LRSMTD: Low-Rank Plus Sparse Multiple-Term Decomposition of Defocusing Target Detection for Single-Channel Single-Band Single-Pass VideoSAR

Machine learning-based automatic target detection in video synthetic aperture radar (VideoSAR) has great potential for raising the reconnaissance capability in dynamic region of interest (DROI). In this article, a novel systematic perspective, called low-rank plus sparse multiple-term decomposition (LRSMTD) for simultaneously single-channel, single-band, and single-pass (SCSBSP) VideoSAR configuration is proposed to track the ground defocusing targets. To address the target features of circular VideoSAR imaging, we extend the polar format algorithm (PFA) via exploiting a priori knowledge. In accordance with both the revealed imaging and imagery characteristics, we solve the systematic LRSMTD with proximal exchange-based alternating directions method of multipliers (PEADMM), which makes the process interpretable for defocusing target detection. Comprehensive circular SCSBSP airborne VideoSAR experiments reveal the superior detection performance of the systematic LRSMTD and its several subalgorithms with PEADMM, outperforming 11 state-of-the-art algorithms.