buried landmines. This model incorporates amplitude, size, position and background amplitude parameters into an automatic analysis process. An iterative regression approach is described which can be used as a major part of the automatic landmine recognition. The estimated parameters, such as the amplitude relative to the background, the size, and the shape of a target are used to make the decision regarding the presence of a mine. Once a positive decision is made, the estimated position parameters are used to localize the target location.

Signatures of buried landmines are often difficult to separate from those of clutter objects. Often, shape information is not directly obtainable from the sensors used for landmine detection. The Acoustic Sensing Technology (AST), which uses a Laser Doppler Vibrometer (LDV) that measures the spatial pattern of particle velocity amplitude of the ground surface in a variety of frequency bands, offers a unique look at subsurface phenomena. It directly records shape related information. Generally, after preprocessing the frequency band images in a downward looking LDV system, landmines have fairly regular shapes (roughly circular) over a range of frequencies while clutter tends to exhibit irregular shapes different from those of landmines. Therefore, shape description has the potential to be used in discriminating mines from clutter. Normalized Fourier Descriptors (NFD) are shape parameters independent of size, angular orientation, position, and contour starting conditions. In this paper, the stack of 2D frequency images from the LDV system are preprocessed by a linear combination of order statistics (LOS) filter, thresholding, and 2D and 3D connected labeling. Contours are extracted form the connected components and aggregated to produce evenly spaced boundary points. Two types of Normalized Fourier Descriptors are computed from the outlines. Using images obtained from a standard data collection site, these features are analyzed for their ability to discriminate landmines from background and clutter such as wood and stones. From a standard feature selection procedure, it was found that a very small number of features are required to effectively separate landmines from background and clutter using simple pattern recognition algorithms. Details of the experiments are included.

The paper presents results of the field test of the nonlinear seismo-acoustic technique for detection and discrimination of land mines. The tests were conducted in summer 2001 at the U.S. Army's outdoor testing facilities. Plastic antitank mines (M19, VS1.6, VS2.2) and plastic antipersonnel mines (M14, VS50, TS50) were confidently detected at their maximum burial depths in both gravel and dirt lanes. Mine M14 is one of the smallest mines and is very difficult to detect by other techniques. The test proved that the nonlinear seismo-acoustic detection algorithm is very sensitive to AT and AP mines, while completely insensitive to false targets, such as rocks, chunks of metal or wood, thus promising to deliver high probability of detection with low false alarm rate. The results of the tests are in good agreement with the developed physical model of the seismo-acoustic detection.

Acousto-to-seismic coupling has proven to be an extremely accurate technology for locating buried landmines. Most of the research to date has focused on linear acoustic techniques in which sound couples into the ground, interacts with the buried mine, and causes increased vibration of the ground above the mine. However, Donskoy has suggested that nonlinear acoustic techniques may be applicable to acoustic mine detection. This technique has recently been used with success in field tests at the University of Mississippi and US Army mine lanes. In the nonlinear acoustic technique, airborne sound is produced at two primary frequencies which couple in to the ground and a superimposed compressional wave interacts with the mine and the soil. Because the mine is compliant, contact between the soil and the mine is maintained during the compression phase of the wave, but they are separate during the tensile phase. This creates a bimodular oscillator that is inherently non-linear. This effect has been demonstrated on inert landmines at the University of Mississippi and at US Army test lanes. Results of these tests indicate that nonlinear measurements over buried landmines have more sensitivity than linear measurements. Non-compliant objects such as concrete disks do not exhibit nonlinear phenomena but can be located using linear techniques.

Acoustic-seismic coupling mine detection offers an alternative approach to distinguishing mines from clutter. The approach is based on the principle that an area with a buried object shows a different response to acoustic excitation from that of the surrounding soil. Prior research shows that the response in the low frequency range can be captured using simple physically based models under certain conditions. According to the models, areas with buried mines exhibit natural frequencies that can be determined from mine types and buried depths. In this paper, we argue that not only are the natural frequencies useful for the purpose of mine detection, but the locations of the transmission zeros are important as well. Under certain conditions, the locations of the transmission zeros are also less sensitive to changes in physical properties of mines. We take advantage of this characteristic and offer a method to improve signal-to-clutter ratio for the purpose of automatic mine detection.

