Raghu G. Raj

SAR Automatic Target Recognition Using Discriminative Graphical Models

The problem of automatically classifying sensed imagery such as synthetic aperture radar (SAR) into a canonical set of target classes is widely known as automatic target recognition (ATR). A typical ATR algorithm comprises the extraction of a meaningful set of features from target imagery followed by a decision engine that performs class assignment. While ATR algorithms have significantly increased in sophistication over the past two decades, two outstanding challenges have been identified in the rich body of ATR literature: 1) the desire to mine complementary merits of distinct feature sets (also known as feature fusion), and 2) the ability of the classifier to excel even as training SAR images are limited. We propose to apply recent advances in probabilistic graphical models to address these challenges. In particular we develop a two-stage target recognition framework that combines the merits of distinct SAR image feature representations with discriminatively learned graphical models. The first stage projects the SAR image chip to informative feature spaces that yield multiple complementary SAR image representations. The second stage models each individual representation using graphs and combines these initially disjoint and simple graphs into a thicker probabilistic graphical model by leveraging a recent advance in discriminative graph learning. Experimental results on the benchmark moving and stationary target acquisition and recognition (MSTAR) data set confirm the benefits of our framework over existing ATR algorithms in terms of improvement in recognition rates. The proposed graphical classifiers are particularly robust when feature dimensionality is high and number of training images is small, a commonly observed constraint in SAR imagery-based target recognition.

Automatic target recognition using discriminative graphical models

Of recent interest in automatic target recognition (ATR) is the problem of combining the merits of multiple classifiers. This is commonly done by “fusing” the soft-outputs of several classifiers into making a single decision. We observe that the improvement in recognition rates afforded by these approaches is due to the complementary yet correlated information captured by different features/signal representations that these individual classifiers employ. We present the use of probabilistic graphical models in modeling and capturing feature dependencies that are crucial for target classification. In particular, we develop a two-stage target recognition framework that combines the merits of distinct and sparse signal representations with discriminatively learnt graphical models. The first stage designs multiple projections yielding M > 1 sparse representations, while the second stage models each individual representation using graphs and combines these initially disjoint and simple graphical models into a thicker probabilistic graphical model. Experimental results show that our approach outperforms state-of-the art target classification techniques in terms of recognition rates. The use of graphical models is particularly meritorious when feature dimensionality is high and training is limited - a commonly observed constraint in synthetic aperture radar (SAR) imagery based target recognition.

Meta-classifiers for exploiting feature dependencies in automatic target recognition

Of active interest in automatic target recognition (ATR) is the problem of combining the complementary merits of multiple classifiers. This is inspired by decades of research in the area which has seen a variety of fairly successful feature extraction techniques as well as decision engines being developed. While heuristically based fusion techniques are omnipresent, this paper explores a principled meta-classification strategy that is based on the exploitation of correlation between multiple feature extractors as well as decision engines. We present two learning algorithms respectively based on support vector machines and AdaBoost, which combine soft-outputs of state of the art individual classifiers to yield an overall improvement in recognition rates. Experimental results obtained from benchmark SAR image databases reveal that the proposed meta-classification strategies are not only asymptotically superior but also have better robustness to choice of training over state-of-the art individual classifiers.

ISAR imaging in sea clutter via compressive sensing

We investigate the application of compressive sensing (CS) to inverse synthetic aperture radar (ISAR) imaging of moving targets. We present our results for a simulated target immersed in different levels of sea clutter. Comparison between traditional and CS approaches to ISAR imaging reveal that our based CS algorithm offers some advantages compared to traditional ISAR imaging under certain limited operating conditions that are nevertheless of practical interest. We conclude by pointing out directions for future work in extending the results of this paper.

Performance studies of emulated multichannel SAR for motion characterization

Conventional single-channel synthetic aperture radar (SAR) cannot directly measure or correct for smearing and distortions due to scene motion. Previous theoretical efforts have pointed out that this deficiency can be overcome using an along-track multichannel SAR (MSAR), but lack an experimental foundation. In this paper we conduct in-depth sensitivity studies that characterize and demonstrate the robustness of an MSAR system for scene motion characterization. These include novel experimental studies that build on our previous work in MSAR emulation. Of significance, to our knowledge this is the first experimental verification of removing motion distortions in shoaling ocean waves using such an MSAR system.We also introduce a new processing mode for MSAR systems called SAVSAR (subaperture velocity SAR), which affords us a much wider range of options in adjusting the spatial and temporal resolution capabilities of MSAR imaging. The insights gained in this paper provide guidance on the choice of suitable configurations and processing methods for scene motion estimation using airborne MSAR systems.

