Current counter-mine and obstacle avoidance technology is inadequate, and limits the Navy's capability to conduct shallow water (SW) and very shallow water (VSW) MCM in support of beach assaults by Marine Corps forces. Without information as to the location or density of mined beach areas, it must be assumed that if mines are present in one area then they are present in all areas. Marine mammal systems (MMS) are an unusual, effective and unique solution to current problems of mine and obstacle hunting. In the US Navy Mine Warfare Plan for 1994-1995 Marine Mammal Systems are explicitly identified as the Navy's only means of countering buried mines and the best means for dealing with close-tethered mines. The dolphins in these systems possess a biological sonar specifically adapted for their shallow and very shallow water habitat. Research has demonstrated that the dolphin biosonar outperforms any current hardware system available for SW and VSW applications. This presentation will cover current Fleet MCM systems and future technology application to the littoral region.

Echolocating bats use sonar to guide interception of insects, recognize objects by shape, and even track prey in clutter. Broadcasts of the big brown bat are 0.5 to 20 ms FM signals in the 20-100 kHz ultrasonic band. Insects consist of several reflecting glints, each equivalent in cross- section to a small sphere of 2 mm to 2 cm radius, while clutter is typically composed of numerous glints distributed over a large volume. The bats' signals extend in space for many target lengths, while ka values for each glint are 0.5 to 30 across the broadcast band. Bats perceive acoustic images having echo delay as their primary dimension, and space is perceived in terms of the distribution of target glints in range. Range disparities between the ears provide two 'looks' at each target from slightly different locations as well as information about azimuth. The bats auditory system encodes the FM sweeps of broadcasts and echoes as linear-period spectrograms with integration-times of 300-400 micrometers . Bats nevertheless perceive individual glints in targets for echo-delay separations well inside the integration-time window. Deconvolution is achieved by spectrogram correlation in the time domain and spectral shape transformation in the frequency-domain, with all output evidently being displayed in the time domina. Neural responses in the bat's auditory system seem limited in time precision to 20-50 micrometers at best and 300 microsecond(s) to 3 ms in a broader sample, and stimulus phase is thought to be lost for frequencies above 1-3 kHz. Yet bats perceive echo delay with an accuracy of 10-15 ns and have two-echo resolution of about 2 microsecond(s) . Moreover, bats perceive echo phase-shifts as the correctly corresponding shifts in echo delay. Successive images are subtracted to enhance perception of shape from multiple 'looks', and echo phase is an integral part of this critical process. Utterly novel time-scale magnification appears in the bat's neural responses to ultrasonic broadcasts and echoes, with time-stretch factors of roughly 10 to 100 that account for the bat's mysteriously high temporal acuity. Bats reconstruct time-domain features of FM broadcasts and echoes in a compact, distributed format that incorporates delay, phase, glint separation, and inter-ear delay differences along the same image dimension.

Acoustic buried object detection by a dolphin was observed by means of a hydrophone placed directly above the object. Dolphins are known to be adept at locating prey and objects that are buried in the sediment. This involves the detection and some level of identification of the prey. Since sediment is opaque, it is unlikely that vision is involved. It is likely that the dolphin uses sound to detect buried objects. Direct observation of the sound pulses arriving at the buried object provides information about the sound pulses employed by the dolphin. The observations show that the acoustic pulses are broad band, spanning the frequency range from a few tens of kilohertz to well over one hundred kilohertz. Modeling of the acoustic penetration of such a broad band pulse indicates that it will change shape as it is transmitted through the water-sediment interface. The change will depend on the sediment type and signal grazing angle. As the pulse penetrates into the sediment, it will undergo further distortion due to the frequency dependence of acoustic absorption. Acoustic penetration of the sediment has been modelled and the signal distortion may be expressed as a filter function, with certain characteristics, particularly a phase shift and a reduction at the higher end of the spectrum. These features provide an indication of whether a detected object is buried or not, and the depth of burial.

Active sonar classification of suspended, bottomed, and buried mines is very important in littoral warfare. Current active sonars are inadequate because they require many emissions per potential target, yield high false alarm rates, and suffer from high clutter interference. New active biosonar models based on bat-like range profiling and dolphin-like image construction may reduce these problems. The performance of one such biosonar algorithm, the spectrogram correlation and transformation model developed at Brown University, has been compared with the performance of a standard matched filter on a data set obtained from NSWC Coastal Systems Station, Dahlgren Division. This data set contains echoes form six objects (two mine-like objects, a water-filled 50-gallon drum, a rough limestone rock, a smooth granite rock, and a water-saturated log). Three neural network architectures were used as classifiers. Discrimination was performed between man-made and non-man- made objects, between mine-like and non-mine-like objects, among the three types of man-made objects, and among the six different test objects using single pings, multiple ping fusion, fusion of the results from different algorithms, and a combination of algorithm fusion and multiple ping fusion.

In processing the Toroidal Volume Search Sonar (TVSS) data, the primary challenge is to detect targets in highly reverberant environments. This may be done by processing each sonar channel separately and estimating the time delays associated with the specular returns and then isolating these components from the background noise prior to beamforming. However, if there is significant reverberation it is difficult to accurately estimate time delays using only one sonar channel. Therefore, a multi-channel method is introduced that exploits the interdependency of adjacent sonar stave data. The method starts by enhancing signal to noise by combining multiple stave data after suitable phase compensation for array geometry. Next, time delays of the specular returns are estimated from the discrete wavelet sub-bands of the combined signal using a least squares method. The new method has been tested on several TVSS data sets, and the results are promising for even the most reverberant cases.

Higher order statistics heave been shown to be useful in applications where non-Gaussian or colored Gaussian noise is present. In this paper, fourth order cumulant-based MUSIC and matrix pencil-type algorithms were presented for the estimation of specular components in the acoustic backscattered data. Higher order Toeplitz matrices are formed from the sample cumulants and the generalized eigen- structure of these matrices are studied and exploited to develop high resolution estimators. The performance of these methods is examined on several data sets.

A mine detection algorithm based on the us of structured templates applied to acoustic backscatter data is proposed. The structured templates correspond to the codevectors of a type of cluster-based compression algorithm called residual vector quantization (RVQ). The RVQ clusters have a hierarchical structure that permits efficient searches for nearest neighbor templates, and efficient dictionary storage for memory cost reduction. The structured templates are generated by a multistage synthesis process that produces a sequence of finite precision representations of training data. This successive approximation process is combined with a sequential classification process to form a new type of classifier called a direct sum successive approximation classifier.

Supported by ONR, for the past two decades many researchers have been involved in 'imaging' underwater with background noise, 'Acoustic Daylight'. In the January 1996 issue of Scientific America, Dr. Buckingham and others published an article indicating that it is feasible to image underwater with background noise. Since then, many authors have published their successes in imaging underwater in Acoustic Daylight. In this paper, the author will discus his acoustic noise experiments in Kaneohe Bay with Dr. W. Au. The author presents his findings based on the theory of resonance and/or resonant scattering theory. Based on resonance technique, it appears that underwater targets can be detected and may be identified in background noise, Acoustic Daylight.

An advanced capability for automated detection and classification of sea mines in sonar imagery has been developed. The advanced mine detection and classification (AMDAC) algorithm consists of an improved detection density algorithm, a classification feature extractor that uses a stepwise feature selection strategy, a k-nearest neighbor attractor-based neural network (KNN) classifier, and an optimal discriminatory filter classifier. The detection stage uses a nonlinear matched filter to identify mine-size regions in the sonar image that closely match a mine's signature. For each detected mine-like region, the feature extractor calculates a large set of candidate classification features. A stepwise feature selection process then determines the subset features that optimizes probability of detection and probability of classification for each of the classifiers while minimizing false alarms.

