Abstract: DARPA/Air Force Research Laboratory Moving and Stationary Target Acquisition and Recognition (MSTAR) program is developing state-of-the-art model based vision approach to Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR). The model-based approach requires using off-line developed target models in an on-line hypothesize-and-test manner to compare predicted target signatures with image data and output target reports. Central to this model-based ATR is the PEMS (Predict-Extract-Match-Search) subsystem. The Search module is critical to PEMS by providing intelligent control to traverse the hypothesis feature space. A major MSTAR goal is to demonstrate robust ATR for variations in targets including partially hidden targets. This paper will provide an update on the technology being developed under MSTAR and the status of this model based ATR research, specifically concentrating on the Search Module.

Abstract: A modular neural network classifier has been applied to the problem of automatic target recognition using forward-looking infrared (FLIR) imagery. The classifier consists of several independently trained neural networks. Each neural network makes a decision based on local features extracted from a specific portion of a target image. The classification decisions of the individual networks are combined to determine the final classification. Experiments show that decomposition of the input features results in performance superior to a fully connected network in terms of both network complexity and probability of classification. Performance of the classifier is further improved by the use of multiresolution features and by the introduction of a higher level neural network on the top of the individual networks, a method known as stacked generalization. In addition to feature decomposition, we implemented a data-decomposition classifier network and demonstrated improved performance. Experimental results are reported on a large set of real FLIR images.

Abstract: Texture measures offer a means of detecting targets in back- ground clutter that has similar spectral characteristics. Our previous studies demonstrated that the ''fractal signature'' (a feature set based on the fractal surface area function) is very accurate and robust for gray- scale texture classification. This paper introduces a new multichannel texture model that characterizes patterns as 2-D functions in a Besov space. The wavelet-based fractal signature generates an n-dimensional surface, which is used for classification. Results of some experimental studies are presented demonstrating the usefulness of this texture mea- sure. © 1998 Society of Photo-Optical Instrumentation Engineers. (S0091-3286(98)01001-0)

Abstract: The MACH and DCCF correlation filter algorithms are evaluated using the publicly released MSTAR data base. These algorithms can be used as a matching engine for automatic target recognition in SAR imagery. In practice, the required filters can be synthesized using model based signature predictions. In addition, the MACH and DCCF algorithms are optimized to be robust to variations (distortions) in the target's signature. Unlike Matched Filtering or other exhaustive template based methods, the proposed approach requires very few filters. The paper describes the theory of the algorithm, key practical advantages and details of test results on the public MSTAR data base.

Abstract: A new algorithm is presented for automatic target recognition (ATR) where the templates are obtained via singular value decomposition (SVD) of high range resolution (HRR) profiles. SVD analysis of a large class of HRR data reveals that the range-space eigenvectors corresponding to the largest singular value accounts for more than 90% of the target energy. Hence, it is proposed that the range-space eigenvectors be used as templates for classification. The effectiveness of data normalization and Gaussianization of profile data for improved classification performance is also studied. With extensive simulation studies it is shown that the proposed eigen-template based ATR approach provides consistent superior performance with the recognition rate reaching 99.5% for the four class XPATCH database.

Abstract: A modular neural network classifier has been applied to the problem of automatic target recognition using forward- looking infrared (FLIR) imagery. The classifier consists of several independently trained neural networks operating on features extracted from a local portion of a target image. The classification decisions of the individual networks are combined to determine the final classification. Experiments show that decomposition of the input features results in performance superior to a fully connected network in terms of both network complexity and probability of classification. Performance of the classifier is further improved by the use of multi-resolution features and by the introduction of a higher level neural network on the top of expert networks, a method known as stacked generalization. In addition to feature decomposition, we implemented a data decomposition classifier network and demonstrated improved performance. Experimental results are reported on a large set of FLIR images.© (1998) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: The MIT Lincoln Laboratory has developed the ATR (automatic target recognition) system for the DARPA-sponsored SAIP program; the baseline ATR system recognizes 10 GOB (ground order of battle) targets; the enhanced version of SAIP requires the ATR system to recognize 20 GOB targets. This paper compares ATR performance results for 10- and 20-target MSE (mean-squared error) classifiers using medium-resolution SAR (synthetic aperture radar) imagery.