Ground penetrating radar has been proposed as an alternative sensor to classical electromagnetic induction techniques for the landmine detection problem. The NIITEK-Wichmann antenna provides a high frequency radar signal with very low noise levels following the ground reflection. As a result, the signal from a buried object is not masked by the inherent noise in the system. It has been demonstrated that an operator can learn to interpret the NIITEK-Wichmann radar signal to detect and identify buried targets. The goal of this work is to develop signal processing algorithms to automatically process the radar signals and differentiate between targets and clutter. The algorithms that we are investigating have been tested on data collected at the JUXOCO test grid as well as on data collected in calibration lanes that are used for evaluating the performance of handheld and vehicular landmine detection systems. We have developed algorithms based on principle component analysis, independent component analysis, matched filters, and Bayesian processing of wavelet features. We have also considered several approaches to ground-bounce removal prior to processing. In this paper we discuss the relative performance of each of the techniques as well as the impact of ground bounce removal on processing of the data.

For downward looking GPR landmine detection systems, the return from the ground surface, i.e., the ground bounce, often surpasses the actual mine return and makes it almost impossible to detect the landmines, especially the buried plastic landmines. The ground bounce is difficult to remove due to the roughness of the ground surface and/or the changing soil conditions. In this paper, a robust and efficient ground bounce removal algorithm, referred to as ASaS (Adaptive Shift and Scale), is presented. ASaS takes into account the variations of the ground bounce by adaptively selecting a reference ground bounce. The shifted and scaled version of the reference ground bounce is used as the estimate of the ground bounce in the current scan. Two adaptive reference selection schemes for ASaS are given and compared with each other. Experimental results based on the data collected by the PSI GPSAR system are used to demonstrate the effectiveness of the adaptive schemes.

Ground penetrating radar (GPR) has been proposed as an effective sensing modality for reducing the excessively high false alarm rates often encountered in landmine detection applications. Ground penetrating radar is sensitive to discontinuities in the interrogated medium, rather than the presence of metal, and thus exploits a different phenomenology than electromagnetic induction (EMI) sensors. Thus, unique signals that are dependent on the composition of the targets can be obtained from buried objects. Consequently, the detection of low metal content targets is improved since the radar responds to non-metallic objects, such as wood, plastic, and stone, as well as metallic objects. When the GPR sensor is mounted on a moving platform, the target signatures are hyperbolas in a time-domain data record. Furthermore, the hyperbolas from different targets often exhibit different characteristics. The goal of this work is to develop robust signal processing algorithms which exploit this knowledge to improve target detection and discrimination. Among the algorithms considered are a Bayesian approach and an approach similar to the Hough transform. The algorithms are evaluated using real data collected with fielded GPR sensors, and are compared in terms of their computational requirements as well as their detection and discrimination performance.

In this paper we discuss fundamental considerations in the design of an electromagnetic induction (EMI) sensor. Simple circuit representations of pulsed and continuous wave EMI systems are presented and the analysis of these circuits leads one to certain conclusions regarding optimal (good) sensor design. Findings reported here are gathered from experimental research conducted over the past year and directed toward the development of an EMI system that not only has reasonable sensitivity but also has the ability to capture a target's low frequency response characteristics. The later capability is important when attempting to discriminate between low metallic content landmines and metallic clutter.

Metal detection has been in use for many years as a method for mine detection. The one major downside to metal detection is that most objects with sufficient metal content will be detected thus increasing the false alarm rate and decreasing the efficiency of metal detection as a sensor for mine detection. Based on land mine detection research carried out by the Defense Science and Technology Laboratory (Dstl), this paper focuses on methods to reduce the negative effect of metallic clutter on sensor performance by using a dual height metal detector array on a vehicle mounted platform to reject unwanted clutter while highlighting objects that are more likely to be of interest. The paper also covers the potential of exploiting the dual height configuration for target classification and identification using feature extraction methods and neural networks.

This paper presents wide bandwidth, time decay signatures from recent unexploded ordnance (UXO) field experiment at a US Government UXO test site. While current technologies have shown the ability to detect buried metal objects, they tend to fail in discriminating the UXOs from metal objects that pose no risk. Metal target time decay measurements have been shown to be an excellent method for target classification and identification. The present paper addresses the research community's need for accurate, wide-bandwidth UXO target signatures. Metal target signatures for a number of important UXO targets are presented in the paper for both vertical and horizontal magnetic field excitation. Target time decay signatures from about 30 microseconds to 8 milliseconds are presented. Target signatures are also characterized using a non-linear parameterization scheme in an effort to develop a compact target signature library.