An automatic, robust, adaptive clutter suppression, predetection level fusion, sea mine detection and classification processing string has been developed and applied to shallow water side-scan sonar imagery data. The overall processing string includes pre-processing string includes pre-processing, adaptive clutter filtering (ACF), 2D normalization, detection, feature extraction and classification processing blocks. The pre-processing block contains automatic gain control, data decimation and data alignment processing. The ACF is a multi-dimensional adaptive linear FIR filter, optimal in the least squares sense, for simultaneous background clutter suppression and preservation of an average peak target signature. After data alignment, using a 3D ACF enables simultaneous multiple frequency data fusion and clutter suppression in the composite frequency-range-crossrange domain. Following 2D normalization, the detection consists of thresholding, clustering of exceedances and limiting their number. Finally, features are extracted and a orthogonalization transformation is applied to the data, enabling an efficient application of the optimal log-likelihood-ratio-test (LLRT) classification rule. The utility of the overall processing string was demonstrated with two side-scan sonar data sets. The ACF, feature orthogonalization, LLRT-based classification processing string provided average probability of correct mine classification and false alarm rate performance exceeding the one obtained when utilizing an expert sonar operator. The overall processing string can be easily implemented in real-time using COTS technology.

At long range, only the highlight echoes are available for classifying targets as minelike; their variation with aspect provides vital classification clues. Gathering this data with experimental equipment presents new challenges for planning, operation and post processing. The course of the ship varies from minute to minute, as it attempts to follow the tracks between waypoints. the sonar array orientation must compensate for the motion, to maintain the sonar footprint over the target. This task is relatively straight forward with the wide sector angels and large range extent of in-service minehunting sonars. Some experimental wide band sonar modules however only process a few tens of meters of range extent. Skilled operators, low seastage and graphic tools are then needed to acquire the target echo. DRA is writing software modules to visualize and plan the mine hunting strategy and fusion of data in such scenarios, aimed at trials planners, sonar operators, and processing engineers.

Research in FY-95 first addressed the problem of combining high-frequency (HF) side-scan sonar imagery, low-frequency (LF) side-scan Sonar imagery, and magnetic gradiometer data in order to detect/classify undersea mines. The first approach developed, termed the "Blob-Pair" based acoustic/magnetic (AM) Fusion system architecture, implicitly assumed that a target manifests itself in both HF,LF imagery, and was based on the fusion of single-sensor derived neural network classifier discriminants at a collection of three "decision" nodes, identified with magnetic (M), HF/LF, and HF/LF/M - data fusion cases, respectively. In order, to remove the restrictive assumption of a target manifesting in both }IF,LF data, the "Generalized" AM Fusion Architecture was developed, with a total of 7 "decision" nodes, identified with M, HF, HF/LF, LF, HF/M, HF/LFIM, and LF/M data fusion cases, respectively. However, the "Generalized" AM-Fusion architecture was found empirically to have significantly increased number of false alarms, relative to the "Blob-Pair" based system. Hence, through two-additional AM-Fusion architecture varaints, involving first the use of Classification Token "Post-Processing", and then both Token "Post-Processing" and decision node statistic modification, the performance "gap" between "Blob-Pair" and "Generalized" AM-Fusion Architecture performance was closed.

The HF/LF SAS is a high resolution SAS developed by COASTSYSTA and Northrop Grumman for the shallow water (SW) and very shallow water (VSW) regimes. This sonar suite has recently been delivered to COASTSYSTA and it is currently undergoing field testing. This article describes this sonar and the type of resolution and acoustical images which are expected from this sonar. The application of this sonar to the SW and VSW regimes required the development of a novel method of motion compensation. A description of this method and the type of accuracy's expected from this technique are presented. Finally, a look at future broad band system and their predicted performance is presented.

An important sonar signal processing problem is the detection of targets under highly reverberant conditions. To address this issue two wavelet-based approaches have been developed. Both have been designed to improve the signal-to- reverberation ratio of each sonar channel prior to beamforming. The first method uses the wavelet shrinkage method to denoise the sonar signals. The second method uses a sub-band adaptive filter to estimate the target location in the sonar signals so that reverberation effects can be easily removed. Both detection methods have proven to be very effective even at high reverberation levels.

Underwater (UW) imagery presents several challenging problems for the developer of automated target recognition (ATR) algorithms, due to the presence of noise, point-spread function (PSF) effects resulting from camera or media inhomogeneities, and loss of contrast and resolution due to in-water scattering and absorption. Additional problems include the effects of sensor noise upon lossy image compression transformations, which can produce feature aliasing in the reconstructed imagery. Low-distortion, high- compression image transformations have been developed that facilitate transmission along a low-bandwidth uplink of compressed imagery acquired by a UW vehicle to a surface processing or viewing station. In early research that employed visual pattern image coding and the recently- developed BLAST transform, compression ratios ranging from 6,500:1 to 16,500:1 were reported, based on prefiltered six- band multispectral imagery of resolution 720 X 480 pixels. The prefiltering step, which removes unwanted background objects, is key to achieving high compression. This paper contains an analysis of several common compression algorithms, together with BLAST, to determine compression ratio, information loss, and computational efficiency achievable on a database of UW imagery. Information loss is derived rom the modulation transfer function, as well as several measures of spatial complexity that have been reported in the literature. Algorithms are expressed in image algebra, a concise notation that rigorously unifies linear and nonlinear mathematics in the image domain an has ben implemented on a variety of workstations and parallel processors. Thus, our algorithms are feasible, widely portable, and can be implemented on digital signal processors and fast parallel machines.

In distributed underwater signal processing for area surveillance and sanitization during regional conflicts, it is often necessary to transmit raw imagery data to a remote processing station for detection-report confirmation and more sophisticated automatic target recognition (ATR) processing. Because of he limited bandwidth available for transmission, image compression is of paramount importance. At the same time, preservation of useful information that contains essential signal attributes is crucial for effective mine detection and classification in shallow water. In this paper, we present an integrated processing strategy that combines image compression and ATR algorithms for superior detection performance while achieving maximal bandwidth reduction. Our reduced-dimension image compression algorithm comprises image-content classification for the subimage-specific transformation, principal component analysis for further dimension reduction, and vector quantization to obtain minimal information state. Next, using an integrated pattern recognition paradigm, our ATR algorithm optimally combines low-dimensional features and an appropriate classifier topology to extract maximum recognition performance from reconstructed images. Instead of assessing performance of the image compression algorithm in terms of commonly used peak signal-to-noise ratio or normalized mean-squared error criteria, we quantify our algorithm performance using a metric that reflects human and operational factors - ATR performance. Our preliminary analysis based on high-frequency sonar real data indicates that we can achieve a compression ratio of up to 57:1 with minimal sacrifice in P_D and P_FA. Furthermore, we discuss the concept of the classification Cramer-Rao bound in terms of data compression, sufficient statistics, and class separability to quantify the extent to which a classifier approximates the Bayes classifier.

We develop lossy compression algorithms for underwater acoustic data and evaluate the effects of the compression on two applications: target-detection and study of ocean floor temperature. We use data from an experiment of sediment transport conducted by the Applied Physics Laboratory at the University of Washington. We apply a variety of wavelet- based vector quantization data compression algorithms to acoustic sonar scans. We sue pruned tree-structured vector quantization (PTSVQ) with the generalized Breiman, Friedman, Olshen, and Stone algorithm to simultaneously prune trees that correspond to different wavelet subbands. We determine that while targets can be detected at compression ratios of over 100:1, compression ratios greater than 4:1 lead to unacceptable loss in accuracy for use of the data in ocean floor temperature studies. We find that PTSVQ applied to wavelet coefficients uniformly gives better results at low bit rates than PTSVQ applied to the untransformed data. For target detection, better compression is obtained by compressing in polar coordinates while for ocean temperature measurement, better compression is obtained by compressing in Cartesian coordinates. Finally although SNR versus entropy measurements are a popular and easy way of measuring the success of compression experiments, they are not good predictors of compression performance for scientific data.

