The problem of mine detection and localization from array- based electromagnetic induction data is addressed. An efficient forward scattering model based on the Born approximation is employed. Using insight obtained from this model, a clutter model in the form of a state space system is developed which describes the correlation of the noise both across the sensing array and from one position of the array to the next as the measurement device proceeds down track. A multiple-model detection scheme based on the whitening properties of the Kalman filter is employed to perform the actual mine detection. This approach allows for the detection and localization of buried objects well before the array physically moves over mines. Examples are provided for mines buried directly in front and off to one side of the array.

An object, made partly or wholly of metals, has a distinct combination of electrical conductivity, magnetic permeability, and geometrical shape and size. When the object is exposed to a low-frequency electromagnetic field, it produces a secondary magnetic field. By measuring the secondary field in a broadband spectrum, we obtain a distinct spectral signature that may uniquely identify the object. Based on the response spectrum, we attempt to 'fingerprint' the object. This is the basic concept of Electromagnetic Induction Spectroscopy (EMIS). EMIS technology may be particularly useful for detecting buried landmines and unexploded ordnance. By fully characterizing and identifying an object without excavation. We should be able to reduce significantly the number of false targets. EMIS should be fully applicable to many other problems where target identification and recognition (without intrusive search) are important. For instance, an advanced EMIS device at an airport security gate may be able to recognize a particular weapon by its maker and type.

Based on technologies developed for the Jet Propulsion Laboratory (JPL) Free-Flying-Magnetometer (FFM) concept, we propose to modify the present design of FFMs for detection of mines and arsenals with large magnetic signature. The result will be an integrated miniature sensor system capable of identifying local magnetic field anomaly caused by a magnetic dipole moment. Proposed integrated sensor system is in line with the JPL technology road-map for development of autonomous, intelligent, networked, integrated systems with a broad range of applications. In addition, advanced sensitive magnetic sensors (e.g., silicon micromachined magnetometer, laser pumped helium magnetometer) are being developed for future NASA space plasma probes. It is envisioned that a fleet of these Integrated Sensor Systems (ISS) units will be dispersed on a mine-field via an aerial vehicle (a low-flying airplane or helicopter). The number of such sensor systems in each fleet and the corresponding in-situ probe-grid cell size is based on the strength of magnetic anomaly of the target and ISS measurement resolution of magnetic field vector. After a specified time, ISS units will transmit the measured magnetic field and attitude data to an air-borne platform for further data processing. The cycle of data acquisition and transmission will be continued until batteries run out. Data analysis will allow a local deformation of the Earth's magnetic field vector by a magnetic dipole moment to be detected. Each ISS unit consists of miniaturized sensitive 3- axis magnetometer, high resolution analog-to-digital converter (ADC), Field Programmable Gate Array (FPGA)-based data subsystem, Li-batteries and power regulation circuitry, memory, S-band transmitter, single-patch antenna, and a sun angle sensor. ISS unit is packaged with non-magnetic components and the electronic design implements low-magnetic signature circuits. Care is undertaken to guarantee no corruption of magnetometer sensitivity as a result of its close proximity with the electronics and packaging materials. Accurate calibration of the magnetometer response in advance will allow removing the effects of unwanted disturbances. Improvements of the magnetometer performance in the areas of the orthogonality, drift, and temperature coefficient of offset and scale factor are required.

A comparison between two forward solving methods, recursive T- matrix and FDTD, is presented. The strengths and weaknesses of both methods are discussed. The recursive T-matrix method is a fast solver that is well suited to solutions in homogeneous media where the scatterers are bodies of rotation. The FDTD method is best suited to complicated and realistic problems involving inhomogeneities, rough surface interfaces and irregular shaped scatterers. Numeric results from both methods are presented.

A principal problem with traditional, narrowband EMI sensors involves target identification. As a consequence, in minefield or unexploded ordinance (UXO) detection, for example, each piece of buried metal must be excavated, causing significant false alarms in regions littered with anthropic clutter. Therefore, the principal challenge for the next generation of EMI sensors is development of electronics and algorithms which afford discrimination. To this end, in this paper we operate in the frequency domain, considering wideband excitation and utilize the complex, frequency-dependent EMI target response as a signature. To test the signature variability of different metal types and target shapes, as well as for calibration of an actual sensor, we have developed a full-wave model for the analysis of wideband EMI interaction with highly (but not perfectly) conducting and permeable targets. In particular, we consider targets which can be characterized as a body of revolution, or BOR. The numerical algorithm is tested through use of a new wideband EMI sensor, called the GEM-3. It is demonstrated that the agreement between measurements and theory is quite good. Finally, we consider development of signal processing algorithms for the detection and identification of buried conducting and permeable targets, using wideband data. The algorithms are described and then tested on data measured using the GEM-3, with results presented in the form of contour plots as a function of the number of discrete frequencies employed.

This paper addresses the issue of identifying conduction objects based on their response to low frequency magnetic fields -- an area of research referred to by some as magnetic singularity identification (MSI). Real time identification was carried out on several simple geometries. The low frequency transfer function of these objects was measured for both cardinal and arbitrary orientations of the magnetic field with respect to the planes of symmetry of the objects (i.e., different polarizations). Distinct negative real axis poles (singularities) associated with each object form the basis for our real-time identification algorithm. Recognizing this identification problem as one of inference form incomplete information, application of Bayes theorem leads to a generalized likelihood ratio test (GLRT) as a solution to the M-ary hypothesis testing problem of interest here. Best performance, measured through Monte Carlo simulation presented in terms of percent correct identification versus signal-to- noise ratio, was obtained with a single pole per object orientation.

The purpose of the present work is finding of opportunities of subsurface sensing with the help of a continuous noise signal. For a basis the reflectometer used for the definition of soil moisture is taken. Model calculations for cases of large flat (l greater than (lambda) ) and small (l less than (lambda) ) spherical objects were carried out in order to estimate the object detection depth and its dependence on the bandwidth of the device and physical properties of the ground. The calculation was made both for monochrome and for broadband noise signals. As a result of modeling and experimental researches it is found out that the use of a broadband signal improves the conditions of recognition such an object which is at a small depth on its reflectivity. For detecting an object in a homogeneous medium without recognition its type the set of 3 - 8 monochrome signals distinguished by their frequency is more suitable. The broadband signal in some cases has an advantage in detecting of object in the ground with large- scale inhomogeneous media of permittivity.

The extent of and horrifying consequences of anti-personnel mines call for urgent improvements in current practices and development of new technologies for detecting mines. Such developments should clearly add value to the various mine clearance activities and should be cost effective investments. This implies that such improvements need to address the real market needs. This article consists of two parts. The first part sets out to formulate the market needs. This is followed by discussions on some features of the market with reference to a number of known frameworks adapted to this end. The need for developments in geographic information systems (GIS) and advanced mine detection products are highlighted. It is argued that these technologies possess the potential to create value for the market and should be thoroughly exploited. The second part deals with the development of advanced mine detectors. Some micro-economic aspects of mine detectors are discussed with reference to unit operations cost and a cost-benefit model is introduced to assist purchase evaluation of advanced mine detectors. The stages of a mine detector life cycle are presented; it is shown that the conception phase is crucial to the successful market development of these detectors. Finally, barriers to advanced mine detection technology development are discussed and some recommendations are put forward to manage the risks involved.

The Mine/Minefield detection community's maturing technology base has become a developmental resource for world wide military and humanitarian applications. During the last decade, this community has developed a variety of single and multi-sensor applications incorporating a diversity of sensor and processor technologies. These diverse developments from the Mine/Minefield detection community require appropriate metrics to objectively bound technology and to define applicability to expected military and humanitarian applications. This paper presents a survey of the test methodology, application and analysis activities conducted by the U.S. Army Communications and Electronics Command's, Night Vision and Electronic Sensors Directorate (NVESD) on behalf of the Mine/Minefield detection community. As needs of world wide military and humanitarian mine detection activities are being responded to by notable technology base advances, a diverse pool of knowledge has been developed. The maturity of these technology base advances must be evaluated in a more systematic method. As these technologies mature, metrics have been developed to support the development process and to define the applicability of these technology base advances. The author will review the diversity of the mine detection technology and their related testing strategies. Consideration is given to the impact of history and global realism on the U.S. Army's present mine detection testing program. Further, definitions of testing metrics and analysis will be reviewed. Finally the paper will outline future U.S. Army testing plans with a special consideration given to the Vehicular Mounted Mine Detection/Ground Standoff Mine Detection System (VMMD/GSTAMIDS) Advanced Technology Demonstration and related issues.

In minefield detection, two main types of operation can be identified. First, there is the detection of surface-laid minefield. This scenario is encountered largely in tactical operations (troop movement, beach landing) where the speed at which the minefield is deployed or the strategic barrier that they represent exceed the need to bury them. Second, there is the detection of buried minefield which is encountered mainly in peacekeeping missions or clearance operations. To address these two types of minefield detection process, we propose an airborne far-infrared minefield imaging system (AFIRMIS). This passive and active imaging system fuses the information from the emissivity, the reflectivity and the 3-dimensional profile of the target/background scene in order to improve the probability of detection and to reduce the false alarm rate. This paper describes the proposed imaging system and presents early active imaging results of surface-laid mines.

The problem of mine and minefield detection continues to provide a significant challenge to sensor systems. Although the various sensor technologies (infrared, ground penetrating radar, etc.) may excel in certain situations there does not exist a single sensor technology that can adequately detect mines in all conditions such as time of day, weather, buried or surface laid, etc. A truly robust mine detection system will likely require the fusion of data from multiple sensor technologies. The performance of these systems, however, will ultimately depend on the performance of the individual sensors. Infrared (IR) polarimetry is a new and innovative sensor technology that adds substantial capabilities to the detection of mines. IR polarimetry improves on basic IR imaging by providing improved spatial resolution of the target, an inherent ability to suppress clutter, and the capability for zero (Delta) T imaging. Nichols Research Corporation (Nichols) is currently evaluating the effectiveness of IR polarization for mine detection. This study is partially funded by the U.S. Army Night Vision & Electronic Sensors Directorate (NVESD). The goal of the study is to demonstrate, through phenomenology studies and limited field trials, that IR polarizaton outperforms conventional IR imaging in the mine detection arena.

The paper discusses a new method of land-mine detection with a multi-stage algorithm for an infrared (IR) sensor mounted on a vehicle. In recent years, IR-based detection systems have gained increasing interest for the land-mine detection problem. We propose a novel processing technique for the IR sensor. The method consists of five algorithmic stages: (1) pre-processing, (2) preliminary detection, (3) feature extraction, (4) classification, and (5) combination of multiple classifiers. The pre-processing step uses a novel adaptive filtering technique that enhances the mines with respect to the background by an online algorithm. The pre- processed IR image is analyzed by parametric and non- parametric methods of testing differences of means for populations from two distributions. We identify candidate regions of interest at the preliminary detection stage, from which we extract features for classification. We use three different classifiers which are based upon a (1) probabilistic neural network, (2) decision tree, and (3) multi-layer feed- forward neural network. Results from multiple classifiers are combined using a new technique of dynamic classifier selection. We apply the complete algorithm on images acquired by IR cameras mounted on a vehicle, and the preliminary test results are very encouraging.

In this paper we present results we obtained in mine detection, in the course of a multi-national European research program. Trials were performed in the Joint Research Center in ISPRA, using polarimetric infrared imagers. Usually the 3 - 5 micrometer spectral band is used for this application, however we explain that the 8 - 12 micrometer band is physically a better choice. We thus obtained information on the polarization of the self emitted radiations of the objects so that our method should be more versatile regarding the environment. Images of the global intensity, the radiation global ellipticity and orientation are presented on several types of mines. The obvious increase of contrast between the observed mines and the clutter demonstrates the usefulness of this technique in mine and UXO detection.

We describe a method for recognizing surface-lying ordnance in test ranges using stereo range information and image edge maps. This method is to be used by an unmanned ground vehicle (UGV) surveying the test range for autonomous clearing of ordnance. We concentrate on a particular type of cylindrical ordnance (BLU-97) in current usage in U.S. military test ranges. In order to locate instances of the ordnance, we employ a stereo pair of cameras to be mounted on top of a UGV. Parallel segments corresponding to the occluding contours of the ordnance are detected in the imagery using robust and efficient model extraction techniques. The stereo range data is used to adaptively select the local scale for edge detection and to place constraints on the search space for the parallel segment extraction. Initial tests indicate that robust recognition is possible in near real-time with a low rate of false positives.

The detection and recognition of submerged objects via airborne or spaceborne imaging platforms can be problematic, due to distortions resulting from refraction at the sea surface as well as absorption and scattering in atmospheric and aqueous media. In previous research, we showed that, given point-to-point measurement of interfacial topography and estimates of media optical parameters, it is often possible to partially invert refractive effects and compensate for first- order optical effects such as absorption and single scattering, thereby producing a visually acceptable, image- based reconstruction of a submerged target. Additionally, information concerning target depth or surface topography is obtained. Although potentially useful for human-in-the-loop viewing scenarios, this method is not universally applicable to automated target recognition (ATR) due to reconstruction errors that tend to cluster in regions of detail. In this paper, we present theory and error analysis pertaining to stereoscopic viewing through a corrugated refractive interface and scattering media. Via stereoscopic imaging, one can produce a stereo pair image from which target depth and salient features can be extracted mathematically. Estimation of errors involved in this process is facilitated by partial knowledge of interfacial topography and is further enhanced by knowledge of media optical parameters. Analyses emphasize effects of sea state, topography estimation error, media optical parameters, and submergence depth. Reconstruction algorithm complexity is discussed in terms of work requirement applicable to sequential workstations and parallel computers.