Analysis of radar dismount signatures via non-parametric and parametric methods

We present novel approaches to the analysis of radar dismount signatures that entail the characterization of the time-frequency (TF) structure of the received radar signal associated with the dismount gait by both non-parametric and parametric methods. We first introduce the concept of Gaussian g-Snakes in order to parametrically characterize the TF distribution of radar signals. In particular, we derive simple steepest descent equations that enable the estimation of the (locally) optimal g-Snake parameters for a given TF distribution. Furthermore the g-Snake modeling methodology gives us an objective unsupervised criterion from which to quantify the quality of the motion curve estimates that have been tracked from the TF data. We then formulate the non-parametric motion estimation for TF signals by a coupling of a simple partial tracking methodology in conjunction with boundary condition enforcement with regularity constraints. Finally we propose a coupling of the above non-parametric approach with g-Snake modeling that result in improved overall modeling of the given real and simulated radar TF data.

A multi-scale and adaptive Track-Before-Detect technique for maritime environments

Track-Before-Detect (TBD) is a detection technique that simultaneously tracks and detects a target. The technique is specially useful for targets whose back-scattered return is significantly lower than the surrounding noise and background environments. Different algorithms proposed to perform TBD include particle filter, dynamic programming and Hough transform algorithms have been proposed. These techniques work well in a highly noise environment with constant velocity targets. However, some of these techniques are not applicable to targets that have acceleration as well as targets embedded in sea clutter. In this paper, we present a beamlet track-before-detect technique for maritime detection. Due to the sea clutter, two different adaptive sea clutter filtering methods are applied before the aforementioned TBD technique. Overall, the paper introduces and studies a beamlet TBD technique for maneuverable maritime targets in sea clutter.

Localized dictionary design for geometrically robust sonar ATR

Advancements in Sonar image capture have opened the door to powerful classification schemes for automatic target recognition (ATR). Recent work has particularly seen the application of sparse reconstruction-based classification (SRC) to sonar ATR, which provides compelling accuracy rates even in the presence of noise and blur. However, existing sparsity based sonar ATR techniques assume that the test images exhibit geometric pose that is consistent with respect to the training set. This work addresses the outstanding open challenge of handling inconsistently posed Sonar images relative to training. We develop a new localized block-based dictionary design that can enable geometric robustness. Further, a dictionary learning method is incorporated to increase performance and efficiency. The proposed SRC with Localized Pose Management (LPM), is shown to outperform the state of the art SIFT feature and SVM approach, due to its power to discern background clutter in Sonar images.

A hierarchical Bayesian-map approach to computational imaging

We present a novel approach to inverse problems in imaging based on a Hierarchical Bayesian-MAP (HB-MAP) formulation. In this paper we specifically focus on the difficult and basic inverse problem of multi-sensor (tomographic) imaging wherein the source image of interest is viewed from multiple directions by independent sensors. We employ a Probabilistic Graphical Modeling extension of the Compound Gaussian (CG) distribution as a global image prior into a Hierarchical Bayesian inference procedure. We first demonstrate the performance of the algorithm on Monte-Carlo trials followed by empirical data involving natural (optical) images. We demonstrate how our algorithm outperforms many of the previous approaches in the literature including Filtered Back-projection (FBP) and a variety of state-of-the-art compressive sensing (CS) algorithms.

A nonlinear compound representation of sea clutter

We explore the applicability of our recently developed Nonlinear Compound Gaussian (NCG) [1] distribution to modeling the statistics of sea clutter data. We first, for completeness, give a self-contained description of our NCG distribution; both its theoretical properties and the algorithmic details for parameter estimation. We then demonstrate the performance of NCG in modeling the range statistics of sea-clutter data [12]. The results clearly demonstrate the superiority of NCG in modeling sea-clutter phenomena over the compound Gaussian (CG) distribution. We conclude with a brief discussion of a phenomenological interpretation of these results together with directions for future research.