Abstract: A pulsed laser radar (ladar) based object recognition system with applications to automatic target recognition is reported. The approach used is to fit the sensed range images to the range templates extracted using laser physics based simulation of Computer Aided Design target models. A projection based pre-screener filters out more than 80 percent of candidate templates. An M of N pixel matching scheme for internal shape matching combined with a silhouette matching scheme is used for recognition. The system has been blind tested on a data set containing 276 real ladar images of military vehicles at various orientations and different ranges. The system achieves above 90 percent accuracy in recognition of 0.4 meters resolution ladar images.

Abstract: Apparatus for automatic recognition of a target object from an infrared or visible light image, includes primary segmentation and classification means (4, 8, 9, 10, 11, 12) in which objects are either recognised or unrecognised and one or more secondary segmentation means (15, 21, 22, 23, 12) applicable to primary segmented image areas in which objects which are unrecognised by the primary segmentation means are further classified and either recognised or rejected.

Abstract: A multiprocessor version of the ORASIS hyperspectral analysis program has been implemented in support of the ASRP. In brief, the long-term technical objectives of the ASRP are to demonstrate the feasibility and military utility of real-time target detection from uncrewed air vehicles using hyperspectral data. This paper presents a preliminary assessment of ORASIS performance and describes the ORASIS development effort designed to meet the ASRP goals. Real-time performance of the analysis program and its potential effectiveness as a target detection method are demonstrated.

Abstract: Under the Mojave configurable computing project, we have developed a system for achieving high performance on an automatic target recognition (ATR) application through the use of configurable computing technology. The ATR system studied here involves real-time image acquisition from a synthetic aperture radar (SAR). SAR images exhibit statistical properties which can be used to improve system performance. In this paper, the Mojave configurable computing system uses field programmable gate arrays (FPGA's) to implement highly specialized circuits while retaining the flexibility of programmable components. A controller sequences through a set of specialized circuits in response to real-time events. Computer-aided design (CAD) tools have been developed to support the automatic generation of these specialized circuits. The resulting configurable computing system achieves a significant performance advantage over the existing solution, which is based on application specific integrated circuit (ASIC) technology.

Abstract: This paper describes progress on an Automatic Target Recognition (ATR) system for Synthetic Aperture Radar(SAR) imagery. The system is based upon a feature extraction, data ordering, and statistical modeling paradigm.Feature extraction is performed by applying image segmentation to convert the SAR imagery into one of four pixelclasses. A description of a real-time image segmentation design is given. The segmented imagery is re-ordered froma two dimensional (2D) spatial representation to a sequential representation through the use of multiple Radon'&ansforms (RT). Finally, the re-ordered data is classified by target type by appling Hidden Markov Model (HMM)decoding techniques. Performance results on the MSTAR public targets database is provided.Keywords: Hidden Markov Modeling (HMM), Target Recognition, Image Segmentation, Synthetic Aperture Radar (SAR) 1. INTRODUCTION This paper contains a description of a target recognition system for SAR imagery. The conceptual design of thesystem was made with three primary goals in mind. First, the recognition performance should be robust to variationsin target appearance and environmental conditions. Second, the system processing response time should be capableof "real-time" operation. Third, the implementation should result in a system which is power efficient, light weight,

Abstract: Orientation estimation is an essential step in autonomous navigation, automatic target recognition, part handling, and many other machine vision applications. A procedure is presented to estimate the orientation of the target using spectral energy distribution in the Fourier domain. The performance of this technique is determined by conducting simulation experiments on a set of infrared images containing military targets.

Abstract: Without successful adaptive multisensor fusion or online registration techniques, automatic target recognition (ATR) algorithms are prone to poor object classifications. Multisensor fusion for a given situation assessment includes identifying measurement information for task completion and reducing image uncertainty in the presence of clutter. By extracting synthetic aperture radar (SAR) image informational features, image registration and target classification is achievable. This paper examines SAR information-theoretic features for a target orientation and proposes a method for target classification.

