Target recognition of SAR images via hierarchical fusion of complementary features

Abstract

A hierarchical fusion of complementary features is proposed with application to target recognition of synthetic aperture radar (SAR) images. Three features, i.e., principle component analysis (PCA) features, attributed scattering centers (ASCs), and target outline, are used. Sparse representation-based classification (SRC) is adopted to classify the PCA features because of its good effectiveness and robustness. A one-to-one matching algorithm is designed for the ASCs and a partially matching strategy for the target outlines are designed. The test sample is first classified using the PCA features. If the decision is judged to be unreliable, the ASC matching is performed afterwards. Similarly, the outline matching is performed when the ASC matching cannot make a reliable decision on the target label. In this way, the advantages of the complementary features can be combined in a unified framework so the overall target recognition performance can be enhanced. Experiments are implemented based on the moving and stationary target acquisition and recognition (MSTAR) dataset. The results show that the proposed method achieves a notably high recognition rate of 98.34 % under the standard operating condition (SOC) and superior robustness under the extended operating conditions (EOCs) in comparison with other methods.