Multi-View SAR Automatic Target Recognition Based on Deformable Convolutional Network
Zhiyong Wang, Chenwei Wang, Jifang Pei, Yulin Huang, Yin Zhang, Haiguang Yang, Zhiwei Xing
Abstract
Recently many deep neural networks have been utilized to learn and extract valuable features from synthetic aperture radar (SAR) images for SAR automatic target recognition (A-TR). However, in actual applications the types and amount of data that can be obtained are limited and difficult, which makes it hard to train the networks effectively. In this paper, we propose a multi-view deep learning framework combined with deformable convolution for SAR ATR. The scattering distribution characteristics and morphological characteristics of the target will be learned by the special structure of the deformable convolution, providing more sufficient information for subsequent fusion of features from the distinct views. Experimental results have shown the superiority of the proposed network based on the Moving and Stationary Target Acquisition and Recognition data set and the better recognition performance in the condition of a small number of raw SAR images.