Past work in multispectral data fusion has focused on either ad-hoc techniques or conventional variants of the multivariate Gaussian model. This work develops a generalized likelihood ratio for pixel-level fusion that utilizes an unconventional region of support for the unknown target signature. The intent is to demonstrate that partial knowledge of target signature or clutter characteristics can be effectively captured and utilized in a fusion algorithm based on this likelihood ratio. The choice of model was motivated by the need to account for the variability of the multispectral target signatures; in essence, the target signature is modeled to be known within a band of spectral intensities. The resulting estimators were determined initially to be computationally complex, growing in complexity at an exponential rate with the number of bands to be fused. One result of the work was that an algorithm was created that kept the complexity cubic with respect to the number of spectral bands. Another intriguing result was that classic linear and quadrati detectors are special cases of the generalized likelihood ratio; the optimum detector structure for orthant support sets is a linear-quadratic detector, where the subsets of bands to be processed in each mode are in effect chosen on-the-fly. Additional results and examples will be presented.

The accurate, comprehensive estimation of computational error has long been a key problem in the design of automated target recognition (ATR) algorithms due to the extensive manual computation customarily required, as well as lack of theory and software support for automated error analysis. For example, ATR algorithms often perform poorly in the presence of sensor noise at the algorithm input, which is propagated through detection, segmentation, or recognition processes. Since computational error increases with each nontrivial computation, tracking of such error in ATR and image processing algorithms would be advantageous for a variety of applications. In this series of two paper, we present an automated, compilation-based technique for profiling computational error. Our method is similar in concept to the well-known software engineering technique of execution profiling. Such performance estimation procedures accept algorithm source code and produce statistical data that describe the frequency with which various user-defined or built-in functions are invoked. Given such data, algorithm performance can be analyzed in terms of cost measures such as space and time complexity, which supports performance optimization. We herein extend the concept of execution profiling to include the profiling of computational error. Our method employs error probability distributions computed from error functions that are derived from the parse tree which represents a given expression or algorithm. Distributions are characterized in terms of customary descriptors derived from a bivariate probability density map. In this paper, we present basic theory and algorithms for automated error analysis, together with a description of our prototype software system called ERSIA. ERSIA, written in ANSI-standard FORTRAN-77, is portable to numerous workstations and parallel processors.

The two primary measures of detection performance are probability of detection (P_D) and false alarm rate (FAR). Since these are statistical measures, the large statistical uncertainties associated with measurements on small data sets impose a severe limitation on the interpretation and extrapolation of the test results. The main difficulty with the data collected from many of the landmine detection experiments is that the number of mine targets is small, and the clutter data is measured on small areas. As a result of analyzing data from several detection tests for both UXO and mine detection technologies, the Institute for Defense Analyses (IDA) has generated a number of suggestions to improve the results obtained from future demonstrations and tests. In the absence of a rigorous error analysis, we can estimate the uncertainty in a probability of detection measurement with a simple model. If detections can be described as a binomial process weighted by the true probability of detection, then uncertainty equals (root)N, where N is the number of mines detected, and uncertainty in P_D (percent) equals (root)N/N X 100 for large N. A similar statistical uncertainty is encountered in the measurement of probability of false alarm. An opportunity for a false alarm can be defined by the amount of ground covered by the projected area of the target plus an allowable miss distance, say 1 m². Thus, for each 1 m² of ground that it passes over, the system has one opportunity to declare a false alarm. A test field that is 1 m by 20 m offers only 20 m² of area and thus only 20 samples for probability of false alarm measurement. A site that is 4 m wide and 4 km long covers 16,000 m². Thus, there will be a substantial increase in the clutter environment samples by the sensor and a concomitant improvement in false alarm measurements. Of course, false alarm density will be highly site dependent, so extrapolation of the result is still uncertain. Unfortunately, this limitation is insurmountable in single test and at a single site.

Mine and unexploded ordnance (UXO) detection systems must function in highly cluttered environments. Clutter leads to false alarms thereby hindering the detection and identification of targets of interest. Since the end user of the mine or UXO detection technology requires both a high detection rate and a low number of false alarms, technology demonstrations and system evaluations are designed to test these measures of performance. Detection rate and false- alarm rate are highly interdependent and must always be evaluated together. The relationship between the two rates directly affects the overall performance of the sensor in the field. Poor performance of a system in either detection rate or false-alarm rate causes a substantial increase in risk of undetected and thus unmarked or unremediated mines or UXO. A system with a low detection rate will leave many mines or UXO undetected. Performance can be traded between probability of detection and false-alarm rate by changing the system threshold. Raising or lowering the threshold will cause both the detections and false alarms to decrease or increase together. A system with a high false-alarm rate result in an increase in the time required to investigate potential targets. Therefore the rate of advance and rate of clearance decrease. With limited clearance resources, site coverage may become too time consuming or costly for operationally effective clearance, resulting in risk from undetected mines and UXO in areas that have not been searched. An assessment of the connection between detection rate nd false-alarm is presented. This relationship is discussed in the context of several government-sponsored in- field technology demonstrations of prototype and commercially available mine and UXO detection technologies, as well as real clearance operations. Implications of the results of these tests and the measures of performance are discussed in the context of real-world operations, including scenarios for clearance of miens in Bosnia and of UXO at DoD sites.

Coastal Systems Station has translated its synthetic aperture sonar beamformer from linear processing to parallel processing. The initial implementation included many linear processes delegated to individual processors and neglected algorithmic refinements available to parallel processing. The steps taken to achieve increased computational speed for real-time beam forming are presented.

Questions of detector synthesis based on the generalized signal processing algorithm for signals with stochastic parameters are considered. Variance and variance estimation of total noise component at the generalized detector output under finite time interval are determined. Comparative analysis of detection characteristics of the optimal and generalized detectors is carried out. Investigation avenues to stabilization of detection characteristics under employment of the generalized detector independent of signal random parameters based on amplitude and phase tracking systems have been proposed.

A superconducting magnetic-field gradiometer developed in the 1980's has been demonstrated infusion with acoustic sensors to enhance shallow water sea mine detection and classification, especially for buried mine detection and the reduction of acoustic false alarm rates. This sensor incorporated niobium bulk and wire superconducting components cooled by liquid helium to a temperature of 4 degrees K. An advanced superconducting gradiometer prototype is being developed to increase sensitivity and detection range. This sensor features all thin film niobium superconducting components and a new liquid helium cooling concept. In the late 1980's, a new class of 'high Tc' superconductors was discovered with critical temperatures above the boiling point of liquid nitrogen. The use of liquid nitrogen refrigeration offers new opportunities for this sensor technology, providing significant reduction in the size of sensor packages and in the requirements for cryogenic support and logistics. As a result of this breakthrough, a high Tc sensor concept using liquid nitrogen refrigeration has been developed for mine reconnaissance applications and a test article of that concept is being fabricated and evaluated. In addition to these developments in sensor technology, new signal processing approaches and recent experimental results have ben obtained to demonstrate an enhanced D/C capability. In this paper, these recent advances in sensor development and new results for an enhanced D/C capability will be reviewed and a current perspective on the role of magnetic sensors for mine detection and classification will be addressed.

A metallic mine detector is one of the most effective pieces of equipment for detection of mines. Their main drawback is their extremely rate of up to 100 percent, but it can also produce a high false alarm rate in many environments. The high false alarm rate reduces the usefulness of the metal detector in the field. In order to keep a high detection rate with fewer false alarms, object/mine characterization or identification must be used. Several techniques have been implemented to reduce the false alarm rate of metal detectors. They are size discrimination, target imaging, and target signatures such as dipole moment characterization. These techniques are applied for large metallic objects/mines. P.V. Czipott and D.A. Waldron each used separate techniques to characterize smaller metallic objects and some anti-personnel mines, in work supported by US Army CECOM, Night Vision and Electronic Sensors Directorate. Dr. Czipott characterized objects/mines by measuring the frequency dependence of magnetic fields caused by electric currents induced in the target. The frequency responses were measured by using a fixture incorporating a solenoid excitation coil, a receiving coil wound as a gradiometer, and a HP 4195A network/spectrum analyzer. Ms. Waldron characterized small objects with different conductivities and orientations by measuring their phase differences using a search head with one transmitter and four receiver coils and a phase-lock analyzer. We believe that target discrimination and identification are the keys to reduce false alarm rates of metallic mine detectors. Thus, we continue to analyze and characterize small metallic targets/mines using a variety of methods.

