Ground Moving Target Detection for Multichannel SAR System Based on Subdomain Adaptation

With the ability to extract potential features and learn rules automatically, deep learning can successfully use labeled data to train networks and achieve target detection. Due to the lack of labeled data in the field of radar, simulated data are usually employed to support the training of networks, which is proved to be a feasible way in practice. However, there are inevitable distribution differences between simulated data and real radar data, which always leads to a performance loss of the network. Motivated by recent advances in transfer learning, this paper proposes a new method for ground moving target detection of multichannel synthetic aperture radar (SAR) systems based on subdomain adaptation. In our method, a detection network is built on the basis of ResNet18 to extract the features of multichannel radar data, and the distribution differences between domains are measured by multi-kernel local maximum mean discrepancy (MK-LMMD). Experiments on real SAR data demonstrate that the proposed method significantly improves the performance of ground moving target detection.