Abstract: A novel method for fusion detection of small infrared targets based on support vector machines (SVM) in the wavelet domain is presented. Target detection task plays an important role in automatic target recognition (ATR) systems because overall ATR performance depends closely on detection results. SVM is a powerful methodology for solving problems in nonlinear classification, function estimation and density estimation. Least-squares support vector machines (LS-SVMs) are reformulations to standard SVMs. The proposed algorithm can be divided into four steps. First, each frame of the image sequence is decomposed by the discrete wavelet frame (DWF). Second, the components with low frequency are performed by regression based on LS-SVM. The one-order partial derivatives in row and column directions are derived. Therefore, feature images of the gradient strength can be obtained. Third, feature images of five consecutive frames are fused to accumulate the energy of target of interest and greatly reduce false alarms. Finally, the segmentation method based on contrast between target and background is utilized to extract the target. In terms of connectivity of moving targets, the majority of residual clutter and false alarms that survive are removed based on 3-D morphological dilation across three consecutive frames along the motion direction of the moving targets. Actual infrared image sequences in backgrounds of sea and sky are applied to validate the proposed approach. Experimental results demonstrate the robustness of the proposed method with high performance.

Abstract: We propose a ground target recognition method based on 3-D laser radar data. The method handles general 3-D scattered data. It is based on the fact that man-made objects of complex shape can be decomposed to a set of rectangles. The ground target recognition method consists of four steps; 3-D size and orientation estimation, target segmentation into parts of approximately rectangular shape, identification of segments that represent the target's functional/main parts, and target matching with CAD models. The core in this approach is rectangle estimation. The performance of the rectangle estimation method is evaluated statistically using Monte Carlo simulations. A case study on tank recognition is shown, where 3-D data from four fundamentally different types of laser radar systems are used. Although the approach is tested on rather few examples, we believe that the approach is promising

Abstract: In this paper, Adaboost and SVM are applied to SAR ATR(Synthetic Aperture Radar Automatic Target Recognition) respectively. The performance of these two classifiers is analyzed and compared in target aspect window with different size. First, PCA (Principal Component Analysis) features are selected as target feature, and then Adaboost. M1 and SVM are used to classify, respectively. Experimental results based on MSTAR data sets show that Adaboost classifier has better robustness than SVM classifier.

Abstract: Principal component analysis (PCA) has been used in many applications ranging from social science to space science, for the purpose of data compression and feature extraction. Usage of PCA for synthetic aperture radar (SAR) image classification, though widely reported by remote-sensing researchers, has not been exploited much by automatic target recognition (ATR) community. In the present paper, PCA has been used in SAR-ATR using the MSTAR data base, and comparison has been made with the conventional conditional Gaussian model based Bayesian classifier (M.D. DeVore and J.A. O'Sullivan, 2002). The results have been compared based on percentage of correct classification, receiver operating characteristics (ROC), and performance with limited amount of training data. By all standards of comparison, the PCA based classifier was observed to outperform the conditional Gaussian model based Bayesian classifier (CGBC) or at the worst it performs at par. And given the computational and algorithmic simplicity of PCA based classifier, the new algorithm was concluded to be a highly prospective candidate for real time ATR systems

Abstract: In modern day warfare, reconnaissance operations such as automatic target recognition(ATR) using unmanned aerial vehicles(UAVs) constitute a strategic war tactic. Traditionally, ATR is performed by UAVs that fly within the reconnaissance area to collect image data through sensors and upload the data to a central base station for analyzing and identifying potential targets. The centralized approach to ATR introduces several problems including scalability with the number of UAVs, network delays in communicating with the central location, and, susceptibility of the system to malicious attacks on the central location. These challenges can be addressed using a distributed system for performing ATR. In this paper, we describe a multi-agent system called COMSTAR-UAV (Cooperative Multi-agent System for automatic TArget Recognition using Unmanned Aerial Vehicles) that uses swarming techniques inspired from insect colonies to perform ATR in a distributed manner.

