Relation Aware Network for Multi-View SAR Target Recognition

Synthetic Aperture Radar (SAR) has unique advantages and plays an important role in the field of remote sensing, which makes the research of SAR Automatic Target Recognition (ATR) very hot. In recent years, due to the rapid development of deep learning networks and hardware computing power, the deep learning based ATR algorithm for SAR has gradually become the mainstream. Compared with the single-view based recognition, multi-view SAR target recognition tends to better utilize the target scattering information of different azimuth angles. In this paper, we propose a Multi-View Relation Aware (MVRA) network for SAR ATR, which can capture the latent relationships between multiple input views, achieving better recognition accuracy than single-view recognition. Additionally, to enable the network to more accurately capture the contour information of SAR image targets, the Laplacian pyramid is applied for SAR image preprocessing. The proposed MVRA network shows a significant improvement in recognition accuracy on the public dataset MSTAR compared to other multi-view SAR target recognition algorithms. Furthermore, on the new Ground Military Vehicle Target (GMVT) dataset constructed by Nanjing University of Aeronautics and Astronautics (NUAA), the proposed MVRA network demonstrates excellent generalization performance even in the scenarios with relatively strong background clutter interference.