This paper describes a spatial scanning time-domain electromagnetic induction (EMI) sensor and presents results from recent field experiments with buried metal and low-metal content (LMC) anti-personnel (AP) and anti-tank (AT) plastic-cased land mines. The EMI sensor is an modified version of the Electromagnetic Target Discriminator (ETD) sensor developed for the US Army CECOM/NVSED by the Johns Hopkins University Applied Physics Laboratory. The spatial scanning ETD sensor has demonstrated the ability to measure metal target decay times starting approximately 6 ms after the transmitter current is turned off and with metal target decay time constants as short as 1 ms. The sensor antenna sweeps 80 cm over a target area and makes time-decay measurements at 14.5 mm intervals. In addition to metal target signatures, the paper describes coincident void and metal signatures from LMC land mines. The detection of coincident void and metal signatures is shown to be an important classification technique for LMC land mines.

One of the long-standing goals in landmine and UXO detection research has been to identify metal objects based on their electromagnetic induction~(EMI) responses. An often-pursued approach is to model the time-domain response of an object with a sum of damped real exponentials whose amplitude and decay coefficients are related to the object's geometric and electromagnetic properties. Using this model, a measured response can be processed, by a number of techniques, in an attempt to extract the associated amplitude and decay coefficients of the constituent exponentials. These coefficients can be potentially related to the object's physical properties. Some years ago the authors investigated this approach using computer simulated data with added noise. Even for the simple case of a sphere, it was not possible to reliably and uniquely estimate the amplitudes and decay constants, particularly in the absence of some a priori information about the object. The basic problem is that damped real exponentials are highly correlated functions. That is, while it is easy to fit a response with a sum of these exponentials, the accuracy of the estimate of their parameters (amplitude and decay constants) is not guaranteed, making it difficult to relate extracted parameters to object properties. The paper illustrates this point using the EMI response of a sphere and the characteristics of fitting exponential sums to data. For objects more complex than the sphere, there will be additional problems such as the dependence of the response on object orientation and depth. In practice, the problem will be exacerbated by low signal strength (particularly for minimum metal landmines), uncertain or unknown object location and depth and the occurrence of a large number of false targets, some of which will have responses which are statistically identical with that of the target being sought. As well, a false target in the proximity of a real target will alter or totally mask the response of the latter.

The use of Electromagnetic Induction (EMI) for land mine detection has been common for over six decades. Recent work in this field has focused on greater discrimination due to the difficulties in detecting mines amongst metal debris. The false alarm rate of such systems could be improved by the ability to discriminate between metal types. This paper presents initial results of a study into metal discrimination using a multi-frequency metal detector. Factors affecting the performance of such a system, including the choice of frequencies, and the mine size, orientation and metal composition are discussed. This work was carried out as part a UK MoD funded Program.

Wideband electromagnetic induction (EMI) data provides an opportunity to apply statistical signal processing techniques to potentially mitigate false alarm rates in landmine detection. This paper explores the application of matched subspace detectors and support vector machines (SVMs) to this problem. A library of landmine responses is generated from background-corrected calibration data and a bank of matched subspace detectors, each tuned to a specific mine type, is generated. Support vector machines are implemented based on the full mine responses, decay rate estimates, and the outputs of the matched subspace filter banks. Different training approaches are considered for the support vector machines. Receiver operating characteristics (ROCs) for the matched subspace detectors and support vector machines operating in a blind field test are presented. The results indicate that substantial reductions in the false alarm rates can be achieved using these techniques.

Decay rate estimation has been proposed as an effective method for landmine and unexploded ordnance (UXO) detection and discrimination when electromagnetic induction (EMI) sensors are employed. The phenomenological basis for this strategy is that every object in the target library (i.e., landmine and/or UXO target) possesses a unique set of decay rates that are dependent upon the physical characteristics of the target of interest. In theory, these decay rates can be estimated from the measured EMI response and then utilized for target detection and subsequent discrimination and/or classification. Since the basis for this approach to target detection and identification is that targets are uniquely characterized by their decay rates, discrimination performance is dependent upon decay rate estimation performance. Unfortunately, decay rate estimation is notoriously difficult, and this difficulty adversely impacts target discrimination performance. We propose a parameter transformation to improve both the accuracy and the robustness of decay rate estimation when the decay rates are estimated using nonlinear least squares techniques. We present simulation results showing the improvement in the both the RMS error and the bias of the estimates achieved with the parameter transformation.

