Neural network based moving targets localization in SAR-GMTI system

Abstract

Moving targets localization is an important challenge in current synthetic aperture radar (SAR) — ground moving target indication (GMTI) system. The along track interferometry (ATI) method is usually used for GMTI in SAR, however, which suffers from ambiguous parameters estimation due to wrapped phase and less robust to estimation errors (of system parameters and motion parameters of the moving targets). Focusing on these, we propose a new neural network based method to improve the moving targets localization performance. After analyzing the main error source of the conventional ATI method, we utilize an error back propagation (BP) neural network to replace the processing after range compression in the ATI method. And the proposed method is proved effective by the experiments, and it performs much better than the conventional ATI method in localization accuracy, efficiency and robustness.