Detection of buried objects presents a formidable challenge which requires many different approaches. Infrared imaging has proven its versatility in a number of applications. Recent advances in technology have opened the door for spectroscopic imaging systems which can produce images of reflectivity or emissivity as a function of two spatial dimensions and wavelength. These images have been largely unexploited for detection of buried and surface-laid landmines. Several promising opportunities exist for this application in different parts of the infrared spectrum. Variations in soil moisture content, vegetation condition, and soil composition may well be related to the presence of shallow-buried objects. In addition, polarimetric signatures appear useful in detecting man-made objects on the surface and may even help in detecting buried objects. This paper will explore both the feasibility of using infrared spectral imagery in the 1-to-2.5 and 8-to-12 micrometer infrared bands to detect surface-laid and buried objects.

Coastal Systems Station under the sponsorship of the Marine Corps Amphibious Warfare Technology Directorate are exploring the use of a Passive Millimeter Wave (PMMW) sensor for stand off airborne mine detection. In the development of any new technology application, there exist a critical need to develop a balanced modeling and measurement capability. Both will complement one another. Nichols Research has established a physics-based image modeling capability for Passive Millimeter Wave (PMMW) systems. This modeling capability has been used to estimate the performance of a PMMW mine detection system. But, in order to accurately predict the performance of a PMMW imaging system, the background clutter characteristics must be characterized and the modeling results verified against measured data. In fact, in the case of a well designed sensor, the background clutter will define the systems overall performance making accurate knowledge of the clutter statistical variations critical. However currently, there is a lack of high resolution PMMW imagery of backgrounds, due to a lack of data collection instrumentation. This paper will present the results from a preliminary PMMW data collection to provide data for the assessment of a PMMW mine detection system. The data collection results will characterize both surface and buried mine detection capabilities under a variety of conditions. It is a well-established fact that no single sensor will be capable of solving the mine detection problem. Instead, a suite of complementary sensors is required. There is however a lack of an extensive data set of sensor modalities collected in a single sample area. Therefore as a secondary objective of this data collection, several sensor modalities will be used to simultaneously collect mine and minefield data. These results will also be presented.

A hybrid technique is presented that simultaneously uses both electromagnetic and acoustic waves in a synergistic manner to detect buried land mines. The system consists of an electromagnetic radar and an acoustic source. The acoustic source causes both the mine and the surface of the earth to be displaced. The electromagnetic radar is used to detect these displacements and, thus, the mine. To demonstrate the viability of this technique, an experimental system has been constructed. The system uses an electrodynamic transducer to induce an acoustic surface wave, a tank filled with damp sand to simulate the earth, a simulated mine, and a radar to measure the vibrations. The technique looks promising; we have been able to detect both simulated antipersonnel mines and antitank mines buried in damp sand from the experimental results obtained with the system.

An experiment series to determine the ability of laser vibrometers to receive and process echoes from the water surface culminates in a 28 inch by 28 inch aperture synthesis. Highly resolved three dimensional imagery of a submerged three liter beverage bottle, suspended in the free field is presented. Excitation is accomplished by a submerged acoustic source transmitting a swept FM pulse near 70 kilohertz with a 10 kilohertz sweep. The receive laser head is suspended 19 inches above the water surface, and moved by an inverted x-y positioning table to create a three dimensional synthetic aperture. Temporal oversampling and common trigger for transmitter and data acquisition system are used to establish coherence. In an effort to better resolve the target, the aperture is made larger than previous attempts. Experimental results are reported.

Possibilities for increased mine detection and classification techniques have established a need for broadband, underwater acoustic projectors. An advanced version of a low frequency synthetic aperture sonar (SAS) for the mine reconnaissance hunter program has recently been developed. The transducer is resonant at 100 kHz but has been designed to deliver constant high sound pressure levels over an operating frequency range of 10 kHz to 100 kHz. This wide band operation is accomplished because of an absence of spurious modes within the operational frequency decade. The actual projector is constructed with a two layered 1 - 3 piezocomposite material stacked in mechanical series and electrically wired in parallel. This arrangement was selected in order to maximize the source level output. The center electrode of the monolithic 1 - 3 piezocomposite layers has been segmented to offer four individual elements such that combinations of the sectors offer the ability to access nine different apertures. A constant source level is maintained through the use of a preshaped transformer between the driver and the projector. The combination of the transformer design with the clean spectrum response of the composite material results in an acoustic projector with constant source level.

The Naval Research Laboratory is using its world-renowned structural acoustics facilities (originally developed for scaled submarine programs) to study the broad band (1 - 150 kHz) acoustic scattering from proud and buried underwater mines. The objective is to discover what information is contained in the broad-band properties of the scattered signal which might be exploited for target identification purposes. Current acoustic mine-hunting systems form acoustic images that replicate the rough geometric shape of the target. To obtain sufficient resolution, these systems must operate at frequencies that are too high for anything but time-consuming, close-in looks at the target. Even then, they often confuse mines with mine-like targets such as oil drums. In contrast, structural acoustic clues such as mine resonances, elastic wave propagation, internal structure scattering, etc., are available at lower frequencies (1 - 10 kHz), allowing for much longer ranges of operation as well as the construction of unique 'fingerprints' by which to identify the target as a mine. Additionally, at lower frequencies the ocean sediment is more readily penetrated by acoustic waves, creating the possibility for buried mine detection. This paper examines the feasibility of exploiting such very low frequency structural acoustic clues for long range identification of proud and buried mines.

Supported by ONR for the past decades, researchers have succeeded in 'imaging' underwater with background noise, 'Acoustic Daylight.' In this paper, the authors will discuss their success in detecting underwater objects in background noise. This detection method is not an imaging technique. It is a broadband resonant scattering detection technique based on the theory of resonance and/or resonant scattering of the elastic sphere illuminated by 'Acoustic Daylight.' Using the resonant detection technique, it appears that underwater targets can be detected and identified in background noise, 'Acoustic Daylight.'

The Mobile Underwater Debris Survey System (MUDSS) is a collaborative effort by the Navy and the Jet Propulsion Lab to demonstrate multi-sensor, real-time, survey of underwater sites for ordnance and explosive waste (OEW). We describe the sonar processing algorithm, a novel target recognition algorithm incorporating wavelets, morphological image processing, expansion by Hermite polynomials, and neural networks. This algorithm has found all planted targets in MUDSS tests and has achieved spectacular success upon another Coastal Systems Station (CSS) sonar image database.

Traditional Receiver Operating Characteristic (ROC) Curve Confidence Bound Construction Methods are based on the use of asymptotic results or idealized models. In this work, an approach to ROC confidence bound construction is developed for both Per-Token and Cumulative ATR system performance, based on the use of the Bootstrap Method. This technique is nonparametric, and has been applied to problems in which traditional confidence bounds would be difficult, if not impossible, to construct. In addition, the technique is able to encompass the effects of training data and evaluation data variability in a single unified approach. Results are presented contrasting parametric and Bootstrap based cumulative ROC curve confidence bounds for three distinct sets of side-scan sonar data. Finally, Bootstrap based bounds are developed in the context of Acoustic/Magnetic fusion systems.

An advanced, automatic, adaptive clutter suppression, pre- and post-detection level fusion, sea mine detection and classification processing string has been developed and tested with sonar imagery data. The overall string includes preprocessing, adaptive clutter filtering (ACF), normalization, detection, feature extraction and classification processing blocks. The ACF is a multi- dimensional adaptive linear FIR filter, optimal in the Least Squares sense, and is applied to low resolution data. It performs simultaneous background clutter suppression and preservation of an average peak target signature (normalized shape of mine highlights -- computed a priori using training set data). After data alignment, using a 3-dimensional ACF enables simultaneous multiple frequency data fusion and clutter suppression in the composite frequency-range- crossrange domain. Following 2-d normalization, the detection consists of thresholding, clustering of exceedances and limiting the number of detections. Finally, features are extracted from high resolution data and a orthogonalization transformation is applied to the features, enabling an efficient application of the optimal log-likelihood-ratio-test (LLRT) classification rule. Various fusion approaches were investigated including pre-detection fusion using the 3-d ACF string, post-detection fusion of the 2d ACF strings and an overall fusion of the two previous strings. The utility of the various processing strings was demonstrated with two new shallow water high resolution sonar imagery data sets. The overall ACF, pre- and post-detection level fusion, feature orthogonalization, LLRT-based classification processing string provided mine classification capability and false alarm rate performance exceeding the one of an expert sonar operator. A wide-sense stationary covariance model was utilized in the ACF algorithm design, significantly reducing the algorithm implementation complexity, thus enabling an easy implementation of the overall processing string in real-time.

This paper presents a novel formulation of an adaptive order- statistic filter, and describes the performance enhancements it provides to an automatic sea mine classification system. Non-linear filters based on order statistics (median, 'largest-of,' etc.) have been shown to be effective in suppressing noise with long, heavy-tailed density functions (e.g., Laplacian), and they have also been successfully used to suppress 'salt-and-pepper' noise in image processing, as well as transients and Raleigh-distributed speckle noise in ultrasound imaging. Such 'order-statistic' filters can be adaptively generalized and optimized, for a given data set, by finding the weights that, operating on ordered data samples, minimize filter output power while preserving signals that are constant within the filter window. Morphological filters can also be optimized in this manner, since they have been shown to consist of combinations of order-statistic filters. A new adaptive order-statistic filter formulation, enabling the preservation of signals that are not constant within the filter window, has been developed and its efficacy demonstrated with side-scan sonar imagery data. Using these filters as a non-linear 'corrector' of the outputs of the linear clutter-filtering stage of a sea mine classification system, reduced the number of false alarms by an order of magnitude.

One of the biggest challenges in distributed underwater mine warfare for area sanitization and safe power projection during regional conflicts is transmission of compressed raw imagery data to a central processing station via a limited bandwidth channel while preserving crucial target information for further detection and automatic target recognition processing. Moreover, operating in an extremely shallow water with fluctuating channels and numerous interfering sources makes it imperative that image compression algorithms effectively deal with background nonstationarity within an image as well as content variation between images. In this paper, we present a novel approach to lossy image compression that combines image- content classification, content-adaptive bit allocation, and hybrid wavelet tree-based coding for over 100:1 bandwidth reduction with little sacrifice in signal-to-noise ratio (SNR). Our algorithm comprises (1) content-adaptive coding that takes advantage of a classify-before-coding strategy to reduce data mismatch, (2) subimage transformation for energy compaction, and (3) a wavelet tree-based coding for efficient encoding of significant wavelet coefficients. Furthermore, instead of using the embedded zerotree coding with scalar quantization (SQ), we investigate the use of a hybrid coding strategy that combines SQ for high-magnitude outlier transform coefficients and classified vector quantization (CVQ) for compactly clustered coefficients. This approach helps us achieve reduced distortion error and robustness while achieving high compression ratio. Our analysis based on the high-frequency sonar real data that exhibit severe content variability and contain both mines and mine-like clutter indicates that we can achieve over 100:1 compression ratio without losing crucial signal attributes. In comparison, benchmarking of the same data set with the best still-picture compression algorithm called the set partitioning in hierarchical trees (SPIHT) reveals that some weak targets can completely disappear in certain situations because SPIHT is not content adaptive.

The Advanced Volume Visualization Display (AVVD) research program is a joint research program between the Fraunhofer Center for Research in Computer Graphics, Inc. and Innovative Research and Development Corp. It is dedicated the application of the human visual system to real-time visualization of high- resolution volumetric sensor data sets. The AVVD program has successfully demonstrated its application to undersea imaging using data from the Naval Undersea Warfare Center -- Division Newport's High Resolution Array (HRA), and from the Naval Surface Warfare Center -- Coastal System Stations's Toroidal Volume Search Sonar (TVSS).

A critical part of automatic classification algorithms is the extraction of features which distinguish targets from background noise and clutter. The focus of this paper is the use of variational methods for improving the classification of sea mines from both side-scan sonar and laser line-scan images. These methods are based on minimizing a functional of the image intensity. Examples include Total Variation Minimization (TVM) which is very effective for reducing the noise of an image without compromising its edge features, and Mumford-Shah segmentation, which in its simplest form, provides an optimal piecewise constant partition of the image. For the sonar side-scan images it is shown that a combination of these two variational methods, (first reducing the noise using TVM, then using segmentation) outperforms the use of either one individually for the extraction of minelike features. Multichannel segmentation based on a wavelet decomposition is also effectively used to declutter a sonar image. Finally, feature extraction and classification using segmentation is demonstrated on laser line-scan images of mines in a cluttered sea floor.

Statistical and neural network algorithms are used to separate mine targets from clutter in side scan sonar images. In these images, a typical target usually contains in excess of 100 pixels filled with salt and pepper noise. This translates into a problem of classifying in a complicated high dimensional space, which is very difficult if not impossible to solve. Therefore, a typical mine detection algorithm contains three stages preceding the classification algorithm: noise reduction, clutter rejection, and feature extraction. These pre-processing steps would reduce the dimension of the feature space by an order of magnitude. Side scan sonar images are known to be contaminated with noise and mine like clutter. The major challenge is to select and measure the features of the potential targets. This is frequently done by fractal and/or Fourier analysis. Recently, wavelet analysis has also been used successfully as a tool for feature extraction. However, there are few analytical rules to guide the selection of features. In this paper, we investigate a new integrated feature extraction and classification algorithm that first enhances a potential target using variational based algorithms, and then transforms the enhanced image into a set of wavelet channels. We use the multichannel information as inputs to a feed-forward neural network. This new classifier has the advantage of extracting not only the local features but also the background features through higher scale wavelet channels. Results are compared for different network designs.