Electromagnetic (EM: Magnetic Susceptibility [MS], Electrical Conductivity) and soil texture characteristics were determined for a Cambodian soil from an area where landmine detection interference has been experienced. The purpose was to collect information for developing techniques to discriminate between EM signals from small metallic particles in landmines and from iron-oxides or ferromagnetic mineral grains in soil. Ferromagnetic minerals are iron-oxides with strong MS characteristics. Results indicate that this soil consisted of four textural components: clasts (2-10 mm), medium-coarse-sand (<2.0 mm), fine-sand (<0.25 mm) and clay-silt (<0.063 mm). The coarse-sand had high MS values (~550x10^-8 SI/kg) due to high ferromagnetic mineral content (~20 wt.%). Some large rounded clasts, however, had considerably higher MS values (~11000x10^-8 SI/kg) due to high ferromagnetic mineral concentrations (30-60 wt.%), a likely source of significant landmine detection interference. The finer components had smaller MS values and iron-oxide contents. Complex electrical conductivity (1-10⁶ Hz) of iron-oxides showed significant frequency dependence due to capacitance effects of electrochemical double layers on their surfaces in contact with soil moisture. This frequency dependence of iron-oxides may provide opportunities for potential EM system's design to discriminate between soil and landmine responses.

We discuss an improved Kalman filter-based algorithm for automatic detection of targets from metal detector data. This innovations process utilizes the difference between measurements and single-stage predicted values. In our previous work a Kalman filter based algorithm was used to detect targets assuming that the metal detector output signal is a constant in the background. In this work we extend the capability of this method to detect targets by assuming the distribution of the metal detector output data is Gaussian. The analysis has been extended by computing state estimation errors, covariance matrices and treating metal detector background data as a discrete-time Gauss-Markov random sequence. The proposed detection algorithms have been applied to Minelab F1A4-MIM metal detector data.

Several sensor fusion approaches are in use for detection of land mines. These are based on different techniques and show different results. Meaningful comparisons of fusion algorithm performance are difficult, because the performance achievable may be limited by the data acquired and the sensors used. Especially in comparing such dissimilar situations, it is important to compare measures that are not dependent of sensor peculiarities or layout of test lanes. Nonetheless, a quantitative comparison of algorithms is necessary to identify the most effective fusion techniques. This comparison must sometimes be done when the algorithms are being applied to different data sets. In this paper we describe a methodology to evaluate the performance of sensor-fused mine-detection systems. This process can be used to compare different mine-detecting sensors in well- defined scenarios.

Humanitarian demining is a very dangerous, cost and time intensive work, where a lot of effort is usually wasted in inspecting suspected areas that turn out to be mine-free. The main goal of the project SMART (Space and airborne Mined Area Reduction Tools) is to apply a multisensor approach towards corresponding signature data collection, developing adapted data understanding and data processing tools for improving the efficiency and reliability of level 1 minefield surveys by reducing suspected mined areas. As a result, the time for releasing mine-free areas for civilian use should be shortened. In this paper, multisensor signature data collected at four mine suspected areas in different parts of Croatia are presented, their information content is discussed, and first results are described. The multisensor system consists of a multifrequency multipolarization SAR system (DLR Experimental Synthetic Aperture Radar E-SAR), an optical scanner (Daedalus) and a camera (RMK) for color infrared aerial views. E-SAR data were acquired in X-, C-, L- and P- bands, the latter two being fully polarimetric interferometric. This provides pieces of independent information, ranging from high spatial resolution (X-band) to very good penetration abilities (P-band), together with possibilities for polarimetric and interferometric analysis. The Daedalus scanner, with 12 channels between visible and thermal infrared, has a very high spatial resolution. For each of the sensors, the applied processing, geocoding and registration is described. The information content is analyzed in sense of the capability and reliability in describing conditions inside suspected mined areas, as a first step towards identifying their mine-free parts, with special emphasis set on polarimetric and interferometric information.

