Abstract: We develop a two-stage, modular neural network classifier and apply it to an automatic target recognition problem The data are features extracted from infrared and TV images We discuss the problem of robust classification in terms of a family of decision surfaces, the members of which are functions of a set of global variables The global variables characterize how the feature space changes from one image to the next We obtain rapid training times and robust classification with this modular neural network approach

Abstract: Morphological processing operations are necessary for many difficult automatic target recognition (ATR) problems All of the basic morphological functions can be performed on an optical correlator This paper emphasizes the hit-or miss morphological transform and how it provides improved part recognition

Abstract: The application of artificial neural network technology to data fusion for target recognition is discussed. The specific application includes airborne target recognition primarily using information available from radar and EO-IR thermal imaging sensors. The artificial neural network based fusion architectures from alternative approaches are discussed, such as alternative learning algorithms, a training set, and massively parallel distributed processing for making real-time automatic target recognition decisions. The experimental results indicate that target recognition performance can be greatly enhanced by using an array of complementary sensors whose outputs are fused to extract information not otherwise available from a single sensor

Abstract: A method for modeling radar target scattering data for the purpose of automatic target recognition is considered. The approach is to estimate a time-domain feature vector which describes the target. The target is characterized by a set of scattering centers. Using the motion of a transient polarization response, scattering centers are estimated along with their polarization. An exponential model for the fully polarized radar return and estimation algorithm are presented

Abstract: Image processing to accomplish automatic recognition of military vehicles has promised increased weapons systems effectiveness and reduced timelines for a number of Depariment of Defense missions. Automatic Target Recognizers (ATh) are often claimed to be able to recognize many different types of vehicles as possible targets in military targeting applications. The targeting scenario conditions include different vehicle poses and histories as well as a variety of imaging geometries, intervening atmospheres, and background environments. Testing these ATh subsystems in most cases has been limited to a handful of the scenario conditions of interest, as is represented by imagery collected with the desired imaging sensor. The question naturally arises as to how robust the performance of the ATh is for all scenario conditions of interest, not just for the set of imagery upon which an algorithm was trained.© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: We are interested in training neural networks to recognize objects in images. An important part of this task is to make the overall system robust with respect to image variations, including rotation, scale, and translation. This paper addresses two major issues associated with feature extraction for this problem; Selection of meaningful features, and data compression. Selection of meaningful features: Even though neural networks are very effective and robust pattern classifiers, they have an important limitation; for any given application, we cannot always explain why they succeeded or why they failed. The black box'' nature of the neural networks makes it difficult to analyze their internal states. Also, their performance is highly dependent on the training data. Our approach, therefore, is to create feature vectors that contain information that is as meaningful as possible. This paper describes the use of Gabor representations to generate feature vectors that are robust to variations in rotation, scaling, and translation. We are also studying ways to make the system robust to variations in perspective, occlusion, contrast, noise, and background. Gabor filters when used as the receptive field in a hierarchical scheme for feature extraction, offer properties that make them promising for providing the desired robustness more » to image variations. Unlike other filters, Gabor filters are optimally localized in both the space domain and the frequency domain, so that their space-bandwidth product is minimum. It has been shown that their properties of spatial localization, orientation selectivity, and spatial frequency selectivity make Gabor filters a good model for biological vision. This paper describes experiments in which we demonstrate the ability of the overall recognition system to classify objects in simulated scenes, even though they have undergone variations in orientation scale, and position. « less

Abstract: A brief review is presented of neural network tools for Automatic Target Recognition (ATR) . These tools include collective computation for implementing a variety of computational-vision techniques learning and adaptation for pattern recognition knowledge integration for expert-system capabilities and beyondsupercomputer- level hardware. As a specific example neural networks for stereo vision are introduced as a potentially fruitful approach to ATR. Preliminary results are presented which show substantial performance improvements over previous stereo algorithms for producing accurate dense displacement maps. These maps can be used in turn to derive accurate geometrical shape information that can result in improved recognition performance. 1.© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: A three unit artificial neural network (ANN) automatic target recognition (ATR) system is integrated within, and compared to, a recently AFIT developed conventional ATR system. The integration of ANN within this existing framework allows the determination of where the benefits of using these biologically motivated processing techniques lie. The integration and testing of ANN within each of the three units constitutes the major contribution of this research. The emphasis of this paper is in the area of effects of learning alternatives on ATR. Several alternative feedforward networks were compared in the classifier unit.