Lateral migration radiography (LMR), a form of Compton backscatter radiography, is applied to the detection and identification of landmines. The LMR system consists of two inner uncollimated detectors positioned to optimally detect first scattered photons and two outer collimated detectors designed to detect primarily photons that have had two or more scatterings. The difference between the collimated and uncollimated detector response to both the landmines themselves and the different types of landmine image masking phenomena, form the basis of the image enhancement and landmine identification procedures. Surface feature information is the primary component of the uncollimated detector response. The collimated detector signal contains information about the surface features as well as the buried objects. The principles of the detection system have been shown in previous work and now the focus has shifted to the preparation for field tests and the associated problems. One of the expected events that the detector system will encounter is the variation of detector height with respect to the ground. This is caused by irregularities in the surface as will as oscillations of the detection vehicle. The collimated detectors and the uncollimated detector react differently to height variations. When the detector height increases the uncollimated detector response will be reduced due to the decrease in solid angle. Although the collimated detector will also be affected by the change in solid angle the dominate reaction is the loss of collimation causing the collimated detectors signal to increase. When the detector height decreases the opposite responses are observed. By using the information from both detector systems, the effects of the detector height variation can be removed.

This paper reports the latest development of a miniature gravity gradiometer. This device consists of two pairs of classical spring-mass sensors and microwave displacement detectors. We have completed a set of two collinear accelerometers and report the initial performance characteristics as well as its potential applications to the detection of buried objects. Gravity gradient changes of 1000 Eotvos units were detected using the present two-axis system in a 10-s sampling mode.

The term 'mine hunter/killer' has frequently served as a catch-all for almost any discussion of the integration of mine detection and neutralization technology. With the Army's recent approval of the mine hunter/killer advanced technology demonstration, the term now takes on a specific set of challenges and goals. In this paper, the author discusses the developing operational and functional requirements for the mine hunter/killer, the mine types considered as targets for the system, and the implications of these factors for detection systems and neutralizers alike. The issue of 'stand-off' is explored with an emphasis on the trade-offs that must be made between it and overall system effectiveness. The author further discusses recent developments in both detection and neutralization technology that make the notion of a stand-off mine detection and neutralization system more realistic. This discussion includes a summary of forward looking radar, forward looking IR sensors, and advanced signal processing architectures and algorithms. However, it focuses on stand-off neutralization technology and fire control techniques, their stand-off and area kill capabilities, and the synergism between them and the detection suites. Recent neutralizer test results are provided as a basis for considering the requirements of stand-off mine detection systems for probability of detection, detection halo, and false alarm rate. The author concludes with a review of the mainstream mine hunter/killer concept.

This paper explores the application of standoff vehicular mounted mine detection technologies. These technologies could support a fieldable standoff mine detection system for the digitized battlefield.In the rapidly evolving countermine field, the standoff detection of mines has become a goal for Government, industrial research and development communities. Standoff mine detection is demanded by worldwide armed forces and extolled by industry. Obvious benefits, such as operator safety, increased reaction time, and enhanced vehicle maneuverability sustain standoff developments. In order to determine those inclusive standoff detection requisites the authors discuss overall worldwide vehicular mounted mine detection requirements. The authors provide a tabulated assessment of these worldwide requirements. Consideration is made of both vertical and horizontal sensors which constitute the industrial community's technical developments. Incorporation of common digitized positioning systems and multiply-correlated technologies, will be described. The resultant digitization of the acquired horizontal and vertical standoff data and detection results are considered with respect to a standard Geographic Information System Universal Transverse Mercator databases. These data and results can be applied to later relocation and remediation activities. Finally requirements and technologies are crosswalked to determine the best technology application at the least operational risk. Concluding remarks provide suggestions for the US Army and industry.

The integrated ground mobile mine detection testbed, developed for the countermine division of the US Army Night Vision Laboratory at Ft. Belvoir, MD, deploys a multisensor suite comprised of a ground penetrating radar, an IR sensor, an electromagnetic sensor and a global positioning system receiver. The physical co-location of these sensors and the positional correlation of their detection results has created a state-of-the-art multisensor fusion platform for real time detection of mines. Government testing of this system in both arid and temperate climates has produced a wealth of data for post processing and performance analysis. In addition to providing real-time detection and marking, the system performs sufficient data storage as to allow subsequent playback of test data, with modified operational parameters and/or alternative integrating techniques, thereby extending the value of the system as a testbed. The result is a laboratory tool using real-world data for algorithmic development. The purpose of this paper is to present some of the test results along with some analytical insights into the performance of the integrating elements of the system, including navigation, elementary sensor fusion, and confirmed sensor detection, plus an initial evaluation of the ability of the correlation procedure to reduce false alarms, and characterize targets as to type. In addition, we now have a platform and a baseline for development and evaluation of more sophisticated fusion techniques. The statistics resulting from these tests will be presented and discussed, along with proposals for extending and/or enhancing the sensor fusion methods.

State-of-the-art in multi-sensor (MSI) application involves extensive research and development time to understand and characterize the application domain; to determine and define the appropriate sensor suite; to analyze, characterize, and calibrate the individual sensor systems; to recognize and accommodate the various sensor interactions; and to develop and optimize robust merging code. Much of this process can benefit from adaptive learning, i.e., an output-based system can take raw sensor data and desired merged results as input and adaptively develop an effective method of interpretation and merger. This approach significantly reduces the time required to apply MSI to a given application, while increasing the quality of the final result and provides a quantitative measure for comparing competing MSI techniques and sensor suites. The ability to automatically develop and optimize MSI techniques for new sensor suites and operating environments makes this approach well suited to the detection of mines and mine-like targets. Perhaps more than any other, this application domain is characterized by innovative and dynamic sensor suites, while nature and interactions are not yet well established. This paper presents such an outcome-based multi-image analysis system. An empirical evaluation of its performance, application, and sensor and domain robustness is presented.

Oak Ridge National Laboratory and Quantum Magnetics, Inc. are exploring novel landmine detection technologies. Technologies considered here include bioreporter bacteria, swept acoustic resonance, nuclear quadrupole resonance (NQR), and semiotic data fusion. Bioreporter bacteria look promising for third-world humanitarian applications; they are inexpensive, and deployment does not require high-tech methods. Swept acoustic resonance may be a useful adjunct to magnetometers in humanitarian demining. For military demining, NQR is a promising method for detecting explosive substances; of 50,000 substances that have been tested, one has an NQR signature that can be mistaken for RDX or TNT. For both military and commercial demining, sensor fusion entails two daunting tasks, identifying fusible features in both present-day and emerging technologies, and devising a fusion algorithm that runs in real-time on cheap hardware. Preliminary research in these areas is encouraging. A bioreporter bacterium for TNT detection is under development. Investigation has just started in swept acoustic resonance as an approach to a cheap mine detector for humanitarian use. Real-time wavelet processing appears to be a key to extending NQR bomb detection into mine detection, including TNT-based mines. Recent discoveries in semiotics may be the breakthrough that will lead to a robust fused detection scheme.

A concept of a modular multisensor system for use on an airborne platform is presented. THe sensor system comprises two high resolution IR sensors working in the mid and far IR spectral regions, a RGB video camera with its sensitivity extended to the near IR in connection with a laser illuminator, and a radar with a spatial resolution adapted to the expected mine sizes. The sensor concept emerged from the evaluation of comprehensive static and airborne measurements on numerous buried and unburied mines. The measurements were performed on single mines and on minefields, layed down according to military requirements. The system has an on-board realtime image processing capability and is intended to operate autonomously with a data link to a mobile groundstation. Data from a navigation unit serve to transform the location of identified mines into a geodetic coordinate system. The system will be integrated into a cylindrical structure of about 40 cm diameter. This may be a drone or simply a tube which can be mounted on any carrier whatever. The realization of a simplified demonstrator for captive flight tests is planned by 1998.