It is widely accepted that single sensors cannot simultaneously achieve both high detection rates and low false alarm rates for the landmine detection problem. Thus, in this paper we consider the fusion of two types of sensors, electromagnetic induction (EMI) and ground penetrating radar (GPR). In its most common instantiation, EMI essentially provides metal detection and thus detects mines with high metal content as well as metal debris in the environment. More advanced EMI systems have begun to show potential for discriminating such debris from landmines. GPR is also used for landmine detection since it can detect and identify low-metallic subsurface anomalies. In our previous work, we have shown that a Bayesian detection approach can be applied to EMI and GPR data and provide improvements in false alarm rates. In this paper, we present results that indicate that statistical signal processing techniques can be applied simultaneously to GPR and EMI data and that reductions in false alarm rates can be achieved. We present results for two landmine detection systems, both handheld, and when possible compare the results to those obtained by a human operator who essentially fuses the outputs of the single sensor systems.

The potential for performance improvements through sensor fusion is explored for two electro-optical (EO) imaging sensors: a passive thermal IR camera and an active polarimetric system. Tests of decision-level fusion using a small data set (roughly 60 mine signatures) suggest that a significant performance improvement can be obtained by using an AND fusion approach. The source of this improvement derives from correlation among the sensors. Specifically, the sensors exhibit a strong positive correlation when a mine is present, and a negligible correlation when viewing clutter. The observed improvement is independent of the local ground clutter, but it depends strongly on the decision thresholds used for the individual sensors.

Landmine detection using metal detector (MD) and ground penetrating radar (GPR) sensors in hand-held units is a difficult problem. Detection difficulties arise due to: 1) the varying composition and type of metal in landmines, 2) the time-varying nature of background and 3) the variation in height and velocity of the hand-held unit in data measurement. In prior research, spatially distributed MD features were explored for differentiating landmine signatures from background and non-landmine objects. These features were computed based on correlating sequences of MD energy values with six weighted density distribution functions. In this research the effectiveness of these features to detect landmines of varying metal composition and type is investigated. Experimental results are presented from statistical analysis for feature assessment. Preliminary experimental results are also presented for evaluating the impact on MD feature calculations from varying height and sweep rate of the hand-held unit for data acquisition.

An effort is underway to develop a fused sensor system for effectively detecting both metallic and non-metallic landmines. This advanced research effort will meld two orthogonal technologies, acoustic-to-seismic coupling and ground penetrating synthetic aperture radar, into a single system with a higher probability of detection and lower false alarm rate than either technology can achieve individually. Previous testing has demonstrated that these two technologies have individually high probabilities of detection and low false alarm rates but exploit disparate phenomena to locate mines. The fact that they both produce similar data makes a high confidence mine/no mine decision possible. Future plans include a stepped development process to build a close-in detector and leveraging that experience to develop a forward-looking system capable of meeting long- term Army requirements.

This paper presents a simple target-tracking algorithm for use on a vehicle platform for landmine detection. Data from three landmine detection sensors mounted at the front of the vehicle enhance the probability of detection and, when combined via data fusion, reduce the false alarm rate to practical levels. The output of the data fusion process is a target position that must be tracked to either a marking nozzle array or point confirmation sensor mounted at the back of the vehicle.

The MsMs project is a major campaign to collect calibrated and well-documented data, suitable for use by workers developing advanced multisensor algorithms for antipersonnel mine detection. The data, together with a full description of the site layout and measurement protocols, are publicly available via the internet site http://demining.jrc.it/msms. Measurements are made on a test lane consisting of 7 plots of different soils, each 6m by 6m, populated with surrogate mines, calibration objects, simulated clutter and position markers. There are 48 targets in each plot, configured identically for all plots. A first report was presented last year. Since then, laser acoustic vibrometer and magnetometer data have been added and the metal detector and thermal infrared data have been augmented. The database has been reformatted to make it more uniform and user-friendly and to remove typographic mistakes. The test site remains essentially unchanged, apart from some equipment upgrades, and is available for further data collection. In particular, the targets have not been moved, so as to provide stable surrounding soil conditions representative of mines left undisturbed for long periods post-conflict. This presentation will describe the new data and data format, the status of the upgrades and the outlook for the future.