We investigate various image enhancement techniques geared towards a specific detector. Our database consists of side- scan sonar images collected at the Naval Surface Warfare Center (NSWC), and the detector we use has proven to have excellent results on these data. We start by investigating various wavelet and wavelet packet denoising methods. Other methods we consider are based on more common filters (Gaussian and DOG filters). In wavelet based denoising we try different approaches, combining techniques that have been successfully used in signal and image denoising. We notice that the performance is mostly affected by the choice of the scale levels to which shrinkage is applied. We demonstrate that wavelet denoising can significantly improve detection performance while keeping low false alarm rates.

Signal features based on multiresolution short-time Fourier transforms (STFT) and the Morlet wavelet transform (MWT) have been developed to classify echo returns from targets ensonified by simulated dolphin echolocation clicks. Spectrogram features are obtained at different scales of resolution using analysis windows of different sizes. A method of compressing the highly redundant time-scale representations provided by the MWT has been developed based on multiscale edge analysis (MSEA) of wavelet local maxima. Neural networks are used to evaluate the efficacy of the various feature sets for target recognition. Hierarchical neural networks are used to combine different feature sets for improved classification performance.

The Local Discriminant Basis (LDB) approach of Saito and Coifman was compared to Principal Components Analysis (PCA) for feature extraction prior to automatic classification. LDB finds an orthonormal basis from a library of local basis vectors (e.g., cosine packets) that maximizes a discriminative measure for the given set of training data. The data consist of broadband sonar data collected from mine-like targets and calibration targets in shallow water on the bottom of Puget Sound in Washington State. The database consists of twelve sequences of data for each of 10 separate objects. Each signal contains 481 samples. The lowest error rate achieved with PCA was 9.2%, or 11 out of 120 errors, using the first 45 principal component vectors and performing classification with Fischer's Linear Discriminant. The corresponding error rate for LDB was also 9.2%, though using fewer (34) coefficients. Comparable results were obtained using CART (Classification Trees) with very few coefficients: 8.3% error rate for LDB with 7 coefficients; 10% for PCA with only 4 coefficients. A modified LDB, which used a robust version of Fisher's separability (the ratio of between-class variation to within- class variation) instead of normalized energy as the discriminative measure, reduced the error rate to 9/120 (7.5%). Another method of improving the basis selection, called 'cycle spinning,' reduced the error rate to 7/120 (5.8%). Thus, LDB yields consistently better classification rates than principal components feature extraction. An added advantage of LDB is speed: it is an O[Nlog(N)²] procedure for cosine packets, while principal components is O(N³).

This paper presents the application of a novel scheme for dynamic structural adaptation for back-propagation neural networks. It utilizes the time and order update formulations in the orthogonal projection theorem to establish a recursive weight updating procedure for the training process and a dynamic node creation procedure during the training process. The effectiveness of the algorithm is demonstrated on a simple multiplexer problem and a real-life application dealing with underwater target classification from the acoustic backscattered signals. It is shown through the simulation results that the dynamic structural adaptation scheme offers better trainability for the networks without requiring prohibitive cost of retraining. In addition, the results on the testing data indicate good classification performance of the network trained in conjunction with the structural adaptation method.

ART (Adaptive Resonance Theory) neural networks for fast, stable learning and prediction have been applied in a variety of areas. Applications include automatic mapping from satellite remote sensing data, machine tool monitoring, medical prediction, digital circuit design, chemical analysis, and robot vision. Supervised ART architectures, called ARTMAP systems, feature internal control mechanisms that create stable recognition categories of optimal size by maximizing code compression while minimizing predictive error in an on- line setting. Special-purpose requirements of various application domains have led to a number of ARTMAP variants, including fuzzy ARTMAP, ART-EMAP, ARTMAP-IC, Gaussian ARTMAP, and distributed ARTMAP. A new ARTMAP variant, called ARTMAP- FTR (fusion target recognition), has been developed for the problem of multi-ping sonar target classification. The development data set, which lists sonar returns from underwater objects, was provided by the Naval Surface Warfare Center (NSWC) Coastal Systems Station (CSS), Dahlgren Division. The ARTMAP-FTR network has proven to be an effective tool for classifying objects from sonar returns. The system also provides a procedure for solving more general sensor fusion problems.

Undetected sea mines in a littoral environment are dangerous threats that must be first detected and then avoided or neutralized in the conduct of strategic and tactical warfare. The U.S. Navy is seeking enabling sensor suites and associated algorithms that allow autonomous underwater vehicles to search, detect and destroy sea mines. Acoustic backscatter is a sensing mechanism that permits searches to be conducted at comparatively long ranges and thus would enable high area coverage rates. The research problem addressed in this paper is the development of an algorithm that allows acoustic backscatter to be used to detect and classify mines and mine like objects (MLOs). This paper presents a novel approach of fusing and classifying multiple acoustic backscatter signals for the purpose of identifying mines and mine-like objects at long ranges. The algorithm relies on an underlying database of measured target signatures for classification purposes and uses a set of quick search templates that encapsulate the target information contained in this 'knowledge-pool' database. The templates are mathematically structured to permit database searches to be performed in real time with low to moderate computational resources. The mathematical structure of the search templates is hierarchical in nature and allows the signal processing tasks of mine detection, discrimination, and identification to be performed by a single integrated system in a progressive manner. This classification system also knows when data of an unknown nature is encountered.

The classification of mine-like and non-mine-like objects using acoustic backscatter data is considered. Backscatter data for 6 objects at 5 degree intervals in aspect is used. The data preprocessing is described; this includes location of the starting point of the received signals and multipath removal. Various features and classifiers are considered. Excellent performance is achieved.

The solubility properties of a series of nitroaromatic compounds have been determined and utilized with known linear solvation energy relationships to calculate their sorption properties in a series of chemoselective polymers. These measurements and results were used to design a series of novel chemoselective polymers to target polynitroaromatic compounds. The polymers have been evaluated as thin sorbent coatings on surface acoustic wave (SAW) devices for their vapor sorption and selectivity properties. The most promising materials tested, include siloxane polymers functionalized with acidic pendant groups that are complimentary in their solubility properties for nitroaromatic compounds. The most sensitive of the new polymers exhibit SAW sensor detection limits for nitrobenzene and 2,4-dinitrotoluene at 3 parts per billion (ppb) and 235 parts per trillion (ppt) respectively. Optimized polymers exhibit low water vapor sorption, and rapid signal kinetics for nitrobenzene, reaching 90% of signal response in 4 seconds. Studies with an in-situ infra-red spectroscopy technique are used to determine the mechanism of interaction between nitroaromatic compounds and the chemoselective polymer.

Response data were collected for a carbon black-polymer composite electronic nose array during exposure to homologous series of alkanes and alcohols. At a fixed partial pressure of odorant in the vapor phase, the mean response intensity of the electronic nose signals varied significantly for members of each series of odorants. However, the mean response intensity of the electronic nose detectors, and the response intensity of the most strongly-driven set of electronic nose detectors, was essentially constant for members of a chemically homologous odorant series when the concentration of each odorant in the gas phase was maintained at a constant fraction of the odorant's vapor pressure. Because the thermodynamic activity of an odorant at equilibrium in a sorbent phase is equal to the partial pressure of the odorant in the gas phase divided by the vapor pressure of the odorant, and because the activity coefficients are similar within these homologous series of odorants for sorption of the vapors into specific polymer films, the data imply that the trends in detector response can be understood based on the thermodynamic tendency to establish a relatively constant concentration of sorbed odorant into each of the polymeric films of the electronic nose at a constant fraction of the odorant's vapor pressure. This phenomenon provides a natural mechanism for enhanced sensitivity to low vapor pressure compounds, like TNT, in the presence of high vapor pressure analytes, such as diesel fuel. In a related study to evaluate the target recognition properties of the electronic nose, a statistical metric based on the magnitudes and standard deviations along Euclidean projections of clustered array response data, was utilized to facilitate an evaluation of the performance of detector arrays in various vapor classification tasks. This approach allowed quantification of the ability of a fourteen-element array of carbon black-insulating polymer composite chemiresistors to distinguish between members of a set of nineteen solvent vapors, some of which vary widely in chemical properties (e.g. methanol and benzene) and others of which are very similar (e.g. n-pentane and n-heptane). The data also facilitated evaluation of questions such as array performance as a function of the number of detectors in the system.

An integrated approach for enhanced detection of nitroaromatic compounds in the vapor phase is being investigated. The approach involves the use of a very low volume cryoconcentrator with rapid desorbing capabilities. The concentrator desorber is coupled to solute modulated detection systems. Modulation of nitroaromatics via electron attachment reactions holds potential for nitroaromatics because of high electron affinity of these compounds. Under suitable conditions, i.e., low solute concentration and high electron population, these reactions occur at coulometric limits. The results show that the concentrator system permits rapid, quantitative enrichment of semivolatile nitroaromatics. Results of electron attachment reaction show that these reactions lead to disintegration of nitroaromatic radicals while the potentially interfering compounds such as polyhalogenated aromatics undergo hydrochlorination and yield distinctive products. These differences can be used for modulation of nitroaromatic compounds at low concentrations.

We report on the development of MEMS devices for detecting explosive particles associated with anti-personnel mines. Because of the affinity of explosive substances for surfaces and owing to the high partition coefficients of explosives in soils relative to water and air, we employ remote stimulation of the soil's surface with a high intensity, focused air ultrasonic beam whose energy can megasonically clean the target area of particles above a designed-for size. We have fabricated a MEMS electrostatic transducer to test the concept. Nanogram particle detection will occur by collecting particles on an array of temperature sensitive MEMS sensors and irradiating the particles with 3 - 5 micrometer wavelength infrared light. Explosive particles will selectively absorb the infrared energy at approximately 1600 cm^-1, decompose, and give off heat which can be detected. Prototype explosive detectors have been fabricated which do not absorb energy in the peak absorption bands of the explosives, thus allowing for selective particle heating without heating the sensor device itself.

Progress on the development of a miniature mass transducer for low level vapor detection is described. Thin-film bulk acoustic wave resonators (TFRs) are fabricated with AlN as the piezoelectric layer onto an acoustically reflecting solid substrate. Q's of 200 at 2 GHz have been measured. Comparison of the sensitivity of these 2 GHz TFRs with 250 MHz surface acoustic wave resonators is made. A set of TFRs with species selective coatings has the potential of detecting TNT vapor at the sub PPT level. Detection of 5 ppb TNT with a signal to noise of 200:1 is demonstrated.

We have previously developed an optically-based 'artificial nose' to detect a wide variety of volatile organic compounds. An optical fiber sensor array is prepared containing a variety of differentially-reactive sensors comprised of polymer/dye combinations. When an analyte is presented in pulsatile form each sensor produces a unique fluorescence vs. time signature. The system employs neural network analysis to discriminate between many organic vapors using pattern recognition. Following an initial training step, the system can recognize 91 - 100% of a training set and greater than 84% of a test set of volatile organics. We are now attempting to detect explosives and explosive-like materials using this system. Prior work has shown that some sensors respond to compounds structurally similar to TNT (e.g. 2,4-DNT and 2,6-DNT) at saturated vapor concentrations. These preliminary results provide grounds for exploring the capacity of these and other new polymer/dye sensing combinations for detecting polynitro- compounds at low concentrations.

A novel sensor concept for detection of polynitroaromatic compounds has been developed in a partnership between Texas Instruments and the University of New Hampshire. The objective for this sensor is to demonstrate an explosive detection system designed specifically for field use. Our approach incorporates manufacturability and low cost while emphasizing field compatibility, usability, hand-held portability, selectivity, and sensitivity. The new device incorporates a novel multi-plate configuration and is based on colorimetric changes that occur when polynitroaromatic compounds react with polyvinylchloride polymer films containing Jeffamine T-403. Response time and characteristic absorbance for the films will be presented along with a description of the device. The results represent a first step toward a potential solution for detection of vapors utilizing chemically sensitive optical polymers.

Trace Explosive Detectors (TED) technologies have been investigated as a means of confirming the presence of a landmine at a given location. A field trial was performed with a landmine detector prototype based on Ion Mobility Spectrometry. The system was based on the detection of the explosives in soil and had a detection limit of 0.4 ppb w/w for TNT and 7.4 ppb w/w for RDX. The minefield was composed of 51 sites on which the detector performance was evaluated. For most freshly buried sites it was found that the level of explosive was below the detection limit of the prototype. In addition, a quantitative analysis of the residual explosive transfer to the soil by hands was performed. Results showed that the level transferred by hands was in most cases below 0.1 ppb for TNT and at 0.8 ppb w/w or below for RDX. However, it was found that the explosive level contained in the soil increased with time to a level around 2 - 8 ppb w/w for TNT, ten month after the landmine burial. These rough tests have yielded some preliminary results concerning the level of explosives detectable after the burial of landmines and the dynamics of the explosive level build up in the soil with time.