Abstract: Simultaneous detection and classification of single/multiple identical and dissimilar targets is very important in automatic target recognition applications. A new approach is proposed for this purpose using a combination of maximum average correlation height (MACH) filter and polynomial distance classifier correlation filter (PDCCF). In this technique, full-resolution MACH filters are applied to the input scene, and the regions of interest (ROIs) containing the probable targets are selected from the input scene based on the ROIs with higher-correlation peak values in the correlation output. Then a multiclass PDCCF is applied to these ROIs to identify target types and reject clutters and/or backgrounds. To increase the robustness of the proposed technique, multiple filters are formulated for multiple ranges of target size and/or orientation variations. The simulation results using real-life imagery indicate the effectiveness of the proposed technique for target detection and classification in the presence of distortion, clutter, and other artifacts.

Abstract: This paper addresses robust speech feature extraction in combination with statistical speech feature enhancement and couples the particle filter to the speech recognition hypotheses. To extract noise robust features the Fourier transformation is replaced by the warped and scaled minimum variance distortionless response spectral envelope. To enhance the features, particle filtering has been used. Further, we show that the robust extraction and statistical enhancement can be combined to good effect. One of the critical aspects in particle filter design is the particle weight calculation which is traditionally based on a general, time independent speech model approximated by a Gaussian mixture distribution. We replace this general, time independent speech model by time- and phoneme-specific models. The knowledge of the phonemes to be used is obtained by the hypothesis of a speech recognition system, therefore establishing a coupling between the particle filter and the speech recognition system which have been treated as independent components in the past. Index Terms: particle filters, automatic speech recognition, speech feature enhancement, phoneme-specific

Abstract: We propose in this paper the standard Hough transform to be changed with analogous polar Hough transform (PHT). The obtained polar Hough detection algorithm is very comfortable for track and target detection because the input parameters of this transform are the output parameters of the search radar. Another important advantage is the transform stability when the target changes its speed and flies at different azimuths.

Abstract: Automatic target recognition (ATR) in target search phase is very challenging because the target range and mobility are not yet perfectly known, which results in delay-doppler uncertainty. In this paper, we firstly perform some theoretical studies on radar sensor network (RSN) design based linear frequency modulation (LFM) waveform: (1) the conditions for waveform co-existence, (2) interferences among waveforms in RSN, (3) waveform diversity in RSN. Then we apply RSN to ATR with delay-doppler uncertainty and propose maximum-likekihood (ML) ATR algorithms for fluctuating target and nonfluctuating target. Simulation results show that our RSN vastly reduces the ATR error comparing to a single radar system in ATR with delay-doppler uncertainty.

