A comparative study of basis selection techniques for automatic target recognition

Often in automatic target recognition (ATR) problems, a small number of representative features that encapsulate image information are usually extracted from the target images prior to the actual classification procedure. In literature, principal component analysis (PCA) is one of the most widely used feature extraction techniques. In this paper, we investigate the capability of basis representations to encode discriminative information for target classification using synthetic aperture radar (SAR) imagery. Specifically, we consider the two different scenarios of shared basis built using all available training and class-specific basis using training from each class separately. We compare the traditional PCA-based technique with basis representations constructed using oriented PCA and non-negative matrix approximations (NNMA). Experiments on the benchmark MSTAR database reveal the merits of basis selection techniques that can model imaging physics more closely and can capture inter-class variability, in addition to identifying a trade-off between classification performance and availability of training.