This paper will present an overview and early results of the QinetiQ Portable Humanitarian Mine Detector project, funded by the UK Treasury Capital Modernization Fund. The project aims to develop a prototype multi-sensor man-portable detector for humanitarian demining, drawing on experience from work for UK MoD. The project runs from July 2000 to October 2002. The project team have visited mined areas and worked closely with a number of demining organizations and a manufacturer of metal detectors used in the field. The primary objective is to reduce the number of false alarms resulting from metallic ground clutter. An analysis of such clutter items found during actual demining has shown a large proportion to be very small when compared with anti-personnel mines. The planned system integrates: a lightweight multi-element pseudo-random-code ground penetrating radar array; a pulse induction metal detector and a capacitive sensor. Data from the GPR array and metal detector are fused to provide a simple audio-visual operator interface. The capacitive sensor provides information to aid processing of the radar responses and to provide feedback to the operator of the position of the sensors above the ground. At the time of presentation the project should be in the final stages of build, prior to tests and field trials, which QinetiQ hope to carry out under the International Test and Evaluation Project (ITEP) banner.

An experimental study was undertaken to measure the absolute reflectance of a large number of landmines using a single- mode laser diode operating at approximately 814 nm. Approximately thirty different anti-tank (AT) and anti- personnel (AP) mines were irradiated with a highly polarized laser beam configured in a monostatic geometric arrangement. The in-plane (p) and out-of-plane (s) linear polarization components were measured as a function of incidence angle of the transmitted laser energy. The relative value of the reflected power of each polarization component was calibrated against various reflecting Spectralon panels to obtain an absolute reflectance. This effort is being undertaken in support of an Army Program using an active laser system.

A tactical unmanned aerial vehicle-size illumination system for enhanced mine detection capabilities has been designed, developed, integrated, and tested at the Coastal Systems Station. Airborne test flights were performed from June 12, 2001 to February 1, 2002. The Airborne Laser Diode Array Illuminator uses a single-wavelength compact laser diode array stack to provide illumination and is coupled with a pair of intensified CCD video cameras. The cameras were outfitted with various lenses and polarization filters to determine the benefits of each of the configurations. The first airborne demonstration of a laser diode illumination system is described and its effectiveness to perform nighttime mine detection operations is shown.

In case of hand-held mine detection, the operator functions in two distinct modes. Namely the scan mode and investigation mode. In scan mode, the operator scans the area to look for potential targets. On identifying a suspect target location, the operator switches to investigation mode where he/she closely scan the area and tries to identify/discriminate target based on consistency, size and strength of the response. The aim of this paper is to look at the various aspects of sensor fusion in scan and investigation mode to fuse information from a collocated metal detector and ground penetrating radar sensors on a hand-held mine detection unit. Different sensor fusion schemes are compared. It is found that the two sensors are complimentary for a set of mine targets while they are supplementary for other set of mine targets. A better detection performance can be achieved by suitable modification to the sensor fusion scheme based on identified electromagnetic characteristics of detected targets.

Shape features based on gray-scale moment invariants are presented for airborne mine detection and discrimination. Eleven shape features are obtained by translation, rotation and contrast normalization of the fourth-order gray-scale moments. Mahalanobis distance between an observed and true (average) shape feature vector is used as a shape metric. Covariance matrix corresponding to the average shape feature vector is obtained analytically using an additive and multiplicative noise model for the MWIR image. Effectiveness of gray scale moment invariant shape features for mine discrimination and false alarm mitigation is shown using MWIR imagery collected for LAMD-I program in May 2000. Successful implementation of the features in an airborne detection depends on the consistency of these shape features over time with change in factors such as solar illumination, ageing, clouds and environmental conditions. A study of the variability of gray-scale moment invariant-based shape features with time is conducted using MWIR time-sequenced imagery acquired in June-July 1998 by E-OIR.

A major difficulty in automatic mine detection arises from the fact that the physical properties of background soil can vary significantly from one location to another. This in turns alters the sensor signals of the buried mines. Hence, a robust ATR algorithm for mine detection requires that the algorithm be adaptable to environmental changes. Moreover, mine features used for detection should be invariant to background variation. We have developed an ATR algorithm that uses only background soil data during the training phase and mine features that are less affected by soil changes. Since the algorithm uses only the background data for training, not only is it much easier to tailor the algorithm to a minefield but the algorithm can also be adapted in real-time during operation. This further improves robustness of the process. The algorithm demonstrated good performance when tested on ground penetrating radar data acquired from U.S. Army test lanes.