Abstract: This paper presents the status of our SAR automatic target recognition (ATR) work on the Moving and Stationary Target Acquisition and Recognition (MSTAR) public database using the minimum noise and correlation energy (MINACE) distortion-invariant filter (DIF). We use a subset of the MSTAR public database for the benchmark three-class problem and we address confuser and clutter rejection. To handle the full 360° range of aspect view in MSTAR data, we use a set of Minace filters for each object; each filter should recognize the object (and its variants) in some angular range. We use fewer DIFs per object than prior work did. The Minace parameter c trades-off distortion-tolerance (recognition) versus discrimination (confuser/clutter rejection) performance. Our filter synthesis algorithm automatically selects the Minace filter parameter c and selects the training set images to be included in the filter, so that the filter can achieve both good recognition and good confuser and clutter rejection performance; this is achieved using a training and validation set. In our new filter synthesis method, no confuser, clutter, or test set data are used. The peak-to-correlation energy (PCE) ratio is used as the correlation plane metric in both filter synthesis and in tests, since it works better than correlation peak height. In tests, we do not assume that the test input's pose is known (as most prior work does), since pose estimation of SAR objects has a large margin of error; we describe our procedure for proper use of pose estimates in MSTAR recognition. The use of circular versus linear correlations is addressed. We also address the use of multi-look SAR data to improve performance.© (2006) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: The Autonomous Operations Future Naval Capability (AOFNC) program developed a 12.75" diameter autonomous underwater vehicle (AUV) and a synthetic aperture sonar (SAS12) payload. This system falls under the lightweight designator of the Unmanned Undersea Vehicle (UUV) master plan. Bluefin Robotics Corporation and the Applied Research Laboratory of The Pennsylvania State University (ARL/PSU) developed the vehicle/payload system. In addition to the previous team members, Naval Surface Warfare Center Panama City (NSWC PC) developed the synthetic aperture image processing. The system will include motion compensation and beam formation software, real time data handlers, and automatic target recognition algorithms. NSWC PC provided test range services and test planning to the project, as well. The AUV design is an open frame that allows modular payloads to be attached. The modules are self-contained and the surround is free-flooded. A plastic fairing covers the payload and vehicle subsystems. The payload power and communications are supplied through common interfaces. The vehicle hosts a suite of inertial, environmental, and heading sensors, as well as, a Doppler velocity log (DVL). Data from this sensor suite is combined to provide the information necessary for proper SAS operation. This data is used both in the SAS ping timing and ultimately in the correction of errors due to aperture misalignment. Vehicle and payload data and logs are recorded and used to evaluate system performance. The SAS payload is designed using COTS data acquisition and communication hardware. The SAS operates at 180 kHz in the side looking mode. A suite of arbitrary waveforms can be transmitted to optimize SAS performance in a given environment. The broadband receiver is designed for minimal channel-to-channel gain and phase errors necessary for acquisition of high fidelity signals. Signals are filtered and decimated then passed to the recorder and processing systems. The individual element aperture determines the ultimate resolution limit. In principle, SAS12 can be processed for 25 mm resolution at all ranges out to a maximum of 150 meters. One advantage of SAS is that the data collected can be processed to whatever resolution is defined by the user, within this limit. This is useful in resolution studies because the same data set can be processed for different resolutions. Typically real apertures have a fixed resolution proportional to the physical length. Depending on the real aperture system, this resolution may be constant or vary as a function of range. The system will include real time automatic target recognition (ATR). The ATR consists of a set of algorithms developed by several different contributors. The master algorithm uses a rule based system to combine the information generated by the individual contributors. The result produces a lower false alarm rate that any single algorithm. The authors present performance for comparison to the existing data bases that relate ATR performance to image resolution. ATR performance is affected by clutter, bottom type, target aspect, and many other characteristics, as modified by the SAS resolution. Imagery is presented with ATR performance measures. Sonar performance is discussed in qualitative terms, and is based on image appearance and knowledge of what targets are present in the field. Quantitative performance measures are also presented in terms of requirements of the ATR, Probability of Detection (Pd), and Probability of False Alarm (Pfa)

Abstract: Aimed at the necessary on the automatic detection level in unattended substation, a new method which monitoring and diagnosis the high-voltage (HV) power equipment running state based on image processing is put forward. Image processing techniques realize and strengthen the capability of the human vision. It is a non-contact measurement method, which can monitor the HV power equipment running state on-line. In recent years, it is widely applied in many fields and gets plentiful and substantial success. The HV power equipment running state detection process is introduced in this paper. After image preprocesses and feature extraction, the automatic target recognition algorithm based on the radial basis function neural network (RBFNN) is presented.