Abstract: In order to take a promising Automatic Target Recognition (ATR) algorithm or system from the development stage to the point where it is robust enough for the application which it is intended for, a considerable amount of testing, training, and evaluation is required. This can take years. For example with seemingly slight distortion in the target/clutter characteristics, the algorithm behavior becomes erratic and random. In addition, no well defined method currently exists for designing image processing algorithms. Sophisticated and intelligent tools are needed in order to perform such a task in a rapid & efficient manner. A prototype system has been developed which uses a blackboard architecture and allows rapid mapping of procedural algorithms into a knowledge based development environment. This system was developed at Honeywell Systems and Research Center. Coupled with the graphics interface, this provides a powerful tool for prototyping & tuning ATR algorithms. Significant improvements were obtained from the knowledge based version over the conventionally coded procedural version.

Abstract: There are two demanding needs which must be fulfilled to further the development of automatic target recognition systems. One is analytical modeling for performance prediction and the other is a disciplined evaluation methodology for automatic target recognition algorithms. Currently both areas are in their infancy. The analytical modeling for automatic target recognition performance evaluation and for prediction of algorithm performances has been investigated, both as they relate to human visual performances and as a tool for algorithm development. The matched filter approach presented a good limiting performance for target detection in uncluttered scenes with complete knowledge- of the target characteristics. The comparision of the matched filter detection performance to human performance model have produced some interesting results.© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: This paper describes two techniques for automatic recognition of surface targets from an airborne platform using an imaging laser radar sensor The first technique rotates a three-dimensional model of the target in real time to enable a generalized Hough transform to match the ladar image to the target''s key discriminating features as a basis for target identification The second technique uses a variation on minimum average correlation energy filters to perform robust target identification Examples illustrating the application of these algorithms are presented along with a description of a real-time implementation of the critical parts of the algorithms on a 40 systolic array based on the Geometric Arithmetic Parallel Processor (GAPP) chip© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering Downloading of the abstract is permitted for personal use only

Abstract: Adequate tools for diagnosis and evaluation of Automatic Target Recognition (ATh) systems are very critical for their successful development. In this paper we describe system called: Automated Instrumentation and Evaluation (Auto-I). Auto-I provides many of the needed capabilities for rapid testing and evaluation of ATR systems. It also provides a module for automatic adaptation of algorithms parameters using algorithms performance models, optimization and Artificial Intelligence techniques. The current design of Auto-I is modular, it is designed so it can be interfaced to other ATh systems.© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: A primary concern of target recognition systems is the actual detection of targets in a scene. Whereas identification assigns each detected target to a class, targets that go undetected are not considered. This leads to deteriorating system performance. This paper describes a neural network approach to the target recognition problem that exploits target composition in conjunction with structure to detect targets. An automatic target recognition architecture is presented to identify how the neural network environment may be integrated into existing systems. The neural network system is described in detail and the training process delineated to show the actual implementation and training issues involved. Performance of the system is documented through evaluative analysis of results generated using an infrared image database. The research performed to date is summarized and a discussion of future developments and their complement to the target recognition process is presented.

Abstract: The Fourier transformation was applied on a set of typed text characters, extracting their unique features and developing an appropriate knowledge base for quick text character recognition. The use of this technique may also allow the development of an adaptive recognizer capable of learning through proper development of the classifier. The proposed technique computes the Fourier transform of the input string derived by the HVP (horizontal-vertical projection) process. In particular, the string created by the HVP scheme is a combination of two strings from the horizontal and vertical projections. The coefficients of the input string-derived Fourier series are compared with the features of the known characters, and classification is performed based on the closeness of the features set. Analysis of test results showed that the Fourier transform approach for feature extraction and the simple classification technique chosen in this project displayed a classification accuracy of over 80% for a limited set of conditions. >

Abstract: A general transformation equation is derived and applied to dynamic and static image differences. It uses translation, rotation, scale, and velocity changes as fundamental object features for image segmentation. A linear minimization algorithm is derived based on a spatial smoothness constraint. The transformation and algorithm constitute an automatic target recognition preprocessor architecture.

Abstract: An assessment of numerous activities in the field of multisensor target recognition reveals several trends and conditions which are cause for concern. .These concerns are analyzed in terms of their potential impact on the ultimate employment of automatic target recognition in military systems. Suggestions for additional investigation and guidance for current activities are presented with respect to some of the identified concerns.

