Vehicle target classification in inverse SAR imagery via multiview feature joint learning network

Radar remote sensing is one of the crucial technologies of intelligent transportation. Among them, inverse synthetic aperture radar (ISAR) has also been considered a vital detection system in intelligent transportation because it can capture rich information about vehicle targets such as dimension, trajectory, and number of wheels. We put forward a target classification method named multiview feature joint learning network (MvFJLN) to achieve vehicle target classification in sequential ISAR imagery. Specifically, we first propose a multiview feature extraction architecture, which appeals to attention-based dynamic convolution and multiview weighted fusion to dynamically mine view-varying and vehicle discrimination information from sequential ISAR imagery. Second, we develop a spatial–temporal feature refinement module to gradually highlight the discriminability of ISAR vehicle’s spatial–temporal features. Finally, we leverage the architecture of long short-term memory to achieve multiview ISAR image features aggregation and vehicle target classification. A series of evaluation experiments illustrate that the proposed MvFJLN is competitive with several state-of-the-art vehicle classifications in sequential ISAR imagery.