Sandia National Laboratories has conducted research in chemical sensing and analysis of explosives for many years. Recently, that experience has been directed towards detecting mines and unexploded ordnance (UXO) by sensing the low-level explosive signatures associated with these objects. Our focus has been on the classification of UXO in shallow water and anti-personnel/anti tank mines on land. The objective of this work is to develop a field portable chemical sensing system which can be used to examine mine-like objects (MLO) to determine whether there are explosive molecules associated with the MLO. Two sampling subsystems have been designed, one for water collection and one for soil/vapor sampling. The water sampler utilizes a flow-through chemical adsorbent canister to extract and concentrate the explosive molecules. Explosive molecules are thermally desorbed from the concentrator and trapped in a focusing stage for rapid desorption into an ion-mobility spectrometer (IMS). We will describe a prototype system which consists of a sampler, concentrator-focuser, and detector. The soil sampler employs a light-weight probe for extracting and concentrating explosive vapor from the soil in the vicinity of an MLO. The chemical sensing system is capable of sub-part-per-billion detection of TNT and related explosive munition compounds. We will present the results of field and laboratory tests on buried landmines, which demonstrate our ability to detect the explosive signatures associated with these objects.

Rapid detection and classification of odor is of particular interest in applications such as manufacturing of consumer items, food processing, drug and explosives detection, and battlefield situation assessment. Various detection and classification techniques are under investigation so that end users can have access to useful information from odor sensor arrays in near-real-time. Feature-level data clustering and classification techniques are proposed that are (1) parallelizable to permit efficient hardware implementation, (2) adaptable to readily incorporate new data classes, (3) capable of gracefully handling outlier data points and failed sensor conditions, and (4) can provide confidence intervals and/or a traceable decision record along with each classification to permit validation and verification. Results from using specific techniques will be presented and compared. The techniques studied include principal components analysis, automated outlier determination, radial basis functions (RBF), multi-layer perceptrons (MLP), and pulse-coupled neural networks (PCNN). The results reported here are based on data from a testbed in which a gas sensor array is exposed to odor samples on a continuous basis. We have reported previously that more detailed and faster discrimination can be obtained by using sensor transient response in addition to steady state response. As the size of the data set grows we are able to more accurately model performance of a sensor array under realistic conditions.

Dogs are capable of detecting and discriminating a number of compounds constituting a complex odor. However, they use only a few of these to recognize a substance. The focus of this research is to determine the compounds dogs learn to use in recognizing explosives used in land mines. This is accomplished by training dogs under behavioral laboratory conditions to respond differentially on separate levers to (1) blank air, (2) a target odor such as an explosive, and (3) all other odors (non-target odors). Vapor samples are generated by a serial dilution vapor generator whose operation and output is characterized by GC/MS. Once dogs learn this three-lever discrimination, testing sessions are conducted containing a number of probe trials in which vapor from constituent compounds is presented. Which lever the dogs respond to on these probe trials indicates whether they can smell the compound at all (blank lever) or whether it smells like the target odor (e.g., the explosive) or like something else. This method was conducted using TNT and C-4. The data show the dogs' reactions to each of the constituent compounds tested for each explosive. Analysis of these data reveal the canine detection odor signature for these explosives.

The Institute for Biological Detection Systems (IBDS) has developed a quantitative vapor delivery system that can aid in characterizing dog's sensitivity and ability to recognize odor signatures for explosives and contraband substances. Determining of the dog's odor signature for detection of explosives is important because it may aid in eliminating the risk of handling explosives and reducing cross-contamination. Progress is being made in the development of training aids that represent the headspace of the explosives. NESTT^TM TNT materials have been proposed as an approach to developing training aid simulates. In order for such aids to be effective they must mimic the headspace of the target material. This study evaluates the NESTT^TM TNT product with regard to this criterion. NESTT^TM TNT vapor was generated by the IBDS vapor delivery system, which incorporates a vapor generation cell that enables the user to control the conditions under which a substance is tested. The NESTT^TM TNT vapor was compared to the headspace of military-grade TNT. The findings identify and quantify major vapor constituents of military-grade TNT and NESTT^TM TNT. A comparative analysis evaluated the degree to which the NESTT^TM TNT mimics the headspace of an actual TNT sample.

The fate and transport of chemical signature molecules that emanate from buried landmines is strongly influenced by physical chemical properties and by environmental conditions of the specific chemical compounds. Published data have been evaluated as the input parameters that are used in the simulation of the fate and transport processes. A one- dimensional model developed for screening agricultural pesticides was modified and used to simulate the appearance of a surface flux above a buried landmine and estimate the subsurface total concentration. The physical chemical properties of TNT cause a majority of the mass released to the soil system to be bound to the solid phase soil particles. The majority of the transport occurs in the liquid phase with diffusion and evaporation driven advection of soil water as the primary mechanisms for the flux to the ground surface. The simulations provided herein should only be used for initial conceptual designs of chemical pre-concentration subsystems or complete detection systems. The physical processes modeled required necessary simplifying assumptions to allow for analytical solutions. Emerging numerical simulation tools will soon be available that should provide more realistic estimates that can be used to predict the success of landmine chemical detection surveys based on knowledge of the chemical and soil properties, and environmental conditions where the mines are buried. Additional measurements of the chemical properties in soils are also needed before a fully predictive approach can be confidently applied.

Nuclear Quadrupole Resonance (NQR) combines the compound specific detection capability offered by chemical detection techniques with the spatial localization capability and convenience of an induction coil metal detector. In the 16 years since NQR was last applied to mine detection in the U.S., there has been considerable improvement in the basic techniques. This paper reviews the progress achieved under a recent initiative to detect landmines by NQR. Two basic technical developments are summarized: the design of a detection coil suitable for probing the ground for landmines buried at typical depths, and an increase in the NQR signal obtained from the explosive TNT. In addition, we report the sensitivity of an NQR detection system to detect the electromagnetic response of metal-cased landmines.

We present an analysis of the use of thermal neutron analysis (TNA) to confirm the presence of energetic materials in unexploded ordnance detection. Our analysis is based on the performance of a prototype built by ANCORE and tested at Socorro, NM, and at Yuma, AZ, as part of the Environmental Security Technology Certification Program (ESTCP). From that data, we were able to develop a semi-empirical model for the expected signal strength as a function of the target's nitrogen content and depth. We found that the dependence on depth differs greatly between the two sites. We expect this simple model to be useful in future assessments of the feasibility of this approach. We also determine the P_d/P_FA performance of the system at the two sites and found it to correspond to a signal-to-noise ratio of order unity. We estimate that an increase in signal-to-noise of roughly three will be necessary to extent the applicability of this technology in unexploded ordnance detection. Such improvements may be possible if the NaI detectors currently employed are replaced with high purity germanium (HPGe) detectors.

Prompt gamma neutron activation analysis (PGNAA) has been proposed for confirming the presence of energetic materials as part of a mine or unexploded ordnance detection system. Ancore Corporation (previously SAIC Advanced Nucleonics Division), funded through Night Vision Electro Sciences Directorate by Environmental Security Test Certification Program, has carried out proof-of-concept demonstrations of PGNAA in this confirmatory role at Socorro, NM, and Yuma, AZ. In this, the first part of a two-part paper addressing the use of PGNAA in the detection of surface and near-surface UXO, we explore the phenomenology of PGNAA signals from surface or near-surface ordnance in soil to gain insight into the results of those demonstrations. PGNAA uses the high-energy gamma ray (10.8 MeV) from capture on N¹⁴ as a signature of the presence of nitrogen. This is one of the highest energy gamma rays resulting from neutron capture, and nitrogen is a major constituent of explosives, but a small portion of soil. Thus, PGNAA might be effective at confirming the presence of explosives. The phenomenology of dry soil is dominated by the two most common elements, oxygen and silicon. Neutrons injected into the soil elastically scatter from nuclei (predominantly oxygen), losing energy and propagating in a random walk fashion. Once slowed, neutron capture on soil elements produces a broad gamma-ray spectrum. Capture on Si²⁹ produces a 10.6 MeV gamma, which is not resolvable from the nitrogen signal of interest using scintillation detectors. Thus, PGNAA will need either good resolution detectors, or robust background subtraction to estimate the silicon contribution. For any system unable to resolve the Si²⁹ (10.6 MeV) and N¹⁴(10.8 MeV) gammas there is an inherently low signal to background, resulting primarily from the silicon in the soil. After background subtraction, there remains a challenging signal to noise level, where the noise is partly due to counting statistics and partly due to the modeling of the subtracted background.

To detect and locate buried landmines, the Canadian Department of National Defence (DND) is developing a teleoperated, vehicle-mounted, multisensor system called ILDP. In operation, a suite of 4 detectors scan ahead of the vehicle. Their outputs are combined through data fusion to indicate the possibility of a mine at a particular location, within a 30 cm radius. A thermal neutron activation (TNA) sensor, mounted behind the vehicle, is used to confirm the presence of explosives via detection of the 10.83 MeV gamma-ray associated with neutron capture on ¹⁴N. The TNA system developed for this uses a 100 microgram ²⁵²Cf neutron source surrounded by four 7.62 cm X 7.62 cm NaI(Tl) detectors. A combination of the use of state-of-the art radiation transport codes for design, judicious choice of specialized shielding materials and development of high-rate, fast pulse processing electronics has led to a system which can; (1) confirm the presence of all surface-laid or shallowly-buried anti-tank mines in a few seconds to a minute (depending on mass of explosive) (2) confirm the presence of anti-tank mines down to 20 cm depth in less than 5 minutes. (3) confirm the presence of large (greater than 100 g Nitrogen) anti-personnel mines in less than five minutes (4) operate in adverse climatic conditions. These results have been verified in field trials using the prototype sensor. Work is now ongoing to miniaturize the electronics, make the system robust and easy to use and investigate the use of an electronic neutron generator expected to enter service by the year 2000.

Locating landmines and UXO by detection of the chemical signature emanating from these devices is extremely challenging due to several of the physical properties of explosives. Because the explosives used in landmines and UXO have extremely low vapor pressures, the concentration of explosive vapors escaping from the ordnance is very low. Fate and transport studies of explosives in soil over buried ordnance have indicated that once released into the soil, virtually all of the explosive vapor is quickly adsorbed onto the surface of soil particles. This behavior is not surprising, since explosives are known to readily adsorb onto most types of surfaces. The adsorption of explosive onto soil particles is to an extent a reversible process, enabling diffusion of explosive through the soil. Unfortunately, because of irreversible adsorption and other processes occurring in the soil that destroy or degrade the explosive, the concentration of explosive reaching the surface of the ground is extremely low. However, dogs are able to locate buried ordnance, indicating that explosive signature compounds are present at or near the surface of the ground at concentrations in excess of the minimum detection limit of canines. Since the fate and transport studies indicate a much higher concentration of explosive adsorbed onto soil particles than in the vapor phase, sampling the explosive adsorbed onto soil particles may be a more efficient approach to sampling explosives than sampling explosive vapor in the air over buried ordnance. An electrostatic particle sampler has been designed which is capable of rapidly and efficiently sampling soil particles. Once the soil particles have been sampled, the explosive is desorbed from the particles, concentrated, and then presented to a sensitive chemical detector for analysis. In its present configuration, the particle sampler delivers a vapor phase sample to the detector, but the device could be adapted to deliver samples in the solvent phase as well. This makes the sampler compatible with a number of sensor technologies.

The U.S. Army has under development a number of systems to detect buried metallic and nonmetallic land mines. Almost all of these systems include a ground penetrating radar (GPR). These systems may be handheld or vehicle mounted and may be designed for close in or for standoff detection. A consensus has not been reached regarding many important system parameters. A discussion of the tradeoffs involving waveform, frequency, bandwidth, downlook angle, scanning methods, polarization, and spatial resolution will be presented. The usefulness of special techniques such as target resonances, synthetic aperture, and multiple polarizations will be discussed. The potential of GPR will be compared with competing sensors. A brief overview of a diverse set of GPR sensors will be presented.

Research progress on the design, construction, and operation of a novel, airborne ground penetrating radar system to detect surface and subsurface landmines is presented. The landmine detection system is unique in that active, electronic projectiles are shot into the ground from an airborne platform to create high power, monopulse radar signals. Intimate contact between the projectile and the ground reduces the amount of reflected radar energy at the air-soil interface and ensures that maximum radar energy is propagated into the surrounding ground. The end result is that the reflected radar signal is of higher energy and possesses a higher signal-to- noise ratio allowing enhanced detectability. The high power, monopulse signal that is reflected off the landmine is received at the airborne platform via scanned antenna array. In comparison, conventional ground penetrating radar systems typically use chirped or long pulse signals and horn type antennas located close to the ground limiting their usefulness in this application. To generate electrical energy, two types of projectiles are used and are based on the principle of magnetic flux compression or by the principle of piezoelectric compression. The performance results of these two projectile types as well as the models used to predict their behavior are presented and discussed. To evaluate the overall performance of the system, a sub-scale radar test range was also constructed. The radar test range consists of a large dirt- filled tank containing a high power impulse source, several targets that simulate buried landmines, and a post scanned antenna array located above the dirt-filled tank. The high power impulse source simulates the radar signal generated when the projectiles impact the ground. The radar cross-sectional data generated in the test range is presented and discussed.

This paper evaluates a non-intrusive buried object classifier developed for a ground penetration radar (GPR) system. The process uses a model based inversion algorithm to generate synthetic data sets which are correlated with real data sets. Recent work has introduced this technique to the community. Accomplishments and deficiencies with the procedure are discussed. Real data sets were collected with a commercially available GPR that is used to locate buried objects in a non- invasive manner. While synthetic data has been generated with a software implementation of a mathematical model developed for electromagnetic returns from a buried object. These real and synthetic measurements have been processed and compared using this technique to measure the similarities and the differences between the process data sets. The processed synthetic data images exhibited similar traits as present in the processed real data. Favorable visible correlation results were observed, yet the analytical comparisons were not conclusive due to lack of adequate data.

