Multi-Aspect Convolutional-Transformer Network for SAR Automatic Target Recognition

In recent years, synthetic aperture radar (SAR) automatic target recognition (ATR) has been widely used in both military and civilian fields. Due to the sensitivity of SAR images observation azimuth, multi-aspect image sequence contains more information for than a single-aspect one. Nowadays, methods mainly use recurrent neural networks (RNN), which rely on order between thus suffer from loss. At same time, training deep learning model also requires lot data, but are expensive obtain. Therefore, this paper proposes method based self-attention, is find correlation semantic images. Simultaneously, improve anti-noise ability proposed reduce dependence large amount convolutional autoencoder (CAE) pretrain feature extraction part designed. The experimental results using MSTAR dataset show that superior various working conditions, performs well with few samples strong anti-noise.