This paper analyzes the UXO classification capabilities of the GEM-3 using data collected for the Advanced UXO Detection/Discrimination Technology Demonstration at the U.S. Army Jefferson Proving Ground (JPG), Madison, Indiana. The approach taken in the US Army Engineer Research and Development Center (ERDC) analysis of the performance of the GEM-3 at JPG was to extract data points collected near each of the actual target locations and compare them to the calibration data acquired with known targets at the beginning of the demonstration. This was done to determine how well the data collected near each actual target matched the calibration signatures for the same ordnance type and the extent to which the data could be differentiated from other ordnance types and non-ordnance clutter. Classification of the targets was performed using a simple template-matching algorithm. This procedure resulted in an exact classification match for nearly half of the targets for which calibration data were available and a match to a similarly sized target for more than two-thirds of the medium and large targets. The sensor coverage of the test areas and the effect of test parameters such as ordnance size and depth on classification performance were also examined. New data were acquired with the GEM-3 to investigate the statistical variability of the instrument.

In ground penetrating radar (GPR) antipersonnel mine sensing, in which the target is small, shallow and often of low dielectric contrast, detection is challenging. One of the difficulties is that it is hard to distinguish the target signal from the omnipresent random rough ground reflection clutter. In this work, a Monte Carlo computational simulation using 2-dimensional (2-D) transverse magnetic (TM) finite difference time domain (FDTD) with multiple rough surfaces is implemented to investigate single TNT target buries in dispersive soil. Based on the effects of the random rough surface on an impulse GPR signal and the knowledge of wave propagation differences in different media - air, soil, and TNT - a special background average process using physics based signal processing (PBSP) is performed to remove the ground clutter signal. This procedure first involves shifting and scaling multiple time signals from target-free random rough ground to establish the nominal (average) ground reflection pulse shape. Next, this nominal pulse shape is correlated in time with each trial signal, then shifted and scaled to match the ground surface clutter of that trial signal. Subtracting this shifted scaled clutter signal from the trial signal ideally leaves the target signal (with some additional multiple scattering between the target and ground surface). The PBSP algorithm reapplied in cases for which surface scattering occurs at multiple points. The statistical results of PBSP surface clutter removal indicate that the detection performance degrades with increasing surface roughness and decreasing burial depth. Hypothesis testing on the processed results proved to be successful in a detection and estimation point of view. This paper presents the detection performances in terms of Receiver Operating Characteristics (ROC) for various ground surface roughness and target burial depth cases. Also demonstrated is the performance improvements expected from multiple views: indicating that a multi-bistatic configuration appears to be superior to multistatic transmitter/receiver geometry with minimum combinations.

Scattering of electromagnetic waves from multilayered random rough surfaces is crucial for subsurface sensing applications. A multiple interaction method of moments (MoM) model is used in this work to analyze scattering from two-dimensional multilayered random rough ground (3-D scattering problem) especially when the underground layer is deeply buried under the air/ground interface. The presented model removes a barrier and enables the application of the Steepest Descent Fast Multipole Method (SDFMM) to certain 3-D non-quasi-planar structures. The conventional SDFMM has been used to analyze electromagnetic wave scattering from quasi-planar structures where the scatterer's height is a fraction of a free-space wavelength. The presented model is based on multiple interactions mechanism between the air/ground interface and the buried underground layer. The basic idea of the proposed multiple scattering model is to decompose the non-quasi-planar multilayered ground into two quasi-planar scatterers where the conventional SDFMM can be applied separately to each one. The interactions between the sub-quasi-planar scatterers are calculated using the electromagnetic vector potentials near-field expressions. This model is tested and validated with the MoM on a variety of geometries. The results show that the strongest signature of the buried scatterer is mainly due to the first multiple interaction mechanism (ground-object-ground) while the contributions from repeating this mechanism become insignificant even for lossless and/or slightly lossy underground.