The University of Nebraska has developed a coherent random noise radar technique that permits phase-coherent processing of random noise radar signals. This is achieved through a heterodyne correlation receiver that retains the phase of the reflected signal during the detection process. A simulation study was carried out to evaluate the system's expected performance during detection of objects buried under lossy dispersive media with different types of complex permittivity characteristics as a function of frequency. It was observed that system performance was degraded under dispersive media conditions. However, by suitably tailoring the random noise transmission characteristics to attain an inverse frequency relationship with respect to the attenuation characteristics of the media, a significant enhancement was observed in detecting obscured objects. This shows that adaptive matched illumination is useful in detecting objects buried under lossy coatings intentionally induced to inhibit detection.

A noise radar is a radar that transmits band limited electromagnetic noise. The radar system cross correlates the received signals with a delayed version of the transmitted signal. The correlation value as a function of the delay time is proportional to the impulse response of the target. The long integration times and the use of coaxial cable as a delay line makes such a radar inherently appropriate for very short range (less than 3 meters) slow moving or stationary targets. This means that such a radar is particularly suited to the land mine detection role. We will discuss the development of such a radar at the Ohio State University, and we will demonstrate the system using data for buried underground targets, including the OSU surrogate mine field. Time domain and frequency domain characteristics of the targets will be discussed with application to buried target classification.

The Army Research Laboratory (ARL), as part of its mission- funded applied research program, has been evaluating the utility of a low-frequency, ultra-wideband (UWB) imaging radar to detect obscured targets such as vehicles concealed by foliage and objects buried underground. This paper concentrates on a specific area of great interest to the Army: the reliable detection of surface and buried mines. Measurement programs conducted at Yuma Proving Ground and elsewhere have yielded a significant and unique database of extremely wideband and (in many cases) fully polarimetric data. We will review recent findings from ARL's modeling, phenomenology and detection efforts. We also included a discussion of an end-to-end detection strategy that has been trained and tested against a significant data set. Performance assessments are included that detail detection rates versus false alarm levels.

A method of moments (MoM) analysis has been developed for the calculation of electromagnetic scattering from and the natural resonances of a dielectric body of revolution (BOR) embedded in a layered medium (the half-space problem constituting a special case). The layered-medium parameters can be lossy and dispersive, of interest for simulating soil. To make such an analysis tractable for ultra-wideband (short-pulse) applications and the calculation of natural resonances, we use the method of complex images to evaluate Sommerfeld-type integrals, characteristic of the dyadic Green's function of a layered medium. Example wideband scattering results as well as results for the fundamental natural resonant frequency are presented for a model plastic mine. Of special interest here is the influence of the background soil environment (e.g., water content, covering snow layer) on the ultra-wideband (UWB) scattering signature and the late-time resonances of plastic mines.

To detect and identify buried mines, the U.S. Army Research Laboratory (ARL) is using its ultra wideband (UWB) radar in a ground-penetrating mode. Operating in the frequency band from 50 to 1200 MHz, the radar is mounted on a mobile boom lift platform (BoomSAR). This enables it to form synthetic aperture radar (SAR) images as well as measure range profiles. As an integral part of the UWB radar project, ARL is developing an in-house modeling capability. In field tests at Yuma Proving Ground, Arizona, a variety of buried and surface targets were imaged with the BoomSAR, including a minefield of buried and surface metal mines. Most land mines of interest can be accurately modeled as bodies of revolution (BORs). Through consideration of the half-space Green's function, we realized that, there is no cross-polarized scattered field for such BOR targets, (theoretically) and, therefore, such targets are characterized only by co-planarized scattered fields. This feature, which, to our knowledge, has not been recognized before, has important implications for polarimetric SAR imaging of minefields, especially in regions with significant natural clutter (e.g., rocks) that are generally not BORs. This theoretical result will be verified using measured and computed data. Mine dimensions are on the order of one wavelength or less for the frequencies in our bandwidth. The modeling techniques we use for this range of wavelengths are method of moments (MOM) and finite-difference time-domain (FDTD). Model results from our buried BOR MOM code will be compared to the measured data.

The Army Research Laboratory (ARL) has several technology development programs that are evaluating the use of ultra- wideband synthetic aperture radar (UWB SAR) to detect and locate targets that are subsurface or concealed by foliage. Under these programs, a 1-GHz-bandwidth, low-frequency, fully polarimetric UWB SAR instrumentation system was developed to collect the data needed to support foliage and ground- penetrating radar studies. The radar was integrated onto a 150-ft-high mobile boomlift platform in 1995 and was thus named the BoomSAR. In 1997, under the sponsorship of the Strategic Environmental Research and Development Program (SERDP), ARL began a project focused on enhancing the detection and discrimination of unexploded ordnance (UXO). The program's technical approach is to collect high-quality, precision data to support phenomenological investigations of electromagnetic wave propagation through varying dielectric media, which in turn supports the development of algorithms for automatic target detection. For this project, a UXO test site was set up at the Steel Crater Test Area -- an existing test site that already contained subsurface mines, tactical vehicles, 55-gallon drums, storage containers, wires, pipes, and arms caches located at Yuma Proving Ground (YPG), Arizona. More than 600 additional pieces of inert UXO were added to the Steel Crater Test Area, including bombs (250, 500, 750, 1000, and 2000 lb), mortars (60 and 81 mm), artillery shells (105 and 155 mm), 2.75-in. rockets, submunitions (M42, BLU-63, M68, BLU-97, and M118), and mines (Gator, VS1.6, M12, PMN, and POM- Z). In the selection of UXO to be included at YPG, an emphasis was placed on the types of munitions that may be present at CONUS test and training ranges.

A numerical algorithm has been developed for the modeling of ultra-wideband (UWB) plane-wave scattering from a class of buried mines. In particular, the model assumes that a mine can be simulated as a body of revolution (BOR). The numerical results indicate that there are particular frequency subbands in which a given target is excited most strongly, with the subband depending strongly on the target type. Moreover, these optimal subbands are also dependent on the depression angle; therefore, the synthetic aperture radar (SAR) aperture must be limited (spatial-spectrum subbanding) so that the desired depression angle is achieved over the aperture used for imaging. These results indicate that the scattered response from particular buried mines can be highlighted by proper processing of the temporal and spatial spectrum. So motivated, we have applied subbanding to data measured by the Army Research Laboratory (ARL) UWB SAR, which collects fully polarimetric data over the 50 - 1200 MHz bandwidth. The measured data, when processed appropriately corroborate the theoretical expectations of where example anti-personnel and anti-tank mines scatter optimally. In particular, we consider the Valmara anti-personnel mine and the M20 anti-tank mine, using data collected at Yuma Proving Grounds. After validating the theory, we present frequency subbanding that will highlight one mine over another, a technique of potential application to the discrimination of targets from clutter.

DARPA sponsored a program during the fall of 1996 to collect backgrounds data (clutter and simulated targets) using several sensor systems. This paper focuses on the processing and interpretation of ground-penetrating (GPR) data collected using the Coleman ToMAS GPR system. High-resolution images generated using a backpropagation imaging (BPI) algorithm are presented for calibration lanes at two sites: one at Ft. A. P. Hill, VA, and the other at Ft. Carson, CO. Separate co- polarized and cross-polarized images are generated and compared. High-resolution cross-polarized images are also generated for an entire 100 meter X 100 meter 'center square' area at one site in 0.1 meter depth intervals. Comparison with Geonics EM-61 magnetometer as well as dig results are also presented.

Detection of mines buried in irregular stratified media is significantly complicated by the presence of signal clutter due to scattering from the rough interfaces of the layered structure. Full wave solutions have been derived for the electromagnetic fields scattered by the two rough surfaces in a realistic physical model of the three media environment of the mines. They account for five different scattering mechanisms that the waves undergo, assuming that both the transmitter and receiver are above the uppermost interface of the irregular media. Two scattering mechanisms are associated with reflection from above and below the upper interface and two are associated with transmission across the upper interface, the fifth is associated with reflection from above the lower interface. In view of the fact that in general the two rough interfaces are characterized by independent random rough surface heights (except where the thickness of the intermediate medium vanishes) the rough surface height joint probability density functions are characterized by a family of probability density functions associated with the gamma functions. Multiple bounces between the two interfaces are accounted for in the analysis. The elements of the incoherent Mueller matrix (that relates the scattered fields to incident Stokes vectors) are obtained from the expressions for the scattered fields. From the simulated data it is possible to determine the optimal polarizations and the incident and scatter angles of the waves as well as the wavelength, for purposes of suppressing the impact of the clutter on the detection of buried objects.

The scattered field from a ramp driving function traces out the transverse cross sectional area profile of a metallic target as a function of time or depth of the target. This concept has been used in the past to generate 'low frequency' images of simple conducting shapes. This concept is currently being examined for dielectric scatterers such as land mines. It has been applied to theoretically computed scattered fields of dielectric spheres and to experimentally measured scattered fields from dielectric targets in free space. There are some advantages to applying this technique to identify land mines. First, the ramp response yields information about target size, orientation and geometrical shape. Second, the early time component of the scattered fields used in this imaging concept is relatively strong as contrasted to the late time response used for resonant target identification. Third, it is a low frequency concept since the ramp response frequency spectrum is proportional to the impulse response spectrum weighted by the inverse of the frequency squared. This is significant since the attenuation in most earth at a given depth increases as frequency increases for the frequency bandwidth used to detect land mines. However, for dielectric targets, the critical angle and internal scattering mechanisms can change the ramp response from their transverse cross sectional areas versus distance. Physical Optics solutions for the ramp response of dielectric spheres have been obtained and have been compared with their corresponding eigen function solutions. For a dielectric target, a way to improve the ramp response in the presence of the internal specular scattering mechanisms has been discussed. Also the ramp response signatures of some dielectric targets including a land mine kept in free space have been extracted from measured data for broadside and endfire excitations.

Waveguide studies of the effectiveness of soil modification techniques for non-metallic mine detection with ground penetrating radar are described. Visibility improvements for a nylon target buried in sand are considered as varying amounts of water or liquid nitrogen are added to the sand. Experiments are performed in an S-band waveguide (2.6 - 3.8 GHz), and results show that increased water content improves shallow buried target visibility initially due to increased dielectric contrast between the target and background medium, but eventually obscures target responses due to increased loss. Initial tests of the addition of liquid nitrogen show reduced loss in the high moisture content case, so that targets can again be made visible. Analytical model results are also presented for the experimental configuration and found to be in good agreement with measured data, and further studies of soil modification effects with the analytical model are performed. A finite difference time domain (FDTD) electromagnetic model for more complicated geometries involving general target shapes and inhomogeneous water contents is also described.

The present paper is a part of a project with the goal of developing a hand-held system for detection of buried landmines. The system is based on an impulse radar operating in the frequency band 0.3 - 3 GHz. Previously a crossed dipole antenna has been used for transmit and receive. This antenna has now been complemented by another crossed dipole antenna, rotated 45 degrees relative the first one. The two antenna pairs are not active simultaneously, but transmits/receives alternately. The antenna pattern of a crossed dipole antenna resembles a 'four leaf clover,' and one advantage with using two rotated antenna pairs is that the dead angles in the antenna pattern will be covered by the second antenna. Another, perhaps more important, advantage is that the detection capability of the system can be improved by using the information in both antenna channels. For example, if the antenna is positioned above a relatively smooth surface with no target present, then the returned signals in the two channels will be almost equal. If, however, a target is present the signals will be different when we switch between the channels. The second channel also provides an independent realization of the background clutter, and by fusing the two channels the clutter level can be decreased considerably. This paper will present different detection algorithms that exploit this information added by the second antenna channel. The algorithms are evaluated on real data from the impulse radar system.

Rough surface clutter is a significant source of interference in non-specular ground penetrating radar (GPR) applications that needs to be suppressed to maintain high performance in the signal processing. Our research is in the directions of (1) development and testing of flexible parametric models for the statistical distribution of clutter that rely on the theories of alpha-stable random processes, (2) establishment of bounds on the performance of signal processing algorithms, and (3) design and analysis of robust, non-Gaussian signal processing algorithms based on the statistical clutter models. Synthetic data simulated with FDTD techniques are extensively used.

Ground-penetrating radars (GPRs) are one technology being investigated for the detection of buried landmines. Full experimental simulations for potential GPRs are complicated, time consuming, and expensive. An alternative is to perform simulations using computational electromagnetic techniques, such as the finite-difference time-domain method (FDTD). Until recently, such simulations were not possible because of the limitations on computer memory and speed. Results are presented from two FDTD simulations for GPRs designed for mine detection. One of the GPRs uses continuous-wave signals (the 'separated aperture' or 'waveguide below cutoff' detector for nonmetallic mines), and the other uses baseband pulses (a proposed detector for small antipersonnel mines). These simulations are fully three-dimensional and include all of the details of the detector (antennas, reflectors, feeds, etc.), lossy earth, and mine. Results from the simulations are compared with experimental measurements made with model detectors.

Simulation code was developed to model monostatic or bistatic radar returns from terrain and discrete objects within the radar's field of view, including subsurface scattering due to complex permittivity inhomogeneities and buried objects, using Born approximations. This code is applied to simulation of return from a mine-like dielectric cylinder buried a few inches below the surface. Clutter sources included are: scattering from the rough surface above the mine and subsurface random permittivity inhomogeneities. Simulated images of received power from a subsurface region are obtained with and without the mine and the results will be compared with experimentally-obtained images of the region. Depth resolution of a few inches is obtained by using a focused linear array of 4 transmitters and a linear array of 4 receivers identical to and parallel to the transmitter array. The illuminated subsurface volume is between the two arrays. The detector scans the volume by varying relative delays between array elements such that, at a given time instant t₁ the signals arrive at a point p₁ and returns from p₁ arrive at the 4 receivers with equal delays. Superposition of received signals at time t₁ favors returns from a small volume around p₁. At time t₂ all energy is similarly focussed on a different point p₂. The process continues until the entire volume has been scanned and an image of the region has been generated. Sensitivity to mine dimensions, composition and burial depth and soil parameters is demonstrated.