Within the concept study, Material Requirements for an airborne minefield detection systems (AMiDS) the following topics were investigated: (i) concept concerning airborne minefield detection technique sand equipment, (ii) verification analysis of the AMiDS requirements using simulation models and (iii) application concept of AMiDS with regard o tactics and military operations. In a first approach the problems concerning unmanned airborne minefield detection techniques within a well-defined area were considered. The complexity of unmanned airborne minefield detection is a result of the following parameters: mine types, mine deployment methods, tactical requirements, topography, weather conditions, and the size of the area to be searched. In order to perform the analysis, a simulation model was developed to analyze the usability of the proposed remote controlled air carriers. The basic flight patterns for the proposed air carriers, as well as the preparation efforts of military operations and benefits of such a system during combat support missions were investigated. The results of the conceptual study showed that a proposed remote controlled helicopter drone could meet the stated German MOD scanning requirements of mine barriers. Fixed wing air carriers were at a definite disadvantage because of their inherently large turning loops. By implementing a mine detection system like AMiDS minefields can be reconnoitered before an attack. It is therefore possible either to plan, how the minefields can be circumvented or where precisely breaching lanes through the mine barriers are to be cleared for the advancing force.

We demonstrate a three stage procedure for mine detection. The first phase locates plausible candidates through a series of hypothesis tests. The second state finds the globally optimal convex set containing each candidate using 'course-to-fine' dynamic programming. The third stage refines the mine shape estimates using the knowledge contained in a mine-shape library.

Laser line scan imagery has been found to have fluctuating brightness regions due to high/low signal strengths when scanning data. The low signal strength regions can obscure visibility of potential mines, sometimes to the point of preventing identification by an operator inspecting the data. Two contrast enhancement routines have been developed to enhance obscured objects in the low signal strength regions: a background equalization routine that equalizes the high/low signal strength regions and a local histogram clipping routine that applies a moving window histogram clip to enhance details in the low strength regions without effecting the high strength regions.

In this paper we introduce a method for detecting landmines from remotely sensed data. Our primary goal is to minimize the number of false negatives, while keeping the number of false positive very low.We concentrate on two different phases of the processing: detection and classification. We construct a feature based detector which utilizes some basic signal processing techniques to produce a detection map which has very few false negatives, and quite a few false positives. The detection phase of this paper concentrates on a background equalized matched filter, along with some measures of symmetry. After the detection phase we will construct a classifier for the problem which concentrates on estimating the density distributions of the pixel values. We will show that a variation of 'Borrowed Strength' as introduced by Priebe, significantly improves the classification process. This classification technique will reduce the number of false positives, without significantly increasing the number of false negatives. The procedure is demonstrated on passive multispectral mine data generously provided by the NSWC Dahlgren Division. Coastal Systems Station, Panama City, Florida. In conclusion, we propose a method for improving the performance of this type of classification. We believe that the proposed method shows a great deal of promise, and will be of use in many situations.

Minefields have point patterns that tend to exhibit regularity such as equal-spacing and collinearity that provide potentially valuable discriminants against natural occurring clutter. Previously, several simple procedures based on the empty boxes test and its extensions have ben shown to be effective detectors of generic regularity in minefields without explicitly taking advantage of collinearity and equal-spacing. Recently, modifications of the Euclidean algorithm have been applied to the problem of determining the period from a sparse set of detection times which arises in pulse repetition interval analysis. Besides the intriguing properties of this approach because of its number theoretic roots, the resulting algorithms are both computationally efficient and robust to errors introduced by missed detections, noise in time location, and false alarms. In this paper, a two-step procedure for detecting minefields is proposed whereby collinear points are first detected using a standard approach, the Hough transform, and the period estimated using the modified Euclidean algorithm. One advantage of this approach is that prior information on minefield spacing can be utilized to exclude collinear points that do not exhibit any periodicity or are spaced too close together or too far apart. The preliminary detection performance of some of these minefield detection methods are quantified for using a point pattern extracted from real sensor data.

In many applications one wishes to perform an analysis of the homogeneity of a point process, often as a precursor to more advanced analysis. In general, rejection of the null hypothesis of homogeneity may imply a requirement for further analysis. In remote sensing for minefield detection, for example, homogeneity may correspond to the 'no minefield' case while regions of nonhomogeneity warrant closer inspection. This paper considers a version of the spatial scan process which uses stochastic and disjoint scan regions. The associated test for nonhomogeneity has the potential for improved power over conventional alternative sin applications where the point process is embedded in a general random field. Specifically, when the locations of any subregions of nonhomogeneity in the point process correspond to regions in the underlying field which can be segmented as distinct from their surroundings, the test derived here is recommended. The application to the detection of point clusters in gray-scale imagery, particularly minefields in multispectral imagery, is investigated.

Mines and mine-like objects are distributed throughout an area of interest. Remote sensing of the area form an aircraft yields image data that represent the superposition of electromagnetic emissions from the mines and mine-like objects. In this article we build a hierarchical statistical model for the reconstruction of mien locations given a point pattern of the superposition of mines and mine-like objects. It is shown how inference on the mine locations can be obtained using Markov chain Monte Carlo methods.

The performance of the likelihood ratio test is considered for point processes with piecewise constant interaction functions, and point processes featuring a reduced number of isolated-points. A Poisson approximation of the sufficient statistics is used to evaluate the performance. The approximations are compared with empirical results generated using Markov chain Monte Carlo simulation.

A monostatic amplitude and phase stepped-frequency radar approach have been proposed to detect small non-metallic buried anti-personnel (AP) mines. An M-56 AP-mine with a diameter of 54 mm and height of 40 mm, only, has been successfully detected and located in addition to small metallic mine-shaped objects. 2D probe-correction and addition signal processing are applied to the raw probe- data. The probe used in this experiment was an open-ended waveguide operating at S-band. The movements of the probe are controlled by two stepmotors via an RS-232 interface. The probe is connected to an HP8753C Network Analyzer through a 5 m long Sucoflex coaxial cable. The data are collected automatically using an HPIB interface. The collected data contains both the amplitude and phase information of the reflection coefficient.Data are measured at up to a maximum of 401 different frequencies at each measurement point using a mesh-grid with a resolution down to 1 mm by 1 mm. The size of the scan area is 1410 mm by 210 mm. Measurements have been performed on loamy soil containing a buried M-56, a non-metallic AP-mine, and various other mine-like objects made of solid plastic, brass, aluminum, steel, and wood. The presented results are based on probe-data measured at 100 different frequencies at each measurement point and a coarser mesh-grid of 10 mm by 10 mm, since it is found that less probe-data is needed. Our experiments show that even less amount of probe-data may be necessary.

United Nations Peacekeeping forces around the world need to transport food, personnel and medical supplies through disputed regions were land mines are in active use as road blocks and terror weapons. A method of fast, effective land mine detection is needed to combat this threat to road transport. The technique must operate from a vehicle travelling at a reasonable velocity and give warning far enough ahead for the vehicle to stop in time to avoid the land mine. There is particular interest in detecting low- metallic content land mines. One possible solutionis the use of ultra-wide-band (UWB) radar. The Australian Defence Department is investigating the feasibility of using UWB radar for land mine detection from a vehicle. A 3 GHz UWB system has been used to collect target response from a series of inert land mines and mine-like objects placed on the ground and buried in the ground. The targets measured were a subset of those in the target set described in Wong et al. with the addition of inert land mines corresponding to some of the surrogate targets in this set. The results are encouraging for the detection of metallic land mines and the larger non-metallic land mines. Smaller low-metallic- content anti-personnel land mines are less likely to be detected.

The Technical Cooperation Program subgroup K Action Group 23 is engaged in collaborative ultra-wideband (UWB) radar land mien detection experiments using instruments located in Australia, Canada, the UK and the US. In order to compare results, it is essential to develop standard characterization techniques to provide an improved basis for comparison of data and radar techniques. This paper discusses some suggested standard soil characterization techniques and standardized targets for UWB GPR mine detection experiments. For land mine detection using ground penetrating radar, it is the electrical characterization of soil over the relevant frequency range that is important. Some results of analysis of soil from test sites are given and discussed. Standardized targets are needed to allow reliable comparison of independently conducted experiments. Surrogate targets for this purpose need to be safely handled and deployed, sufficiently realistic to allow estimation of mine detection performance, and easily transported across national boundaries. It is desirable to have simple targets to allow comparison of numerical models with experiment. The design of a limited target set is discussed and sample signature measurements are presented in this and companion papers.