Abstract: In this paper, the application of Joint Transform Correlator (JTC) techniques for use in automatic target recognition using actual sensor data is addressed. The problem of interest is the detection and classification of objects in forward looking infrared (FLIR) images. A Joint Transform Correlator architecture using a single magnetooptical spatial light modulator (MOSLM) and a charge coupled device (CCD) camera is designed and tested. A unique technique for binarizing the fringe pattern, commonly called the joint transform power spectrum (JTPS), that enhances the application to actual sensor images is presented. The modification of the architecture to allow for scale and rotation invariant target recognition is also presented.© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: This paper describes the comparison of three classifiers for use in an automatic target recognition (ATR) system for millimeter wave (MMW) radar data. The three classifiers were the quadratic (Bayesian-like), the multilayer perceptron using a backpropagation training algorithm (termed backpropagation for short), and the counterpropagation network. Two data sets, statistical with four classes and real radar data with three classes, were used for training and testing all three classifiers. Three experiments were performed including: comparing the performances between the three classifiers on both the statistical feature set and the real radar data; optimal configuration for the backpropagation network; and the number of training iterations required for optimal performance using the backpropagation network before overtraining occurred.© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: The availability of validated, synthetic multisensor databases is crucial for the development, testing, and evaluation of advanced Automatic Target Recognition algorithms. Real multisensor scene combinations reflecting operational requirements, missions, and targets are in many cases unavailable and economically difficult to obtain in the variety and quantity needed. This per describes the Texas Instruments (TI's) Synthetic Multisensor (IR, TV, Laser Radar) Image Generation System which has been developed to address this database problem and the procedures used to validate the synthetic imagery. The system generates single frames or image sequences which are based upon scenarios derived directly from DMA digital map data. A user interface designed to allow the user to control the operation of the synthetic image generation system, and to create, modify, and control the associated scenario is described along with the sensor models, atmospheric environment models, and scene rendering software utilized. In addition, validation of the synthetic imagery with respect to real imagery using a feature based methodology is addressed.© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: This paper briefly reviews current methods intended to test automatic target recognition systems (ATRS) and the databases needed to support these methods. A new approach to field testing ofATRS is described.1 This approach can also provide means ofgenerating needed realistic, closely documented imagery for ATRS development and testing.

Abstract: A generalized quadratic (Bayesian-like) classification system has been devel-oped for evaluating the performance of other classifiers, such as neural networks,in automatic target recognition (ATR). The system was tested using multispec-tral real data as well as computer generated data sets. The classifier employsthe covariance matrix and centroid of the feature set to describe each region.The system then calculates the likelihood associated with an unknown objectbelonging to a defined region. A multivariate normal distribution is assumed incalculating this likelihood. The system utilizes a learning algorithm to continu-ously upgrade performance and has shown near 100 percent accuracy even aftervery short training periods. 1. INTRODUCTION Recent research in pattern recognition is motivated by the desire to have aninsight into the pattern recognition capabilities in humans. The ultimate goalof this research is to develop machine vision systems of similar capabilities'. Inorder to develop an adaptive pattern recognition system modeled after the hu-man visual system, it is first necessary to thoroughly understand the statisticalnature of the input data sets.The intent of this work has been to develop a proven classifier against whichother classifier performance can be compared and upon which more sophis-ticated systems may be built.The maximum likelihood classifier implementedhere uses the estimated statistical parameters associated with training samplesin order to classify an unknown object2'3.

Abstract: Automatic road detection is an important part in many scene recognition applications. The extraction of roads provides a means of navigation and position update for remotely piloted vehicles or autonomous vehicles. Roads supply strong contextual information which can be used to improve the performance of automatic target recognition (ATh) systems by directing the search for targets and adjusting target classification confidences. This paper will describe algorithmic techniques for labeling roads in high-resolution infrared imagery. In addition, realtime implementation of this structural approach using a processor array based on the Martin Marietta Geometric Arithmetic Parallel Processor (GAPPTh) chip will be addressed. The algorithm described is based on the hypothesis that a road consists of pairs of line segments separated by a distance "d" with opposite gradient directions (antiparallel). The general nature of the algorithm, in addition to its parallel implementation in a single instruction, multiple data (SIMD) machine, are improvements to existing work. The algorithm seeks to identify line segments meeting the road hypothesis in a manner that performs well, even when the side of the road is fragmented due to occlusion or intersections. The use of geometrical relationships between line segments is a powerful yet flexible method of road classification which is independent of orientation. In addition, this approach can be used to nominate other types of objects with minor parametric changes.

Abstract: : A large number of pattern recognition require the ability to recognize patterns within data when the character of the patterns may change with time. Examples of such tasks are remote sensing, autonomous control, and automatic target recognition in a changing environment. Titterington et al give a list of tasks to which mixture models have been applied. Many of these tasks, and their variants, fall into the above categories.) These tasks have a common requirements: the need to recognize new entities as they enter the environment. A pattern recognition system must be able to recognize and develop a representation of a new pattern in the environment as well as to change its representation of the statistics of the pattern dynamically. The adaptive mixtures approach presented here uses density estimation to develop decision functions for supervised and unsupervised learning. Much work in performing density estimation in supervised situations has been done. For the most part, this research has centered on approaches that use a great deal of a priori information about the structure of the data. In this work, we apply statistical pattern recognition concepts to the problem of recursive nonparametric pattern recognition in dynamic environments.