At present the most effective mechanical aids for the post conflict hand clearance of anti-personnel mines are metal detectors and probes. These are effective against the majority of current mine threats but clearance rates are limited because of the high incidence of false targets in post conflict areas. Such false targets must be exposed and removed with the same care required for handling genuine ordnance. Clearance rates would be substantially improved if false targets detected by metal detectors could be distinguished from mine threats and thus left in place. One possible approach to the problem of differentiating between metal fragments and anti-personnel land mines is the use of multiple sensors. In this paper we discuss the design of a GPR for such a multi-sensor detector head. One of the challenges for combined metal detectors and GPR is the design of the GPR antenna so that it can operate effectively in the presence of metal detector coils. For a practicable device the GPR antennas must operate with the metal detector coils in their near field and coupling between sensors is of primary importance. The antennas must also be designed so that their influence on the metal detector's sensitivity is minimized. In this paper we present one solution for this problem and present experimental results showing the how the proposed GPR design operates in the presence of metal detector coils and in the presence of a resistive transducer located below the antenna array. The GPR concerned uses a 3x3 antenna array and post reception synthetic aperture processing to provide a 3d image of the ground underneath the sensor. Focussed images of various targets are presented, and images to demonstrate the effects of the other sensors on the GPR are shown.

Minimum classification error (MCE) training is proposed to improve performance of a discrete hidden Markov model (DHMM) based landmine detection system. The system (baseline) was proposed previously for detection of both metal and nonmetal mines from ground penetrating radar signatures collected by moving vehicles. An initial DHMM model is trained by conventional methods of vector quantization and Baum-Welch algorithm. A sequential generalized probabilistic descent (GPD) algorithm that minimizes an empirical loss function is then used to estimate the landmine/background DHMM parameters, and an evolutionary algorithm based on fitness score of classification accuracy is used to generate and select codebooks. The landmine data of one geographical site was used for model training, and those of two different sites were used for evaluation of system performance. Three scenarios were studied: apply MCE/GPD alone to DHMM estimation, apply EA alone to codebook generation, first apply EA to codebook generation and then apply MCE/GPD to DHMM estimation. Overall, the combined EA and MCE/GPD training led to the best performance. At the same level of detection rate as the baseline DHMM system, the proposed training reduced false alarm rate by a factor of two, indicating significant performance improvement.

Planning Systems Incorporated (PSI) has developed a promising Ground Penetrating Synthetic Aperture Radar (GPSAR) system to detect buried landmines. GPSAR can be used to generate three-dimensional (3-D) mine images. It has been shown that the SAR processing in the PSI GPSAR system can greatly improve the image quality and hence the mine (especially plastic mine) detection performance. In this paper, two special issues on SAR processing for the PSI system are addressed. One issue is the analysis of the effect of the underground electromagnetic (EM) wave propagation velocity uncertainty on SAR processing and the other is channel mismatch on SAR processing. Since the EM wave propagation velocity in the soil depends on many factors and changes from one location to another, velocity uncertainty is inevitable. However, we have found that the PSI GPSAR system is very robust against the velocity uncertainty. More specifically, velocity uncertainty does not defocus the image but only scales the image along the depth dimension, and hence will not affect the mine detection performance. Another issue is how to select a good SAR processing scheme for the PSI system. Because the radar footprint is 2-D (along-track and cross-track dimensions), 2-D SAR processing may be used. However, the effectiveness of the 2-D SAR processing depends on the coherence of the radar antenna system. Moreover, the computational expense of the 2-D SAR processing is much higher than that of the 1-D SAR processing (along-track dimension only). We have found that due to the channel mismatch of the PSI system, the 2-D SAR processing does not greatly improve the quality of the SAR images when compared with 1-D SAR processing. Hence, without proper antenna calibration, the computationally more efficient 1-D SAR processing may be preferred for the PSI system.

Detection of landmines by vapor-phase sensing of key chemical signature compounds was first demonstrated in 1998 using a sensor we developed as part of the DARPA Dog's Nose Program. This sensor utilizes novel fluorescent polymers to detect ultra-trace concentrations of nitroaromatic compounds emanating from explosives contained in landmines. Much has been learned about the chemical signature of landmines in recent years. For example, it has been shown that the landmine chemical signature tends to be heterogeneous and can be dispersed in the environment near the mine location. This makes it difficult to pinpoint the exact location of the mine using trace chemical detection methods. However, evidence currently available indicates that it may be possible to isolate a mine location to within a small, well-defined area. Data supporting this conclusion have been obtained using our sensor, and the conclusions drawn have been supported using other accepted laboratory analysis methods. Often, minefields contain relatively few mines. Methods of sampling suitable for rapidly isolating the mined areas from large, mine-free areas are being pursued. High-volume vapor sampling and soil particle sampling strategies are being refined for this application. Preliminary data from field tests using prototype samplers and sensors will be presented.