Ultra-wideband returns from surface-scattered and buried land mines give distinctive echoes that depend on the target type and soil environment. These echoes remain relatively stable for land mines with non-metallic casings but vary with the soil environment for land mines with metallic casings. This fact indicates that the signature resulting from the non-metallic-cased land mine is due to internal structure, whereas the signature of the metallic-cased target is due to interaction of the illuminating pulse with the metal case and the surrounding medium. The shape of these echoes can be described by a series of damped sinusoids. Certain parameters of the functions are believed to be stable for the same target in a variety of environments. Thus the echo can theoretically be used to identify targets with non-metallic casings by determining the position of the fixed poles that describe the exponential dampled sinusoids associated with a particular target type. This paper examines extraction of these poles, by the Prony method, for different types of non-metallic land mines and land-mine-like targets.

Land contamination by unexploded mines is a serous problem in many countries. Although a large portion of these mines is plastic, the problem caused by unexploded metallic mines is serious enough to warrant a cleanup effort. These metallic mines may be detected using either a proximity or a standoff sensor. This article discusses the detection of metallic miens using a standoff ultra-wideband SAR sensor. In particular, it analyzes the expected RCS of metallic mines, and the T/C as functions of frequency, depression angle, and soil. The bases of the analysis are theoretical RCS models, an empirical clutter model of bare soil, and limited soil and ground penetrating SAR measurements. The analysis indicates that the SAR should operate in the UHF band, and that the preferred environments are dry soil and possibly frozen soil and snow. Under these conditions, the results indicate that the detection of surface metallic antitank mines should be robust, that the detection of buried metallic antitank mines is viable, and that the detection of metallic anti-personnel mines may be viable with ultrahigh resolution SAR. The false alarm issue is not addressed in this article.

Several mine detection systems are currently under development which will provide airborne mine detection capabilities. For example, the COBRA program utilizes a multispectral video camera which will provide an interim, clear weather, daylight capability when deployed in a Pioneer unmanned aerial vehicle (UAV). The ASTAMIDS program has a dual approach, one contains an active polarized source as well as a passive IR camera and the other only a passive IR imager. Either ASTAMIDS system will provide day/night and limited visibility operation. The use of a PMMW imaging sensor promises to provide day/night and al weather mine detection performance. Attenuation in the MMW regime is not dramatically effected by adverse weather. In addition there is a large contrast between metal targets and the background for air to ground scenarios. Furthermore, due to the long wavelengths, vegetation and soils are not completely opaque in the MMW regime, offering the possibility to detect buried targets under specific conditions. This paper will describe the assessment of an imaging PMMW sensor for mine detection. The results of data collection and modeling analysis will be presented as evidence to the utility and capabilities of the technology to perform under adverse weather conditions.

A review of physical properties of passive sensors working in IR and millimetric wavelength points out main advantages and drawbacks of each technology. To verify and to assess qualitatively the behavior of senors working in these spectral bands, we have collected data with 3 IR sensors and a passive millimeter wave (PmmW) imager. Priority was given to plastic AT mines and blast AP mines laid or buried in sand, compost or in grass. With the MIR and the FIR sensors used, most of the plastic mines were detectable when laid and unobscured. With the PmmW sensor the detection of laid mine is improved, compare to IR, and the number of false- alarms is reduced, but because of the better spatial resolution of IR sensors their classification capability is better than with the PmmW imager. Each technology can take benefits from the fusion or combination of IR and PmmW sensors. An attempt of data fusion at a pixel level is presented.

Different types of anti-tank surface mines were measured on a turntable in a n anechoic chamber, and modelled with the German SIMPRASS model. In addition, several flights over a realistic minefield were performed with a 35 GHz and a 94 GHz radar operated in parallel in a SAR configuration. It is found that typical RCS values of AT mines at mmW frequencies are in the region between -18dBsm and -30dBsm. Therefore, they are detectable by airborne mmW radar only for a resolution cell size comparable to the physical size of the mine itself depending on the background clutter type.

At Sweden's Defence Research Establishment, FOA, a hand-held mine detection device is under development. The system is based on impulse radar technique, which due to its large band width, has shown to be an effective means to find objects buried shallow in the ground. Working with radar technique gives an obvious advantage compared to e.g., metal detectors when searching for plastic mines, or when the search is performed in an area highly contaminated with metal fragments or when the soil itself is rich of ferrite. The paper concentrates on detection and classification of minelike objects from measurements in an indoor testing environment. The focus is on evaluating how methods, partly already successfully proven, work in a 'difficult' environment, namely extremely dry sand. The result shows that metal objects and a stone that was used as object, are fairly easy to detect and to classify correct. The fact that the stone is classifiable, means that the false alarm rate can be reduced. It is also possible to detect a nylon cylinder, but here is the result quite sensitive to parameters of the detection algorithm. This is due to that the permittivity of the sand in the experiment is extremely low and close to the one for nylon. For the same reason, a non-metallic AP mine is not detectable or classifiable in the dry sand. The results indicate that even thought he methods work in more normal environments, other detection and classification algorithms than the presented ones have to be used in extreme cases like this one, in order ensure the function of the device.

Problems of the detection of conductive and dielectric targets buried below the surface with active-mode microwave radiometer are discussed. The broadband noise source is used to increase the subsurface detection possibilities of microwave radiometry at active mode. Additionally the bifocal receiving-emitting antenna is used for the increasing of the spatial resolution. The microwave radiometer is designed for measurements of reflection coefficient distribution of searched subsurface part and for synthesis of the radio images. Some of the proposed approaches have been used for detection of anti-infantry mines buried in the sand. The results of laboratory trials are presented. Similar technology was used by us for searching objects like guns, knives, plastic elements, etc. hidden under clothing.

Extending earlier successful radiometric detection of both metal and plastic mines, both buried and unburied at 44 and 12 GHz, a lower frequency radiometer was tested to determine its effectiveness under soil conditions of higher water content. In addition, the ability to generate 2D images was evaluated using a novel image processing technique, and the results were compared to model predictions.

We describe herein the construction of a simple, low-power, broadly responsive vapor sensor. Carbon black-organic polymer composites have been shown to swell reversibly upon exposure to vapors. Thin films of carbon black-organic polymer composites have been deposited across two metallic leads, with swelling-induced resistance changes of the films signaling the presence of vapors. To identify and classify vapors, arrays of such vapor-sensing elements have been constructed, with each element containing the same carbon black conducting phase but a different organic polymer as the insulating phase. The differing gas-solid partition coefficients for the various polymers of the sensor array produce a pattern of resistance changes that can be sued to classify vapors and vapor mixtures. This type of sensor array has been shown to resolve common organic solvents, including molecules of different classes as well as those within a particular class.

We report on a new methodology for chemical detection of explosive particles associated with anti-personnel mines. Trace particle detection can be used to complement vapor detection of explosives with an electronic noise. Our approach is to remotely stimulate a target area 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 designed 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 micrometers wavelength IR light. Explosive particles will selectively absorb the IR energy at approximately 1600 cm^-1, decompose, and give off heat which can be detected.

Microelectromechanical Systems (MEMS) is the integration of actuators, sensors, and electronics onto a silicon substrate through the utilization of microfabrication technology. MEMS is an enabling technology, in that it allows the realization of smart devices and systems by augmenting the computational ability of electronics with the perception and control capabilities of microsensors and microactuators. In the most general form, the sensors gather information from the environment, the electronics analyze the sensor information and direct the actuators to control or manipulate the environment for some desired outcome or purpose. The benefits of MEMS technology for realization of demining sensors include: high sensitivity, low-cost due to batch fabrication; high accuracy and reliability, small size and weight; low power consumption; high levels of functionality; and the ability to integrate on-chip electronics. Combining MEMS sensors onto low-cost robotic platforms which can autonomously scan large areas for mines without placing humans at risk is an attractive approach to the unexploded ordinance problem.