Abstract: The problem of target recognition using the high resolution radar range profiles is discussedTwo feature extraction methods based on differential power spectrum(DPS) and differential cepstrum,originally used in the research area of speech signal processing and homomorphic signal processing are respectively introduced to the radar target recognition communityTwo differential power spectrum based features are applied to target classificationA multi-layered feed forward neural network with SARPROP(simulated annealing resilient propagation) algorithm is selected as classifierThe range profiles are obtained with step-frequency technique and the two-dimension backscatter distribution data of four different scaled aircraft modelsSimulations are presented to evaluate the classification performance with the above featuresThe results show that the differential power spectrum based feature is effective and robust for the radar target recognition

Abstract: Missing Data Techniques have already shown their effectiveness in dealing with additive noise in automatic speech recognition systems For real-life deployments, a compensation for linear filtering distortions is also required Channel compensation in speech recognition typically involves estimating an additive shift in the log-spectral or cepstral domain This paper explores a maximum likelihood technique to estimate this model offset while some data are missing Recognition experiments on the Aurora2 recognition task demonstrate the effectiveness of this technique In particular, we show that our method is more accurate than previously published methods and can handle narrow-band data Index Terms: speech recognition, missing data techniques, convolutional distortion, channel estimation

Abstract: A multi-stage maximum likelihood target estimator for use with radar and sonar systems is provided. The estimator is a software implemented algorithm having four computational stages. The first stage provides angle smoothing for data endpoints thereby reducing angle errors associated with tie-down times. The second stage performs a coarse grid search to obtain the initial approximate target state to be used as a starting point for stages 3 and 4 . The third stage is an endpoint Gauss-Newton type maximum likelihood target estimate which determines target range along two time lines. The final refinement of the target state is obtained by the fourth stage which is a Cartesian coordinate maximum likelihood target estimate. The four-stage processing allows the use of target historic data while reducing processing time and computation power requirement.

Abstract: An automatic target recognition system has been assembled and tested at the Research Institute for Optronics and Pattern Recognition in Germany over the last years. Its multisensorial design comprises off-the-shelf components: an FPA infrared camera, a scanning laser radar und an inertial measurement unit. In the paper we describe several possibilities for the use of this multisensor equipment during helicopter missions. We discuss suitable data processing methods, for instance the automatic time synchronization of different imaging sensors, the pixel-based data fusion and the incorporation of collateral information. The results are visualized in an appropriate way to present them on a cockpit display. We also show how our system can act as a landing aid for pilots within brownout conditions (dust clouds caused by the landing helicopter).

Abstract: Synthetic aperture radar (SAR) automatic target recognition (ATR) based on the extended maximum average correlation height (EMACH) distortion invariant filter (DIF) is presented. Prior work on the EMACH filter addresses 3-class and 10 class classification with clutter rejection. However, the ability of the EMACH filter to reject confusers is not well known. This paper addresses this. We follow a benchmark procedure which involves classification of three object classes over 360° aspect angle differences and with depression angle and variant differences and rejection of two unseen confusers from the Moving and Stationary Target Acquisition and Recognition (MSTAR) public database. We present a scheme to select which training set images to include while making the filters, since it is not necessary to use all training set images to make the filters. Results for classification with both confuser and clutter rejection are presented. We also compare our work with prior EMACH MSTAR work. We find EMACH filters to have poor confuser and clutter rejection. We also correct prior EMACH clutter rejection performance results.

Abstract: Image measure is a very important part of automatic target recognition algorithm performance evaluation. Whether the image measure accurately relates with algorithm performance will affect directly evaluations. In this paper current image measures are introduced, and the deficiency of the target to background contrast (TBC) measure is analyzed, which is the best single measure and the representative of current general measures. A new texture-based image clutter measure (TIC) is derived from gray level cooccurrence (GLC) matrices, which embody important texture information. The result of testing two measures TBC and TIC shows that the relation between TBC and segmentation algorithm performance is monotonic rising in given scenario condition, but in complex condition it will fail, and that in both conditions TIC has very good monotonic relation with the segmentation algorithm performance. TIC is a fairly robust indicator of segmentation algorithm performance and better suited than TBC.

