Compression based class-specific target recognition using SAR images

Abstract

Recognizing complex targets with unknown pose and scale remains an unsolved problem even after half a century of research in the field of synthetic aperture radar (SAR) based automatic target recognition (ATR). Feature extraction and the high-dimension of the feature vectors are two major issues in the field of ATR. Class-specific classification algorithms address the dimensionality issue to some extent, but feature extraction is a problem with such classifiers. Compression can be used to extract the features of synthetic aperture radar image for classification, but has not been exploited much by the ATR community. Using compression for feature extraction not only avoids the problems associated with high-dimensional feature space but also minimizes the storage and computational overheads. However, the disadvantage of using compression based ATR is that classification performance suffers. The proposed technique, compression based class-specific ATR algorithm, is a modular classifier which uses class-specific compression for classification to circumvent the dimensionality problem and at the same time achieve optimal classification results.