Research on demining includes many different aspects, and in particular the design of efficient and intelligent strategies for (1) determining regions of interest using a variety of sensors, (2) detecting and classifying mines, and (3) searching for mines by autonomous agents. This paper discusses strategies for directing autonomous search based on spatio-temporal distributions. We discuss a model for search assuming that the environment is static, except for the effect of identifying mine locations. Algorithms are designed and compared for autonomously directing a robot, in the case where a single search engine carrying a single sensor.

Time-domain scattering from a fixed target in the presence of a rough surface is considered, with application to ground penetrating radar. The time-domain scattering data are computed via a 2D finite-difference time-domain algorithm. In addition to examining the statistics of the time-domain fields scattered from such a surface, we investigate subsurface-target detection by employing a matched-filter detector. Such a detector is characterized by its receiver operating characteristics, which quantifies the probability of detection and probability of false alarm. Plane-wave incidence is considered, with emphasis on low-grazing-angle backscattering.

The fundamental goal of mine detection is to achieve a high detection rate along with a low false alarm rate. While many mine detectors achieve the first of these goals, it is often at the cost of a prohibitively large false alarm rate. In this paper, a Bayesian decision-theoretic approach to the detection of mines, which incorporates the physical properties of the target response to an electromagnetic induction device, is presented. This approach merges physical modeling of the evoked target response with a probabilistic description that represents uncertainty in the ground surface, composition of the mine, and its placement in the surrounding environment. This approach provides both an optimal detection scheme, and performance evaluation measures in the form of probability of detection and false alarm rate. We present results in which the model-based, Bayesian approach significantly outperforms the energy detector and matched filter detectors on data obtained from the DARPA backgrounds clutter data collection experiment. In addition, the model-based, Bayesian approach is also shown to outperform a detector which estimates the eddy-current decay rate from the data. Results are also presented to illustrate the amount of sensitivity of the matched filter detector for a known environment to incorrect prior knowledge of uncertain parameters in the demining scenario, as well as the robustness of performance and performance bounds realizable by the optimum detection algorithm that properly accounts for uncertainty within a Bayesian framework.

The airborne standoff minefield detection systems (ASTAMIDS) is an airborne imaging system designed for deployment as a modular mission payload on an unmanned aerial vehicle and capable of detecting surface and buried anti-tank mines under all-weather, day/night conditions. Its primary mission is to support a forward maneuver unit with real time intelligence regarding the existence and extent of minefields in their operational area. The ASTAMIDS development effort currently consists of two parallel technical approaches, passive thermal IR sensor technology in one case and an active multi-channel sensor utilizing passive thermal IR coregistered with a near IR laser polarization data for the other case. The minefield detection capability of this system is a result of signal processing of image data. Due to the large quantities of data generated by an imaging sensor even at modest speeds of an unmanned aerial vehicle, manual exploitation of this data is not feasible in a real time tactical environment and therefore computer aided target feature extraction is a necessity to provide detection cues. Our development efforts over the past several years have concentrated on mine and minefield detection algorithms, the hardware necessary to execute these algorithms in real time, and the tools with which to measure detection performance.

Experiments were conducted to estimate the probability of detection (Pd) and false alarm rate (FAR) for detection of buried mines using a casi hyperspectral imager scanned from a personnel lift.Surrogate mines and blocks of explosives were buried under various vegetative covers and bare soil using standard mine laying methods. Images were obtained at various times from 1 2/3 to 15 1/2 months after burial. Mines under short and medium length vegetation and bare soil were detectable by calculating the linear correlation coefficient or fractional composition, derived from orthogonal subspace projection, for each pixel of the image. Pd for surrogates decreased with increasing vegetation length and appeared to vary somewhat with environmental conditions. Short vegetation gave better detection results than bare soil which gave better results than medium length vegetation. Pd was typically in the range of 55 to 94 percent and FAR varied rom about 0.17 to 0.52 m^-2. There were insufficient data to assess the relative importance of stress due to explosive vapor leaching into the soil compared to stress induced by soil disturbance.

This paper describes design and characteristics of an underwater laser imaging systems (ULIS) developed for detection of underwater mines and mine-like targets. The ULIS uses a diode pumped Nd:YLF laser, frequency doubled to 523.5 nm as a transmitter and an image intensifier tube (IIT) as a receiver. The transmitter uses an innovative technique of frequency stabilization to reduce laser line- width so that the receiver can use a less than 1 nm narrow band pass filter. This results in a significant reductio of the back-ground noise in the receiver. Also, the IIT based receiver is range gated to reduce back-scatter noise. A polygon scanner is used in the ULIS with the transmitter and the receiver to obtained range resolved imagery. Description of the system components and performance characteristics of the system as ladar are provided.

IR imagers are being investigated by several groups for use in landmine detection. The ability to predict detection performance is necessary to establish confidence for single sensor systems or to allow appropriate weighting of detector output for data fusion algorithms in multiple sensor systems. Preliminary studies had shown that the in-ground vertical temperature gradient was a good indicator of mine/background contrast in IR images if temperature measurements and imager wee collocated and limited data suggested that remote performance monitoring might be possible. To establish practicality of remote monitoring, temperature probes were buried at 5 sites separated by various distance between 30 m and 5.8 km, in asphalt, sand and gravel, both on and off road. Vertical temperature profiles were automatically recorded at all sites simultaneously with IR images of buried thermal IR surrogate mines located at a gravel road site. The in-ground vertical temperature gradient was confirmed to be a practical indicator of the performance of an IR imager, for probes buried in all materials at distances up to almost 6 km from the imager. A five element probe with thermocouples uniformly placed at depths from -3 to -11 cm would be sufficient to predict detection performance.

Noise and sensitivity measurements are compared for five intensified cameras used for airborne sensing of mines. Individual camera variations are discussed, as well as global similarities useful in camera modeling. The cameras generally have behaviors that closely match each other, but differ in sensitivity by factors of up to 12 for equal noise. The implication is that these intensified cameras-being short production-run items-display a great deal of individuality and that each camera must be carefully characterized before use. Reciprocity (product of radiance and exposure) and the variation of noise with radiance were also tested and found to be flat over three orders of magnitude, thus simplifying modeling.

This paper gives brief overview of the signal processing problem about image and video coding using new types of discrete orthogonal transforms whose matrices are described with finite group character tables. The new effective algorithm of digital information compression based on discrete transform and convolutions on finite algebraic structures have been developed for reducing temporal and energy expenditures, increasing precision and processing reliability of large information volumes in the computer systems.

This paper presents findings based on the IR thermography testing of the solar heated, quasi-natural fields with four anti-tank miens, buried at various depths. Experiment has been carried out last year in different seasons, weather, time and terrain conditions. This baseline data was collected for predicative needs based on modeling and to determine the level of time-dependent adaptability required in the detection algorithms suite. In a single and raw thermogram it was possible to identify mine with a naked eye only in very carefully chosen, rather rare, circumstances. We stated that there is an interesting 'new' kind of IR signatures of mines 'their dynamic phenomena in time'. The obtained contrasts were often similar level as the ones obtained in industrial thermal nondestructive testing, based on the time-domain treatment method. Results show that distinguishing the mines based on time sequence of the IR images is more reliable than on single thermogram, despite the fact that used procedure is much more complicated. Time- domain treatment method based on IR thermography seems to be the potential for further improvement of the localization of the buried mines especially when covered by grass and the like. At present state of the IRT, RPV and GPS technologies, the same as increasing of the signal processing possibilities, technical realization of such idea can be taken into consideration.

Detecting and identifying targets immersed in draught zone are necessary for a safe navigation. But with today technologies it is very difficult to do it from a ship. Nowadays, this detection still stays a challenge, because sonars are inefficient in the zone just under the waterline. In our laboratory we succeeded in using lidar as a complementary tool of the sonar. With optical detection, active imagery, various immersed targets were detected, identified and classified according to a Mueller matrix and whatever the perturbations on the sea surface and the wind conditions. The possibility of using a particular polarimetric codage with optical signal is discussed in this paper. The aim was to demonstrate that our lidar tool can be improved by choosing the best polarimetric couple in emission and detection.