Abstract: Automatic Target Recognition (ATR) is often based on statistical methodologies rather than using symbolic information relationships and properties In this paper, a possible approach for the syntactic outline representation for radar target classification based on non-radar templates is presented, describing the outline extraction and line detection algorithms to define a code-book of structures representing the target radar shape The outline target recognition is attempted from a multi-perspective point of view: only the illuminated side of the target is preserved and then combined with other partial shapes collected from different nodes of a hypothetical radar network

Abstract: A method to quantify clutter in hyperspectral infrared (HSI) images in a framework similar to work done on single-band images is presented. Hereby, all objects in a scene that may be mistaken for targets by an automatic target recognition (ATR) algorithm are considered clutter. A hyperspectral image contains a number of contiguous discrete bands within the spectrum. The aim is to obtain a measure of complexity for hyperspectral images, which will indicate the inherent difficulty for an ATR to detect targets. We implemented 129 different image clutter metrics, and computed them for a database of synthesized HSI images. A matched filter ATR was used to determine the amount of clutter in the images as a baseline. We developed a method to select a subset of the metrics that in combination correlated best with the amount of clutter in an image, and defined this as the clutter complexity measure (CCM). Multiple runs of this selection procedure for different training image groups show a dominance of a further subset of metrics that best predict the CCM. Our results also show that the CCM obtained from a varying number of random sample images generalizes well for the entire database.

Abstract: This paper reports some preliminary results of an examination into the feasibility of recognising the Doppler signatures of targets using speech recognition processing techniques The rationale is that human operators typically listen to the Doppler audio output from the surveillance radar to detect and possibly identify targets A feature of speech recognition is that pre-processing is used that takes account of the voice mechanisms that produce speech and the characteristics of the human ear Three different recognition techniques, with identical pre-processing, were implemented After validating the recognition algorithms with speech the recognisers were retrained with Doppler signals from a number of sources It was found that the best of the speech recognisers, HMM-GMM, was also the best of the Doppler recognisers with 88% recognition The work has been compared with that of others using a similar technique and a good agreement has been found Some recent discoveries in neuroimaging are quoted that suggest that the human brain and that of several other mammals performs visual recognition in a manner common in speech recognition

Abstract: This paper presents a new approach to data fusion for automatic target recognition, surveillance, and tracking research areas growing in an intelligent transportation system. Robust data alignment (RDA), finding relational maps among a sequence of invariant feature data sets, is one of the key requirements for a successful data fusion. To achieve RDA for a multi-modal data fusion, we construct a cost criterion based on information theory and solve an optimization problem with a mixed search strategy combining the Nelder-Mead simplex and random search method. We evaluate the cost criterion and search strategy by a numerical stability test, and demonstrate experimental results on a video sequence collected from an unmanned aerial vehicle (UAV). The results indicate the potential of aerial monitoring and tracking system built upon our information-theoretic RDA

Abstract: Distortion-tolerant correlation filter methods have been applied to many video-based automatic target recognition (ATR) applications, but in a single-frame architecture. In this paper we introduce an efficient framework for combining information from multiple correlation outputs in a probabilistic way. Our framework is capable of handling scenes with an unknown number of targets at unknown positions. The main algorithm in our framework uses a probabilistic mapping of the correlation outputs and takes advantage of a position-independent target motion model in order to efficiently compute posterior target location probabilities. An important feature of the framework is the ability to incorporate any existing correlation filter design, thus facilitating the construction of a distortion-tolerant multi-frame ATR. In our simulations, we incorporate the minimum average correlation energy Mellin radial harmonic (MACE-MRH) correlation filter design, which allows the user to specify the desired scale response of the filter. We test our algorithm on real and synthesized infrared (IR) video sequences that exhibit various degrees of target scale distortion. Our simulation results show that the multi-frame algorithm significantly improves the recognition performance of a MACE-MRH filter while requiring only a marginal increase in computation. We also show that, for an equivalent amount of added computation, using larger filter banks instead of multi-frame information is unable to provide a comparable performance increase.© (2006) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Abstract: This paper treats the problem of target recognition as a decision process The nature of the decision to be made has a bearing on the data gathered and the subsequent processing A key factor in the processing is the separability, ie, the ability to distinguish, of radar images of similar but distinct objects A number of recognition algorithms are considered and their suitability for data sets of various types is discussed In addition some simple measurements of the transfer functions of two targets are considered Observation suggest that the examples have characteristics that may make them readily separable As with all recognition techniques the quality and quantity of training data available will place a limit on the performance of any recognition technique and this discussed in the text The view is formed that a single technique is unlikely to be successful and several techniques cued by gross-features of the image may be more appropriate