The effects of soil and mine parameters on microwave imaging are investigated based on Finite-Difference Time-Domain (FD- TD) simulation and laboratory studies. Range of soil permittivity considered is from 2.6 to 7.6 corresponding to dry soil and a volumetric soil moisture of 16%. Both plastic and metallic mines are imaged. The diameter of the mines considered ranges from 9 cm to 32 cm and their height ranges from 4 cm to 20 cm. It is found that when a mine diameter is larger than a wavelength, its image is discernible and becomes clearer when its diameter exceeds two wavelength. It is also found that a plastic mine with a dielectric constant of 3.9 embedded in a dry soil medium with a dielectric constant of 2.6 can generate an image with the correct geometric shape.

A method of neutralizing landmines in which the integrity of the surrounding terrain is retained is herein described. High pressure waterjets which can be used to detect the presence of landmines can then be used to remove the soil and other cover in a plane immediately adjacent to and around the mine so that the side of the mine can be visually inspected through a remote television camera. At that time the flow of water is channeled through a line in which small particles of sand are added to the waterjet which is at a pressure of between 3,000 and 10,000 psi depending on the device which is used. Jet flow rates are on the order of 5 gpm depending on the nozzle configuration used. By bringing this abrasive stream in along a lateral plane through the mine it is possible to intersect, and neutralize, the fusing systems most likely to be used to initiate the charge, in a single pass. At higher flow rates, as the cut is made the jet will generate significant turbulence in the mine body, sufficient to remove a considerable quantity of the explosive which is resident within the mine at the same time as the mine is being dissected. The precision of cut achievable is shown by the longitudinal cutting into two parts of live detonators, as well as representative mine bodies.

The impact of a waterjet stream emits an acoustic signal when the jet strikes a buried target object. The structure of the sound emitted depends on the energy in the jet, the target which is struck, and the surrounding media. A high pressure (less than 5000 psi) waterjet of diameter 0.01 inches can penetrate into the ground to a depth of greater than 8 inches in less than 0.02 seconds and will generate a distinct acoustic signal from any obstruction it encounters as it makes that hole. By analyzing the sounds generated from a series of jets, located at 2 - 3 inch intervals over the path width of a remotely controlled detection unit, it is possible to identify the location of suspicious objects ahead of a vehicle. Flow rates for such a system, which can cover a path width of 36 inches at walking speeds, are anticipated to be around 5 gallons per minute. We have made initial measurements on acoustic signals generated by waterjet impact and found that the response from buried land mines appears to be Gaussian, thus allowing second order statistics to characterize the signals. We have shown that under laboratory conditions, the power spectrum can be used to discriminate land mines from other underground objects such as rocks and metal pipes.

We have investigated the use of acoustic energy produced by a pulsed CO₂ laser to detect objects underwater or buried in sand. The CO₂ laser produced 150 mJ pulses of duration 100 ns. The resulting acoustic pulses were detected with an audio microphone with a response to 15 kHz or a PZT transducer with a resonant frequency at 28 kHz. With the laser incident on the surface of a water-filled tank, acoustic echoes were observed from the tank walls and from objects in the tank. For objects buried in sand, changes in the acoustic lineshape related to the presence of subsurface objects were observed. Analysis of the data to extract clear signatures of the mine are in progress.

The feasibility of acoustic resonance for detection of plastic mines was investigated by researchers at the Oak Ridge National Laboratory's Instrumentation and Controls Division under an internally funded program. The data reported in this paper suggest that acoustic resonance is not a practical method for mine detection. Representative small plastic anti- personnel mines were tested, and were found to not exhibit detectable acoustic resonances. Also, non-metal objects known to have strong acoustic resonances were tested with a variety of excitation techniques, and no practical non-contact method of exciting a consistently detectable resonance in a buried object was discovered. Some of the experimental data developed in this work may be useful to other researchers seeking a method to detect buried plastic mines. A number of excitation methods and their pitfalls are discussed. Excitation methods that were investigated include swept acoustic, chopped acoustic, wavelet acoustic, and mechanical shaking. Under very contrived conditions, a weak response that could be attributed to acoustic resonance was observed, but it does not appear to be practical as a mine detection feature. Transfer properties of soil were investigated. Impulse responses of several representative plastic mines were investigated. Acoustic leakage coupling, and its implications as a disruptive mechanism were investigated.

In this work we examine the feasibility of active thermal sensing of buried objects. A 1.5 kW carbon dioxide laser is used to provide a thermal impulse to the surface of a sand test bed containing simulated metallic landmines. Time- dependent thermal images of the induced surface temperature differentials are obtained using an infrared focal plane array imaging system. Experimental results are reported for two target sizes and four thermal pulse conditions. Quantitative evaluation of the induced surface temperature differentials as a function of time are presented.

Numerous active landmines buried around the world have prompted work on various technologies for locating these mines. One promising technique directs a beam of x-rays into the ground, and detects the fraction scattered back. An image of the detected photons reveals the subsurface content. In this experiment, the effect of photon cross-talk between adjacent x-ray beam/photon detector systems was investigated. If feasible, multiple beam/detector systems would allow a single landmine detection system to survey the ground much faster. The results of the examination of the segmented detector system showed that this system is quite capable of producing very recognizable images of surface buried landmines, in spite of significant limitations imposed by the required setup of this particular experiment. Therefore, the segmented detector system is an option that should be strongly investigated in the development of a landmine detector system if there is a critical emphasis on speed.

Lateral migration radiography (LMR) employs scattered photons to get detailed images of covered objects. Images of real mines buried in soil using LMR have shown dramatic differences compared to images generated using simulated mines. The major characteristic that allows for the discernibility of land mines to the degree of actual type identification is the presence of voids (air volumes) required for the operation of the fuse assembly or for blast direction control. Air volumes greatly modify the detected field of both once and multiple- scattered photons. The LMR system consists of two uncollimated detectors positioned to detect once-scattered photons and two collimated detectors designed to detect primarily multiple- scattered photons. Air volumes modify both exit paths and the position of first-scatter events; they also modify the migration paths of multiple-scattered photons, thus producing different images in the two detector types. The burial mode (below surface or laid on the surface) of the land mine can also be discerned by LMR due to a shadowing effect seen for surfaced-laid land mines. The presence of even a minute amount of metal in the land mine also aids in discerning the mine, because metal produces a signal decrease in both types of detectors. Monte Carlo calculations have been performed with the MCNP code to obtain an understanding of the details of the photon lateral migration process. Images generated from these Monte Carlo calculations are in agreement with the experimental measurements. The real mine images confirm that LMR is capable not only of mine detection, but also of mine identification.

The implementation of a backscattered x-ray landmine detection system has been demonstrated in laboratories at both Sandia National Laboratories (SNL) and the University of Florida (UF). The next step was to evaluate the modality by assembling a system for field work. To assess the system's response to a variety of objects, buried plastic and metal antitank landmines, surface plastic antipersonnel landmines, and surface metal fragments were used as targets. The location of the test site was an unprepared field at SNL. The x-ray machine used for the outside landmine detection system was a Philips industrial x-ray machine, model MCN 225, which was operated at 150 kV and 5 mA and collimated to create a 2 cm diameter x-ray spot on the soil. The detectors used were two BICRON plastic scintillation detectors: one collimated (30 cm X 30 cm active area) to respond primarily to photons that have undergone multiple collision and the other uncollimated (30 cm X 7.6 cm active area) to respond primarily to photons that have had only one collision. To provide motion, the system was mounted on a gantry and rastered side-to-side using a computer-controlled stepper motor with a come-along providing the forward movement. Data generated from the detector responses were then analyzed to provide the images and locations of landmines. Changing from the lab environment to the field did not decrease the system's ability to detect buried or obscured landmines. The addition of rain, blowing dust, rocky soil and native plant-life did not lower the system's resolution or contrast for the plastic or the metal landmines.

The cost, and the closely related length of time, spent in searching for mines or unexploded ordnance (UXO) may well be largely determined by the number of false alarms. False alarms result in time consuming digging of soil or in additional multisensory tests in the minefield. In this paper we consider two area based methods for reducing false alarms. These are (1) the previously known 'declaration' technique, and (2) the new (delta) -Technique which we introduce. We first derive expressions and lower bounds for false alarm probabilities as a function of declaration area, and discuss their impact on receiver operation characteristic (ROC) curves. Secondly we exploit characteristics of the statistical distribution of sensory energy in the immediate neighborhood of targets and of false alarms from available calibrated data, to propose the (delta) -Technique which significantly improves discrimination between targets and false alarms. The results are abundantly illustrated with statistical data and ROC curves using Electromagnetic Induction Sensor data made available through DARPA from measurements at various calibrated sites.

This paper presents a nonparametric classification procedure based on a generalization of classical rank-based statistics and a preliminary investigation of the method's applicability to mine detection. The classifier is particularly relevant to high-dimensional applications and can improve performance characteristics such as discriminatory power. The common practice of considering ranked interpoint distances is generalized to point-to-subset distances and a recurrence for the exact distribution of this generalized Wilcoxon-Mann- Whitney (GWMW) test statistic is available. From a classification standpoint, GWMW represents a class of generalized weighted (k,l)-nearest neighbor rules. The GWMW classifier is applied to multispectral minefield data collected under the The Coastal Battlefield Reconnaissance and Analysis (COBRA) Program. A truthed detection map obtained from the multispectral image set and provided by NSWC Coastal Systems Station, Panama City, Florida, contains both true mines and false positives. The GWMW classifier is compared to classical classification methods on this data via cross- validation. The preliminary experimental results indicate that GWMW yields a significant improvement in discriminatory power for this important practical application.

We present an efficient algorithm for computation of the maximum likelihood estimate of the location of a known target from short pulse scatter data measured in a suite of tomographic experiments. The algorithm consists of a three step procedure: (1) data filtering, (2) dime-domain backpropagation, and (3) coherent summation and employs of a number of forward and inverse Radon transforms integrated in a tomographic scheme. A computer simulation is included for illustration purposes.

Based on remote sensing of a potential minefield, point locations are identified, some of which may not be mines. The mines and mine-like objects are to be distinguished based on their point patterns, although it must be emphasized that all we see is the superposition of their locations. In this paper, we construct a hierarchical spatial point-process model that accounts for the different patterns of mines and mine-like objects and uses posterior analysis to distinguish between them. Our Bayesian approach is applied to COBRA image data obtained from the NSWC Coastal Systems Station, Dahlgren Division, Panama City, Florida.

This paper presents an extension of the research work on the wavelet-based classification scheme developed to discriminate underwater mine-like from non-mine-like objects using the acoustic backscattered signals. Based on the single-aspect classification results, the robustness and discriminatory power of the selected features, and the generalization ability of the trained network are demonstrated on several cases. To further improve the overall classification accuracy, the classification results of multiple aspect angles are fused together. Two different fusion approaches are considered and their performance is tested on ten different realizations. The final results show excellent classification accuracy of 96% for only a 4% false alarm rate.

Typically, a human agent or a robotic device may sweep a suspected minefield in a systematic up and down pattern to search for explosive mines with the help of an appropriate sensor or sensor system, such as an EMI (Electromagnetic Induction) sensor. In this paper we consider alternative search patterns which take advantage of a priori knowledge of the minefield. In previous work, a gradient based search algorithm has been designed and shown to be an effective search strategy using simulations on hypothetical minefield data. This paper considers a suite of fast search heuristics based on a hierarchical two level approach, and evaluates these algorithms with the realistic sensory data, specifically the Electromagnetic Sensory Data from DARPA. Heuristics considered include a hierarchical version of our gradient based algorithm, a nearest neighbor type greedy heuristic, and a heuristic which is inspired from an approximate solution of the traveling salesman problem.

Detecting minefield point patterns is an important problem for the Navy, Marines and Army. Because of the difficulty and uncertainty associated with accurately modeling enemy mine laying procedures, robust and flexible family of statistics are needed to detect minefields as deviations from complete spatial randomness. In this paper, a large family of minefield detection statistics are presented and compared using their asymptotic relative efficiency for testing multinomial and minefield mixture alternatives. A slightly modified version of the widely-used power-divergence statistics are introduced that are appropriate under sparseness assumptions. This family includes the empty boxes test which has been advocated previously as a simple and effective approach. Another family, called VC statistics, is presented that provides a low- complexity statistic with optimal performance. The efficiency of these methods are compared analytically and with a minefield benchmark used in previous work.

In this paper we model time-domain plane-wave scattering from targets buried under a rough (random) air-ground interface. The properties of the interface are parametrized as a random process with known statistics. Since the fields incident upon a buried target first penetrate the rough interface, they are random processes as well, and so are the scattered fields. An optimal detector is built based on this model, which takes into account both the clutter and target-signature statistics (the former due to scattering at the rough surface, and the latter due to transmission); the statistics of these two processes are in general different. Detector performance is compared to that of a matched filter, which assumes the target signature is known exactly (i.e., is non-random). The results presented here, as a function of angle and polarization, show that, when the target signature is properly treated as a random process, a gain in detector performance can be obtained. Also, we explain under which conditions this improvement is expected to be significant.