The method of minimal embedding dimension definition of the chaotic attractor on the basis of topological structure dynamics analysis of phase trajectories is proposed. It is shown that the suggested method yields reduction of experimental data quantity and computer resources about an order in comparison with traditional methods. This reduction is achieved due to localizing topological analysis of the phase trajectories. The proposed method can be used of investigation of nonlinear dynamical systems with chaotic behavior in technique, biology, medicine.

An investigation was recently undertaken to determine the applicability of thermal imaging to remote airfield assessment. Thermal images were acquired with an IR scanning radiometer having a minimum detectable temperature difference of 0.1 degree centigrade and a spectral bandpass of 3-12 microns. Areas of a test runway, at Wright Laboratory, Tyndall AFB, FL, were prepared to assess the capability of detection for small cracks and subsurface voids. We wee able to detect voids that had been excavated beneath six inches of asphalt, millimeter width cracks, holes and other surface anomalies. During periods of diurnal heating and cooling,the surface anomalies radiate more energy than the bulk material and are easily detectable in the thermal images. We report on the image contrast ratios as a function of temperature throughout the diurnal period.

We introduce two statistical models designed to detect discrete objects in sidescan SONAR which consider complimentary approaches to the problem. The first considers a complex textural model for the objects and implements detection through a dual hypothesis on texture class presence, while the second implements a complex Gibbs field model of the image and utilizes prior knowledge of typical object morphologies to support its detection rate. The models are demonstrated on examples of different seabed sediments and object types, and are shown to be reliable in operation. The common theme of the two models is use of spatial context in analysis, which, we argue, is a very powerful technique for improving the flexibility and reliability of any analysis system.

A continuously operating, scanning x-ray machine is being developed or landmine detection using backscattered x-rays. The source operates at 130 kV and 650 mA. The x-rays are formed by electrons striking a high Z target. Target shape is an approximate 5 cm wide by 210 cm long racetrack. The electron beam is scanned across this target with electromagnets. There are 105, 1-cm by 1-cm collimators in each leg of the racetrack for a total of 210 collimators. The source is moved in the forward direction at 3 mi/h. The forward velocity and collimator spacing are such that a grid of collimated x-rays are projected at normal incidence to the soil.T he spacing between the collimators and the ground results in a 2-cm by 2-cm x-ray pixel on the ground. A unique detector arrangement of collimated and uncollimated detectors allows surface features to be recognized and removed, leaving an image of a buried landmine. Another detector monitors the uncollimated x-ray output and is used to normalize the source output. The mine detector is being prepared for an advanced technology demonstration (ATD). The ATD is scheduled for midyear of 1998. The results of the source performance in pre ATD tests will be presented.

The use of backscattered x-rays to image buried land miens and distinguish between surface and buried features has been well documented. Laboratory imaging experiments, being conducted at Sandia National Laboratories/New Mexico, have been used to develop preliminary data acquisition hardware and software for an upcoming advanced technology demonstration (ATD). In addition image processing techniques, developed by the Department of Nuclear Engineering at the University of Florida, are utilized. Previous buried land mine imaging studies focused on antitank mines buried in screened sand and have included well defined surface features such as a board or a small diameter rock. In the present study we have examined imaging under a variety of practical environmental conditions. We have successfully imaged antitank miens (ATM) buried in sand and rocky New Mexico soil. Images have been obtained for bare surfaces as well as four surfaces covered with limestone road coarse base, snow, water, and native grass. In addition, we have imaged buried ATM and surface antipersonnel mines covered with debris consisting of various size rocks, a log, and leaves such that no mine was visible to the eye. Contour plots of the images obtained for the various environmental conditions are presented.

A new, monostatic, broadband, electromagnetic sensor, known as the GEM-3M, has been prototyped and tested at various sites containing buried unexploded ordnance and mines. The instrument consists of a pair of concentric, circular coils that transmit a continuous, broadband, digitally controlled, electromagnetic waveform. The two transmitter coils, with precisely computed dimensions and placement, create a zone of magnetic cavity at the center of the two coils. A third receiving coil is placed within this magnetic cavity such that it senses only the weak, secondary field returned from the earth and buried target(s). The mine detector is designed for ordinary soldiers, and, when completed, will have the following unique features: (1) one-man portable high-tech mine detector: intelligent, realtime data interpretation and display on a color LCD screen; (2) targets can be metallic mines, nonmetallic mines, or a disturbed soil condition; (3) detection and characterization based on visual color displays, rather than the tonal changes common to current metal detectors; (4) broadband; selectable frequencies for depth scanning suitable for regional soil conditions and geology, and; (5) all sensor coils housed in a light, circular disk - very similar in appearance to a conventional metal detector.

Automated target recognition (ATh) algorithm performance is often degraded by the propagation of sensor noise and input error through an algorithm's computational operations. In lieu of a detailed error analysis, algorithm perfonnance estimates are usually couched in terms of probability of hits (Ph) or rate of false alarms (Rfa). Although such measures may be useful to an algorithm's end users, they may not be meaningful mathematically to a designer or theoretician. That is, deficits in nonstandardized measures such as Ph and Rfa cannot usually be related analytically, in closed form, to noise and error at an algorithm's input. In Part 1 of this series of two papers [1], we discussed requirements for theory and software systems that support automated error analysis. We showed how the dataflow graph of an algorithm can be relabelled with error functions of each constituent operation. The resultant dataflow graph describes the error function of the algorithm. In this paper, we demonstrate the application of our novel error estimation software (called ERSIA -- ErrorReduction and Synthesis in Image Algorithms) to several practical image/signal processing (ISP) and ATh algorithms that were successfully employed in the detection of small targets in cluttered multispectral imagery [2]. For example, we analyze (a) the Adaptive Double-Gated Filter, (b) a statistical texture detector, and (c) gradient-based edge-enhancement algorithms. By perturbing the input with various types and degrees of noise (i.e., Gaussian, Lorentzian, and Poisson distributions), we show that ERSIA can accurately predict the output noise in each algorithm, and that these predictions agree with theory. We analyze the computational cost associated with each algorithm and with the automated error analysis, in terms of sequential (von Neumann) and SIMD-parallel computational paradigms. We also discuss the computational cost associated with the application of ERSIA to a given algorithm. Occasionally, this cost can be high, since ERSIA can be programmed to compute a two-dimensional error distribution for each image operand and scalar variable. Planned implementations of ERSIA on SIMD- and MIMD-parallel processors are discussed in terms of computational accuracy, cost, and portability.

The purpose of this paper is to present an application of global optimization techniques to the problem of the enhancement of images. After reviewing the general strategy which is common to most modern enhancement method based on the solution of a variational problem, we explain the rationale behind the choice of our penalization formulation and we give the details of the practical implementation. Finally, we illustrate the efficiency of our approach by presenting the results obtained in preprocessing side-scan sonar images of the sea-bottom for the purpose of mine detection.

Discrimination problems differ in nature from reconstruction tasks. While in reconstruction, it is the mean squared error that is often used to measure the quality of the scheme, classification requires a different measure which often is not related to the former. The discrimination power of a certain basis or a set of basis function is not necessarily connected to the quality of reconstruction associated with this set. Furthermore, the degree of relevance of the orthonormality constraint to the quality of the discrimination is questionable. For example, linear discriminant analysis [1] searches for linear projections which maximize the between-class variance divided by the sum of within-class variance. Such projections do not necessarily coincide with the principal components of the data which are the directions that optimize MSE reconstruction. A successful approach to discrimination is based on an appropriate preprocessing to create an efficient signal representation, which then leads to an efficient dimensionality reduction. The next step is again some combination of feature extraction and classification. In this paper we briefly review several methods for finding data representation via optimal decomposition of wavelet basis functions and discuss their reconstruction properties. We then discuss some signal decomposition methods for the purpose of discrimination. This is followed by a brief discussion on a combination of feature extraction and classification scheme and with discrimination results on two acoustic data-sets.