Monostatic and Bistatic SAR Automatic Target Recognition with Multi-Aspect Feature Fusion in Open Environment

Synthetic aperture radar (SAR) automatic target recognition (ATR) in an open environment full of out-of-library (OOL) targets is more technically challenging than SAR-ATR concerning only known targets. Leveraging the Synthetic and Measured Paired Labeled Experiment (SAMPLE) dataset in the public domain and a proprietary bistatic SAR database featuring multiple types of representative military vehicles, we present a novel deep neural network (DNN) for categorizing the OOL military vehicles by jointly exploiting the multi-aspect target features and the prior knowledge regarding the scene of interest. The proposed network has a reconfigurable structure that could easily incorporate two extension modules specifically designed for the mission at hand, a cross-modality context analysis module for target prescreening and a local feature matching module for target variants and hybrid target identification. Preliminary experiment results show that the proposed network outperform the existing networks in categorizing the OOL targets in open environment.