Automatic mine detection is a critical issue in battle field management. This is expected to lead to better technologies that provide accurate and reliable detection of mines embedded in clutter. In this paper, we review a procedure for automatic mine detection in multispectral data provided by the Coastal Battlefield Reconnaissance and Analysis (COBRA) program. Our procedure is essentially a two-step method that employs the Maximum Noise Fraction (MNF) transform, a powerful enhancement tool for multispectral data, combined with nonlinear morphological operators that do the actual detection. Mathematical morphology is also used to account for the critical step of clutter estimation required by the MNF transform. Results obtained with available, truthed data, show the high success of the proposed method in meeting performance requirements. A low number of midsections is observed, whereas only a small number of false alarms is introduced by the algorithm. The results are better than the ones obtained by means of a constant false alarm rate (CFAR) algorithm provided along with the data.

Most technologies in use or proposed for use to detect landmines and unexploded ordnance (UXO) suffer from unacceptably high false-alarm rates, even at modest probabilities of detection. High false-alarm rates are a consequence of the inability to discriminate real UXO and landmines from man-made and naturally occurring clutter. The goal of the Defense Advanced Research Project Agency (DARPA)- sponsored Background Clutter Data Collection Experiment is to provide data which will support the development of techniques that are more adept at discriminating UXO from benign, man- made objects. During the fall of 1996, high areal density site surveys were completed using the following sensor types: magnetometer, infrared, electromagnetic induction, and ground- penetrating radar. Preliminary analysis of the data confirmed that a large number of anomalies in the sensor data are visually indistinguishable from anomalies that are a result of emplaced inert UXO or landmines. The Firing Point 20 site at Fort A. P. Hill exhibits the largest number of these ordnance- like anomalies. To determine the source of a subset of these sensor response anomalies, a 1-week excavation effort was conducted. This paper presents an analysis of the data to determine the candidate locations for, the procedures used during, and the results of the excavation.

Detection of mines is an important problem to the Army. A stand-off detection radar has been developed to detect mines in real time mode at stand-off distances ranging from 5 to 30 m. This active radar consists of three horn antennas, one transmitter and two receivers, carried by a moving vehicle. The transmit horn generates 36 sinusoidal carrier signals which have frequencies ranging from 0.5 to 4 GHz, in 100 MHz steps. At each frequency, the carrier signals are modulated by a train of 16 pulses having a 10 ns pulse width and a repetition rate of 3.9 MHz. These signals resonate targets including non-metallic and metallic mines and other objects. The mines and objects' return signals will be detected by the two receive antennas. The detection algorithm is used to identify an anomaly if it exists above background. Then, the discrimination algorithm is used to distinguish between mines and other objects. Finally, the location algorithm and the differential global positioning satellite system are used to mark the position of the detected mines. Test results from last year showed that some of the mines were not detected and the positions of some of the detected mines were not marked precisely enough. This resulted in missed detections and in increasing false alarm rate. Therefore, we propose to investigate Kalman filtering to improve the performance of the radar system. A preliminary study on the use of Kalman filtering algorithm for a single target tracking is provided in this paper. We believe that accurate tracking will result in accurate locating the detected target. Kalman filtering algorithm is selected for target tracking because we consider the fact that the source as well as the receivers are moving relative to the mine. This causes the delay to be also varying with time. Therefore, a dynamic algorithm such as Kalman filtering is a good technique to track the target by estimating the variable time delay.

Early time returns affect the estimation of complex natural resonance (CNR) frequencies associated with a target. This is especially true when there is a small separation between the early time returns and the late time response of the target and the CNRs are low Q mechanisms. A good example of this scenario is antipersonnel mines. In this situation, it helps to remove the early time returns from the total scattered field. A new technique to accomplish this is presented here. Using some numerical data and some experimental data, it is demonstrated that this technique is very effective in removing the early time returns. The modified scattered field data then yields better estimates of CNR frequencies.

In our previous work, we have shown that a model-based Bayesian approach to the detection of mines affords significant performance gains for both time-domain and frequency-domain electromagnetic induction (EMI) sensor over standard thresholding techniques. Our methodology merges physical models of the evoked target response with a probabilistic description of the clutter, and provides both an optimal detection algorithm, under a specific set of assumptions, and performance evaluation measures in the form of probability of detection (P_d) and probability of false alarm (P_fa). This approach was validated on data obtained during the DARPA Backgrounds Clutter Data Collection Experiment. In this paper, we review these results, and also present theoretical results that show that utilization of the entire multi-channel time-domain or frequency-domain waveform always affords an improvement in detection performance over single-channel systems. In addition, we present results that indicate that including spatial information into the processor substantially reduces the false alarm rates over processors in which this spatial information is not included. Finally, we discuss the performance of a more advanced detector, also formulated using signal detection theory, in which uncertainty in the placement of the mine in the environment is incorporated into the detector.

We present a statistically-based method for the enhancement and detection of mines and mine-like targets, in multi-channel imagery. Standard approaches to such multi-channel image processing take advantage of the correlation across channels within a pixel, but typically do not exploit the spatial dependency between pixels. This work aims to construct appropriate spatial statistical models for multi-channel mine imagery and apply these models to allow both image enhancement as well as direct and improved detection of anomalies (i.e., targets) in such data. We base the method on a Markov Random Field (MRF) model that incorporates a priori information about both the target's and the background's spatial characteristics. In particular, we find a Maximum A Posterior (MAP) detector of mine targets in background under the prior assumption target pixels are locally spatially dependent. We implement our algorithm on polarimetric and thermal data obtained from the Remote Minefield Detection System (REMIDS), with favorable results compared to a Maximum Likelihood (ML) detector that performs detections on a pixel-by-pixel basis, i.e. without spatial correlation.

Clutter is the largest factor contributing to the poor detection rates and high false-alarm rates for mine and unexploded ordnance (UXO) detection systems. The source of this clutter can be either naturally occurring or anthropic. Because the standard detector technologies are anomaly-based systems, few features within the sensor data permit mitigation of false alarms or provide an avenue to enhance detection rates. To achieve operational detection performance, a better understanding of clutter statistics is required at the single pixel level and at the feature level. This paper presents an in-depth assessment of the statistical properties of clutter and target signatures for a specific test site. This assessment uses data collected during the Defense Advanced Research Projects Agency (DARPA) Background Clutter Data Collection Experiment. Pixel-level statistics for electromagnetic induction detection systems are discussed. The resulting statistical distribution functions for clutter and targets exhibit poor separation. Improved separation of the distribution functions is achieved if features are employed. For example, by measuring the particular size and shape features of target signatures, the false-alarm rate can be reduced with minimal decrease in the detection rate. By using feature-level information, improved system performance can be achieved. This improved performance is dependent on the feature-level statistics of a specific site and is always limited by the overlap between the distribution functions of the clutter and target signatures. The resulting performance enhancement -- although significant -- is still far below the level required for very high detection rates and low false- alarm rates.

Current ground penetrating radars (GPR) have been tested for land mine detection, but they have generally been costly and have poor performance. Comprehensive modeling and experimentation must be done to predict the electromagnetic (EM) signatures of mines to access the effect of clutter on the EM signature of the mine, and to understand the merit and limitations of using radar for various mine detection scenarios. This modeling can provide a basis for advanced radar design and detection techniques leading to superior performance. Lawrence Livermore National Laboratory (LLNL) has developed a radar technology that when combined with comprehensive modeling and detection methodologies could be the basis of an advanced mine detection system. Micropower Impulse Radar (MIR) technology exhibits a combination of properties, including wideband operation, extremely low power consumption, extremely small size and low cost, array configurability, and noise encoded pulse generation. LLNL is in the process of developing an 'optimal' processing algorithm to use with the MIR sensor. In this paper, we use classical numerical models to obtain the signature of mine-like targets and examine the effect of surface roughness on the reconstructed signals. These results are then qualitatively compared to experimental data.

The GEO-CENTERS, INC. transverse electromagnetic rhombus (TEMR) antenna is a complex, broad-band antenna used in a ground-penetrating radar (GPR) system. The TEMR antenna consists of different types of material, has much fine detail and curvature in each of three dimensions. We have used FDTD to analyze the performance of this antenna in both the transmit and receive mode. A three-dimensional analysis was required to evaluate performance for the launch and incidence angles that were imposed by the GPR application. In this paper, a transmitting-receiving TEMR antenna pair in free space is simulated with FDTD. The transmitting antenna is excited with the waveform generated by the pulser electronics. The near zone fields are propagated into the far zone fields by transformation. Then, the far zone fields are used as incident fields for the receiving antenna. Voltage waveforms at the outputs of the simulated antenna compare well with the experimental results.

In recent years there has been considerable interest in the use of ground penetrating radar (GPR) for the non-invasive detection and localization of buried objects. In a previous work, we have considered the use of high resolution array processing methods for solving these problems for measurement geometries in which an array of electromagnetic receivers observes the fields scattered by the subsurface targets in response to a plane wave illumination. Our approach uses the MUSIC algorithm in a matched field processing (MFP) scheme to determine both the range and the bearing of the objects. In this paper we derive the Cramer-Rao bounds (CRB) for this MUSIC-based approach analytically. Analysis of the theoretical CRB has shown that there exists an optimum inter-element spacing of array elements for which the CRB is minimum. Furthermore, the optimum inter-element spacing minimizing CRB is smaller than the conventional half wavelength criterion. The theoretical bounds are then verified for two estimators using Monte-Carlo simulations. The first estimator is the MUSIC-based MFP and the second one is the maximum likelihood based MFP. The two approaches differ in the cost functions they optimize. We observe that Monte-Carlo simulated error variances always lie above the values established by CRB. Finally, we evaluate the performance of our MUSIC-based algorithm in the presence of model mismatches. Since the detection algorithm strongly depends on the model used, we have tested the performance of the algorithm when the object radius used in the model is different from the true radius. This analysis reveals that the algorithm is still capable of localizing the objects with a bias depending on the degree of mismatch.

It has been shown that, for non-conductive backgrounds and non-conductive spatial heterogenities, two- and three- dimensional images of spatial variations in wave speed can be reconstructed from broadband ground penetrating radar (GPR) measurements. In diffraction tomography, the reconstructed image is the so-called object function which, for non- conducting heterogeneities, is one minus the square of the real refractive index. In cases where spatial variations in electrical conductivity exist, the object function is complex with real part related to relative changes in the real refractive index and the imaginary part representing the spatial variations in conductivity. Thus, by considering the complex object function, it is possible to reconstruct images of both wave speed and conductivity. The procedure presented here is used to reconstruct images of wave speed and conductivity for several buried targets.

The Improved Landmine Detector Project (ILDP) was initiated in Autumn 1994 to develop a prototype teleoperated vehicle mounted mine detector for low metal content and nonmetallic mines to meet the Canadian requirements for rear area mine clearance in combat situations and peacekeeping on roads and tracks. The relatively relaxed requirements, such as low speed and reduced detectability of completely nonmetallic mines, greatly increase the likelihood of success. The ILDP system consists of a unique teleoperated vehicle carrying a forward looking infrared imager, a 3 m wide down-looking highly sensitive electromagnetic induction detector and a 3 m wide down-looking ground probing radar, which all scan the ground in front of the vehicle. Scanning sensor information is combined using a suite of navigation sensors and custom designed navigation, spatial correspondence and data fusion algorithms. Suspect targets are then confirmed by a thermal neutron analysis detector. A key element to the success of the system is the combination of sensor information. This requires coordinated communication between the sensors and navigation system and well designed sensor co-registration, spatial correspondence and data fusion methodologies. These complex tasks are discussed in detail. The advanced development model was completed in October 1997 and testing and improvements are ongoing. Results of system performance during extensive field trials are presented. A follow-on project has been initiated to build four to six production units for the Canadian Forces by the year 2000.

The complexity of the detection and discrimination problem for land mines requires that multiple sensors and algorithms be employed. We have been attempting to develop a range of new sensors and sensor combinations that will contribute to the solution to this problem. The sensors discussed in this paper include heated waterjets combined with infrared images, non- contact acoustic focusing, and the detection of electromagnetic emissions from smart land mines.

No single sensor modality will solve the problem of detecting small, buried objects in soil in the presence of typical clutter. Techniques involving fusion of data from multiple sensors can improve detection statistics. At the next level of complexity is multi-modal sensing in which an excitation with one modality is detected by another. Here we consider one such example, microwave heating of the soil followed by infrared imaging. The technique promises to produce signatures of buried objects which have contrast with respect to the surrounding soil in (1) absorption at the microwave frequency resulting in a change in the amount of energy absorbed, (2) dielectric constant resulting in alteration of the field distribution in the soil, or (3) thermal properties resulting in changes in the heat distribution. Indirect detection may be possible, through changes in the microwave or thermal properties of the soil caused by disturbance during placement of the object, or caused by changes in the soil properties resulting from alterations in water content caused by the object. We discuss wavelength selection, expected sensitivity, and techniques for enhancement of the signal, as well as overall system requirements, and will show some preliminary results.

A new hybrid remote-sensing method using active high-power microwave (HPM) illumination and passive infrared (IR) detection is presented for the detection of shallow buried landmines. A 2.45 GHz, 5 kW microwave source was used for illumination. The thermal signature of the mine at the soil surface was detected in the 8 - 12 micrometer region both in near real-time as well as after a brief time-delay following illumination. The thermal signature at the soil surface is primarily made up of two components. A thermal signature occurs at the soil surface in near real-time due to the interference of the incident beam and the beam reflected by the mine. A second thermal signature is generated when the variations in heating due to differential microwave absorption by the mine and the surrounding soil is conducted upwards from the mine location to the surface. Both signatures are dependent on the complex dielectric constants of the mine and the soil. Results will be presented from laboratory experiments and field trials with different types of metallic and non-metallic mine surrogates, dummy mines without explosives and live mines with explosives but without fuses.

