Multi-view SAR Target Recognition Using Bidirectional Conv-LSTM Network

Abstract

Deep neural networks are widely used in synthetic aperture radar (SAR) automatic target recognition (ATR) due to their excellent performance. The imaging mechanism of SAR images differs from that of optical images in that SAR images are highly angle sensitive. However, current SAR ATR methods based on deep learning frameworks generally lack the use of spatial correlation information between multi-view SAR images. In this paper, we propose a multi-view SAR image recognition method based on a bidirectional convolutional long and short term memory (Conv-LSTM) network. Firstly, we use a Log-Gabor filter to extract angle-stable monogenic features to reduce inter-class differences. Secondly, feature dimensionality reduction is performed using a multilayer perceptron (MLP) network. Finally, a bidirectional LSTM network is used to integrate the SoftMax classifier for target recognition. The experimental results on the MSTAR dataset and the self-made dataset show that the average recognition accuracy of our proposed method can reach more than 99%. The results of our method outperform other existing methods, indicating the effectiveness and application potential of our algorithm.