Abstract: Most current artificial vision systems lack robustness and are applicable only to a narrow range of tasks. Crevier (1997) has suggested that this is due to their reliance on a small number of vision mechanisms and to lack of knowledge about how vision algorithms should be integrated. We suggest a systems approach to artificial vision based on computational vision research. The capabilities of biological vision systems are contrasted with those of current piecemeal approaches to artificial vision. A mature, comprehensive vision system, called the Georgia Tech Vision (GTV) simulation is described. GTV incorporates quasilinear filter mechanisms to simulate the processing performed by simple and complex cortical cells. The outputs of these mechanisms are adaptively combined to discriminate targets from clutter and/or one another. GTV outputs predictions of human search and detection performance and/or targeting metrics for automatic target recognition (ATR) applications. Studies validating GTV as a model of human search and detection performance and demonstrating its performance as an ATR are presented.

Abstract: This paper discusses the application of hidden Markov models (HMMs) to solve the translational and rotational invariant automatic target recognition (TRIATR) problem associated with SAR imagery. This approach is based on a cascade of these stages: preprocessing, feature extraction and selection, and classification. Preprocessing and feature extraction and selection involve successive applications of extraction operations from measurements of the Radon transform of target chips. The features which are invariant to changes in rotation, position and shifts, although not to changes in scale are optimized through the use of feature selection techniques. The classification stage successively takes as its inputs the multidimensional multiple observation sequences, parameterizes them statistically using continuous density models to capture target and background appearance variability, and thus results in the TRIATR-HMMs. Experimental results have demonstrated that the recognition rate is as high as 99% over both the training set and the testing set.

Abstract: This paper presents a method for detecting and classifying a target from its foveal (graded resolution) imagery using a multiresolution neural network. Target identification decisions are based on minimizing an energy function. This energy function is evaluated by comparing a candidate blob with a library of target models at several levels of resolution simultaneously available in the current foveal image. For this purpose, a concurrent (top-down-and-bottom-up) matching procedure is implemented via a novel multilayer Hopfield (1985) neural network. The associated energy function supports not only interactions between cells at the same resolution level, but also between sets of nodes at distinct resolution levels. This permits features at different resolution levels to corroborate or refute one another contributing to an efficient evaluation of potential matches. Gaze control, refoveation to more salient regions of the available image space, is implemented as a search for high resolution features which will disambiguate the candidate blob. Tests using real two-dimensional (2-D) objects and their simulated foveal imagery are provided.

Abstract: Conventional radar uses the Fourier transform to generate a radar target image. Constraints on the use of Fourier methods requiring point scatterers to remain in their range cells and requiring Doppler frequency shifts for point scatterers to be stationary are impractical due to a moving object's inherent non-uniform motion and rotation. Time varying motion-induced Doppler frequency shift spectra represented with Fourier transform methods smears radar target image. Representing time-varying Doppler spectrum using joint time-frequency transform methods is desirable. Replacing conventional radar Fourier transform with a high resolution time-frequency transform, a 2-D range-Doppler Fourier image frame becomes a 3-D time-range-Doppler cube. By sampling in time, a time sequence of 2-D range-Doppler images with superior resolution can be viewed. Smears from time-variance of Doppler spectrum may be removed to enhance target image.

Abstract: This paper discusses the application of Hidden Markov Models (HMMs) to the Automatic Target Recognition (TRI-ATR) problem in Synthetic Aperture Radar (SAR) imagery. Related research with applications of the HMMs to solve SAR Automatic Target Recognition (ATR) problems can also be found in Kottke et al.5 Our approach is based on acascade of three stages: preprocessing, feature extraction and selection, and classification. Preprocessing and featureextraction and selection involve operations performed on the Radon transform of target chips. The features, whichare invariant to changes in rotation, position and shifts, although not to changes in scale, are optimized throughthe use of feature selection techniques. The classification stage takes as its inputs the multidimensional multipleobservation sequences and parameterizes them statistically using continuous density models to capture the targetand background appearance variability. Experimental results have demonstrated that the recognition rate can be ashigh as 95% over both the training set and the testing set, in certain cases.Keywords: HMMs, ATR, SAR, invariant, Radon Transform, and rotation.