A sensor-fused system is being developed for detection of buried land mines. The system uses a ground-penetrating radar, an infrared camera, and an electromagnetic induction sensor. In the current implementation each sensor is used independently, and fusion is performed during post-processing. We briefly describe the sensors and a data collection involving buried mine surrogates. Algorithms for preprocessing and feature extraction are reviewed. To deal with non- coincident sampling we have developed a new feature-level fusion algorithm, which does not require detection and subsequent association of putative targets. Results are presented for fusion of simulated data.

Increasingly, commercial off the shelf (COTS) products are finding direct application to DOD needs providing immediate, state-of-the-art technology solutions at a fraction of development contract costs. The terabyte/day data processing requirements of petroleum exploration has led to the development of very fast N-dimensional voxel (3D pixel) rendering software engines. Seismic processing methods also can be directly applied to ground penetrating radar (GPR) data using inverse fft on raw data to produce time verses amplitude and data quality control using signal, noise, waveform, wavefield tools to analyze signal and noise characteristics. Applying 3D seismic reconstruction methods to the 2D ground penetrating radar transects, volumetric (voxel based) models of entire sites (plus or minus 10,000 m²) can be examined quickly. Fused data sets can be simultaneously displayed in voxel space using cross sections, bench cuts, color pallets and variable transparencies to emphasize targets. Applied to the DARPA background clutter data collection experiment, independent data sets simultaneously displayed (fused) in voxel space allow evaluation of target detection and position accuracy of each sensor system (GPR, magnetic/EM, IR). Voxel false color renderings of the fused data can then be used to visually highlight targets co-detection by different sensor systems, helping to eliminate false positives and select the most reliable detection methods. False color voxel renderings from data fusion could lead to rapid unambiguous landmine threat identification for the warfighter.

Fuzzy Clustering is applied to the problem of detecting landmines in Ground Penetrating Radar (GPR). The DARPA Backgrounds Data provides a rich source of signatures derived from a cluttered environment with a variety of sensors. One sensor used in the Backgrounds collection was the GPR developed and fielded by Geo-Centers, Inc. This GPR provides a three-dimensional array of intensity returns corresponding to a volume underneath the ground. In this paper, a novel approach to processing that GPR is described. The approach relies on computing edge direction and magnitude features in the volume and comparing them to prototypes generated using fuzzy c-means clustering. A confidence map is generated corresponding to the surface traversed by the system. The confidence map is thresholded to produce detections. Experimental results show a reduction in false alarm rates from about 40% using the standard processing method to about 4% using the three-dimensional, fuzzy clustering method.

Sensor data fusion techniques are currently being investigated for potential applications in Handheld Standoff Mine Detection System (HSTAMIDS). Overall HSTAMIDS's performance must be optimized and improved over individual sensor's performance. In addition, HSTAMIDS's performance must be carefully traded off with hardware complexity, packaging challenges, increased cost associated with the sensor fusion process and operational requirements in support of military combat missions.

Infrared spectra from the several typical pollution clouds are observed remotely using a passive Fourier Transform Infrared Spectrometer. The primary purpose of the study is to determine an efficient method to distinguish hazardous-cloud from several clouds. Spectral pattern recognition techniques are employed to suppress the strong and highly varying background to extract the very weak emission features from measured spectra. It can be used to discriminate specific pollution cloud between several smokes and interferents.

Electro-optic (EO) sensors can be used for mine identification in turbid shallow water and very shallow water domains in which spatially- and temporally-variant conditions severely limit the effectiveness of acoustic and visual sensors. Also, EO sensors produce images with significantly finer resolution than acoustic sensors, making identification of mines possible rather than just classification of minelike targets. This paper presents image processing and enhancement techniques that improve mine identification probability. Due to the large variation in brightness, contrast, and texture across EO images, virtually all processing performed on these images must be local in nature. In an effort to compensate for the differences in image quality, several preprocessing techniques for noise reduction and contrast enhancement are applied to the images. A texture analysis is performed on the enhanced images in an effort to locate likely man-made surfaces. Of particular interest, the SUSAN (Smallest Univalue Segment Assimilating Nucleus) principle is investigated as a tool for both image enhancement and texture analysis. The performance of this technique, recently introduced by researchers from Oxford University, is compared with that of more traditional methods. The incompatibility of processes that produce visually appealing images with those that perform target detection is illustrated.

An important constraint upon the design of unmanned demining vehicles is the pressure they exert on the ground. To provide for the safety of the expensive equipment carried, vehicles designed to detect land mines must disturb the ground in a mine field as little as possible. Currently vehicle designs concentrate on properly integrating sensors onto a vehicle, paying less attention to whether the vehicles are appropriate concerning safety. This paper/presentation will describe a ground pressure measurement device which measures ground pressure as experienced by a land mine. Inside a rugged case similar an anti-tank mine, the device accurately measures pressures exerted by a vehicle which could detonate a mine. The device, with a wide pressure range and its sealed case, can be used under a wide variety of conditions. The tests performed to validate the device will be described, and a theoretical analysis of the results using terramechanics will be given. Finally, possible usage of this device will be given, including its applicability in the vehicle design process as well as possible usage as a training device.

In deciding on a radar processing algorithms for detection systems of mines and minelike targets the essential attention is given to the problem of resolution and precision of these systems. The corresponding signal-to-noise ratio is determined for the high quality detection and accuracy of measurements. In this event the problem of signal amplitude precision that is very important for many other cases is not considered as usually. This article concerns to the problems of determination of information losses for detectors constructed in accordance with the generalized approach to signal processing. The information losses are an effect of interferences. The relation between the information losses and the ambiguity Woodward function is determined. Results of experimental researches for the generalized receiver are presented.

We examine a ground-penetrating radar system that uses two antennas with their boresight direction maintained at or near the Brewster angle for the air-soil interface. For signals going to and returning from the buried target, this arrangement minimizes reflections from the air-soil interface for TM waves and can be used for both ground-based and aircraft-mounted systems. A major problem associated with reliable detection of mines is the weak return signal from the target. This system maximizes signal return from the target by minimizing all reflections except the desired signal reflection from the target. We analyzed some soil samples to measure their complex permittivity versus frequency. The real part of the complex permittivity ranged from 3 to 8 times that of free space, thus, the Brewster angle is between 60 and 70.5 degrees. For relatively dry soil and for damp soil, we present calculations of transmission loss versus frequency, due to reflection and absorption, from transmitter to receiver for both TM and TE waves.

Airborne imaging of submerged targets is key to a variety of applications in environmental, law enforcement, and military surveillance and interdiction. As shown in previous research, imaging through the Marine Boundary Layer (also called trans- MBL imaging) incurs distortions due to refraction at the air/water interface and scattering or absorption within the water column. In Part 1 of this series of papers, we presented theory and algorithms for model-based trans-MBL stereoscopic imaging of submerged targets that can provide information concerning target depth and, hence, range-to-target. Additionally, an error analysis was presented which highlights the effect on estimated target depth of key errors in various trans-MBL sensor and imaging model parameters. In this paper, we present simulation results and analysis that illustrate the consequences of off-nadir versus nadir or near-nadir imaging configurations discussed in Part 1. We also discuss and analyze techniques for referring depth estimation error to the focal plane, based on the analysis of Part 1. Finally, we analyze simulation results wherein submergence depth of reconstructed targets is estimated from stereo imagery. Analyses emphasize accuracy of and variability of target depth estimation for various nadir or off-nadir viewing configurations.

The main disadvantage of the IR thermography method for detection of buried mines is the presence of plentiful false indications in a thermogram. This often makes the level of false alarm too high. In a single and raw thermogram it was possible to identify mine with a naked eye only in rather rare and very carefully chosen, circumstances. We stated, that more attention should be directed on recognizing time evolution of detected signals, the same as recognizing usefulness of concepts and methodologies used in Dynamic Quantitative IR Thermography and active Thermal NDT. This paper presents the time dependent aspects of phenomenology potential detection mechanisms of buried mines, the same as selected methodologies developed for Thermal NDT and Dynamic Quantitative IR Thermography research and industrial applications. Results of the first trials in application software specialized for typical dynamic QIRT problems are shown, too.

At present there are approximately 110 million land-mines scattered around the world in 64 countries. The clearance of these mines takes place manually. Unfortunately, on average for every 5000 mines cleared one mine clearer is killed. A Mine Detector Arm (MDA) using mechatronics approach is under development in this work. The robot arm imitates manual hand- prodding technique for mine detection. It inserts a bayonet into the soil and models the dynamics of the manipulator and environment parameters, such as stiffness variation in the soil to control the impact caused by contacting a stiff object. An explicit impact control scheme is applied as the main control scheme, while two different intelligent control methods are designed to deal with uncertainties and varying environmental parameters. Firstly, a neuro-fuzzy adaptive gain controller (NFAGC) is designed to adapt the force gain control according to the estimated environment stiffness. Then, an adaptive neuro-fuzzy plus PID controller is employed to switch from a conventional PID controller to neuro-fuzzy impact control (NFIC), when an impact is detected. The developed control schemes are validated through computer simulation and experimental work.

The innovative technique for detection of artificial objects, such as mines, pipes, containers, etc., buried in the ground, is developed and tested. The technique does not depend upon the material from which the object is fabricated whether it be metal, plastic, wood, or any other material. It depends upon the fact that a mine is a 'shell' whose purpose is to contain explosive materials and associated detonation apparatus. The mine shell is in contact with the soil in which it is buried. The shell is an acoustically compliant article, which compliance is notably different from the compliance of the surrounding soil. This difference is responsible for the mechanically nonlinear behavior of the soil/shell interface making it the detectable entity. Thus for this new technology, the fact that the mine is buried is turned to a detection advantage. Because the technique intrinsically detects buried 'shells,' it is insensitive to rocks, tree roots, chunks of metal, bricks, etc. which was confirmed experimentally. The paper discusses physical mechanisms of the nonlinear behavior of the soil-mine interface, the results of experimental investigation of the observed nonlinear interaction, and demonstration of landmine detection technique based on the discovered phenomenon.

The Savannah River Technology Center (SRTC) is developing the Microbial Mine Detection System (MMDS), a cost-effective, safe and reliable method to detect land mines using microorganisms as the primary biosensor detector. SRTC research has shown that various naturally occurring microbial species are stimulated by nitrogen, trinitrotoluene (TNT), dinitrotoluene (DNT), nitrates, nitrites, nitrous oxide, and the chemical components found in explosive materials. Several of the 10,000 indigenous bacteria already existing in the SRTC Subsurface Microbiology Culture Collection (SMCC) possess characteristics that would support discrete detection of land mines during metabolic activity or growth. SRTC scientists are screening and identifying bacteria residing in the SMCC, and other collections associated with specific land mines, for their attraction to explosive off-gasses. After contacting explosives or off-gasses, the micro-organisms will activate via bioluminescence and identify the location of the land mines. Once identified, down selected and mesocosmly defined, the micro-organisms can then be prepared for field deployment. This deployment process requires minimal user training and is envisioned to be administered in hand-held, vehicular mounted and airborne platforms. Microbial detection systems are a renewable resource, easy to preserve, inexpensive to maintain under field conditions, and provide a high-probability response recognition technology.

Surface-enhanced Raman scattering has been measured for solutions of 2,4-dinitrotoluene (2,4-DNT), trinitrotoluene (TNT), and hexahydro-1,3,5-trinitro-s-triazine (RDX) adsorbed onto metal foils, island films, coated microspheres, and colloids. The wavelength selectivity of the method was also investigated for lasers which potentially may be used in a field instrument. Preliminary room temperature vapor-phase SERS detection of 2,4-DNT adsorbed on gold foil has been achieved. The results demonstrate the potential of SERS as a detector of buried landmines when coupled to compact man- portable Raman instrumentation.

Mines present a severe and ongoing threat to operations of the Navy. They are available to any potential adversary. Dedicated mine countermeasures assets are not likely to be available in sufficient numbers to effectively counter the threat. Several systems can be mounted organically in surface combatant units. They include onboard systems and Remotely Operated Vehicles carrying Advanced Sensors capable of seamless mine detection in all waters. No one system can perform the entire mission, and a mix of appropriate systems will be required. This paper will describe the nature of the mine threat. It will discus various available systems that can be used organically with the Fleet, the considerations that were operative in the design of each system, and the special role that each system can play given its peculiar characteristics. It is stressed that the Navy will be forced to deal with mines for the indefinite future, and that solutions can be and have been found for this problem.

Several studies are under way at the U.S. Army Cold Regions Research and Engineering Laboratory (CRREL) to define environmental effects on detection and classification of buried mines and unexploded ordnance (UXO). Ground that is very wet, frozen, or snow covered can pose severe constraints on demining operations. The qualitative and quantitative nature of chemical signatures of buried land mines is being documented. Research to date indicates that although 2,4,6- trinitrotoluene constitutes over 99% of military-grade TNT, it is a minor component of the vapor signature at ground level. CRREL operates a year-round test site to determine the effect of weather on radar and IR systems used to detect buried mines. The New England site experiences many of the weather conditions likely to interfere with mine detection around the world. Short-pulse ground penetrating radar (GPR) was used to profile both ordnance and non-ordnance targets at the 40-acre UXO site at Jefferson Proving Ground. Analysis of the data indicates that future systems will have to operate at faster data acquisition rates. Radar modeling is being used to simulate the effects of the environment and identify new techniques for finding and classifying buried ferrous objects.