Target Detection in Sea Clutter via Contrastive Learning
Senlin Xia, Yukai Kong, Kui Xiong, Guolong Cui
Abstract
This paper considers the target detection problem using limited labeled samples in the nonhomogeneous sea clutter environment and proposes an effective method of radar visual representations via contrastive learning and its application on target detection. First, the signal features of radar echo segments are extracted by contrastive learning without supervised information. Second, the classification results can be obtained through supervised training the fully connected layer with a small number of labeled samples, and thus the target can be identified in that segment. Finally, the performance of the proposed method is evaluated via measured and simulated data. The results show that in the cases of training with a large amount of unlabeled data and few labeled samples, the proposed method can effectively extract features of clutter and targets and achieve better classification performance compared to state-of-the-art methods.