Abstract: In this paper we present an information sensing system which integrates sensing and processing resulting in the direct collection of data which is relevant to the application. Broadly, integrated sensing and processing (ISP) considers algorithms that are integrated with the collection of data. That is, traditional sensor development tries to come up with the "best" sensor in terms of SNR, resolution, data rates, integration time, etc. and traditional algorithm development tasks might wish to optimize probability of detection, false alarm rate, class separability, etc. For a typical Automatic Target Recognition (ATR) problem, the goal of ISP is to field algorithms which "tell" the sensor what kind of data to collect next and the sensor alters its parameters to collect the "best" information in order that the algorithm performs optimally. We demonstrate an ISP system which utilizes a near Infrared (NIR) Hadamard multiplexing imaging sensor. This prototype sensor incorporates a digital mirror array (DMA) device in order to realize a Hadamard multiplexed imaging system. Specific Hadamard codes can be sent to the sensor to realize inner products of the underlying scene rather than the scene itself. The developed ISP algorithm uses an ATR metric to send codes to the sensor in order to collect only the information relevant to the ATR problem. The result is a multiple resolution hyperspectral cube with full resolution where targets are present and less than full resolution where there are no targets. Essentially, this is compressed sensing.

Abstract: A lot of efforts have been pursued in Automatic Target Recognition (ATR), including, based on: Fourier transform, wavelet transform, novelty filtering, and many others. Unfortunately, in all these methods, a target, either on-the-move (OTM), or static, has to already be identified. Such a Target Identification (ID) pre-ATR process, however, requires significant data reduction that must be done before the ATR process starts. The pre-ATR ID process becomes complex if it is performed by RF-wireless visual sensor networks, based on Unmanned Ground Vehicles (UGVs), or other ground vehicles. The visual sensors include: TV cameras, artificial animal eyes [1] (fish eye, bug eye, lobster eye [2], etc.), and other video-like sensors. These sensors need to work not only autonomously autonomously, but also in cooperation , through RFwireless inter-communication which should be continuous to preserve constant cooperation. Such constant video communication should avoid video image breakdown (in the form of heavy pixeling, or complete image blackout) under abrupt worsening of digital data transfer conditions, in terms of increasing environmental noise (or, reducing SNR), and/or increasing of BER (bit-error-rate) of the video signal transfer. Thus, replacement of image breakdown by its graceful degradation (i.e., preserving image continuity at the expense of quality reduction) is a central issue of the realtime pre-ATR video data reduction. In this paper, we will first discuss the conditions of continuous video RF-communication, based on graceful image degradation, and then analyze some important examples of the real-time pre-ATR video data reduction, based on cooperative video networks, with TV-cameras, as visual sensors.

Abstract: This paper presents a comparative study between different automatic target recognition (ATR) approaches in the application of synthetic aperture radar (SAR) target recognition Four different categories of approaches are investigated and compared The first is distribution-based where a statistical data model is assumed for the SAR image data The second category contains one approach that is based upon principal component analysis (PCA) The third category employs different neural network architectures The last category utilizes support vector machines (SVM) It contains the classical SVM implementation and an enhanced implementation proposed elsewhere by the authors in which the traditional Euclidean kernel is replaced by a new one that is more suitable for the application in question Experimental results are presented It is shown that the enhanced SVM approach outperforms all other investigated approaches in both the classification performance and the confuser rejection