Abstract: High range resolution (HRR) radar is important for its all- weather, day/night, long standoff capability. Additionally, it is an excellent sensor for identifying moving ground targets because it produces high resolution target signatures and because targets can be separated from ground clutter using Doppler processing. Ongoing research under the System Oriented HRR Automatic Recognition Program has led to an increased understanding of the HRR data, the target separability, and a baseline assessment of target recognition algorithms using template based approaches.

Abstract: In this paper, we introduce a new set of image features for use in the discrimination algorithm of a baseline automatic target recognition (ATR) system. These new features are designed to capture the changes in spatial dispersion of the high-intensity pixels in the input image as the image is threshold at different intensity levels. We show that significantly better performance can be obtained when the new features are used in place of the baseline discrimination features. In particular, we demonstrate with a large set of high-resolution synthetic aperture radar imagery that, when the probability of detection is between 0.5 and 1.0, the false alarm density obtained using the new features is approximately 30 to 50 times lower than that obtained using the baseline features. For medium-resolution imagery, the false alarm density has been reduced by a factor of 3 to 5 using the new features.© (1998) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: : The goal of the research is to develop a multiple model extended Kalman filter to perform simultaneous target identification and tracking using high range resolution (HRR) radar measurements. The idea is to use a multiple model estimator (MME) with multiple Kalman filters (one for each target type) to process HRR measurements. The resulting algorithm will provide estimates of the target kinematics (e.g. position, velocity, and possibly acceleration) and predict model probabilities, which correspond to the target probability. Hence, the approach proposed here is named the automatic target recognition (ATR) and tracking filter.

Abstract: A statistical modeling paradigm for automatic machine recognition of speech uses mixtures of nongaussion statistical probability densities which provides improved recognition accuracy. Speech is modeled by building probability densities from functions of the form exp(−tα/2) for t≧0 and α>0. Mixture components are constructed from different univariate functions. The mixture model is used in a maximum likelihood model of speech data.

Abstract: We propose an enabling algorithm for automatic target detection wherein the targets are almost similar but differ in fine details. Multiple translated and scaled target images are processed in the Mellin transform domain and subsequently detected using a learning vector quantization (LVQ) neural network.

Abstract: Automatic target recognition (ATR) involves processing two-dimensional images for detecting, classifying, and tracking targets. The first stage in ATR is the detection process. This involves discrimination between target and non-target objects in a scene. We discuss a novel approach which addresses the target detection process. This method extracts relevant object features utilizing principal component analysis. These extracted features are then presented to a multi-stage neural network which allows an overall increase in detection rate, while decreasing the false positive alarm rate. We discuss the techniques involved and present some detection results that have been implemented on the multi-stage neural network.

Abstract: : A modular neural network classifier has been applied to the problem of automatic target recognition (ATR) of targets in forward-looking infrared (FLIR) imagery. The classifier consists of several independently trained neural networks operating on features extracted from a local portion of a target image. The classification decisions of the individual networks are combined to determine the final classification. Experiments show that decomposition of the input features results in performance superior to a fully connected network in terms of both network complexity and probability of classification. The classifier's performance is further improved by the use of multiresolution features and by the introduction of a higher level neural network on top of the expert networks, a method known as stacked generalization. In addition to feature decomposition, we implemented a data decomposition classifier network and demonstrated improved performance. Experimental results are reported on a large set of FLIR images.

Abstract: Detecting objects in images containing strong clutter is an important issue in a variety of applications such as medical imaging and automatic target recognition. Artificial neural networks are used as non-parametric pattern recognizers to cope with different problems due to their inherent ability to learn from training data. In this paper we propose a neural approach based on the Random Neural Network model (Gelenbe 1989, 1990, 1991, 1993), to detect shaped targets with the help of multiple neural networks whose outputs are combined for making decisions.

Abstract: The problem of combining multi-source information in applications related to automatic target recognition (ATR) is addressed. A mathematical approach is proposed for fusing the (possibly dependent) outputs of multiple ATR systems or algorithms. The method is derived from statistical principles, and the fused decision takes the form of an hypothesis test. The distribution of the test statistic is approximated as gamma, with parameters estimated from available training data. In a brief simulation study, the proposed method outperforms several alternative fusion techniques.