Self-Supervised Contrastive Learning on Cross-Augmented Samples for SAR Target Recognition

Deep learning has led to significant improvements in synthetic aperture radar (SAR) target recognition and further optimized the deployment in real SAR applications. Most SAR target recognition methods are based on supervised learning and require a large number of labeled training SAR data. Due to the time-consuming and laborious work of labeling samples, only a tiny part of the existing radar samples are labeled data, and there is still a large amount of unlabeled radar data. Therefore, relying only on the data inherently without label information, self-supervised contrastive learning is quite prospective for advancing SAR target recognition. In this paper, we propose a self-supervised contrastive learning method via learning feature representation of cross-augmented samples for SAR target recognition. It contains a pre-train self-supervised contrastive learning network to learn a similar feature representation for targets of the same class without the labeled samples, and a downstream classification network for the target recognition task. Our model is pre-trained with a large amount of unlabeled data to achieve better clustering of features of the same target, benefitting the subsequent target recognition. Experiments are conducted on the moving and stationary target acquisition and recognition (MSTAR) benchmark dataset, and the experimental results indicate that the proposed method can effectively extract target features and achieve promising recognition performance.