LAUSR

Motivated by recent advances in deep learning, a novel deep complex-valued convolutional neural network (CV-CNN)-based method is proposed for ground moving target indication (GMTI) in a multichannel synthetic aperture radar (SAR) system. The proposed method integrates the SAR-GMTI task into a blind inverse problem solved by a deep CV-CNN named CV-GMTINet. To take advantage of the amplitude and phase information of complex multichannel SAR images, both feature maps and network parameters are extended into the complex domain. The proposed CV-GMTINet is designed by adopting complex-valued residual dense blocks (CV-RDBs) to adaptively learn complex hierarchical features. The trained CV-GMTINet, as a GMTI processor, can be applied to complex multichannel SAR images to discriminate moving targets from stationary clutter and refocus the moving target images simultaneously. Experiments on TerraSAR-X data show that the proposed method achieves significant improvements over existing state-of-the-art GMTI methods in both detection performance and refocusing accuracy, especially for the slow-moving target and the moving target with only along-track velocity.