An enhanced SiamMask network for underwater target tracking in sonar

In order to solve the problem of underwater noise interference and target boundary blurring in sonar image imaging, this paper proposes a new underwater target tracking method based on SiamMask. By combining hybrid attention and cross-correlation mechanisms to enhance the network’s ability to perceive underwater target boundaries, the method mitigates the interference caused by noise. Furthermore, a ranking loss optimization strategy is employed to impose additional constraints on the original loss of the network. In particular, the discrepancy between the classification and regression branches of the network is diminished by integrating the prospective confidence scores of the positive samples with the IoU (intersection over union) values, which effectively mitigates the risk of misalignment. The evaluation results indicate that the proposed method has achieved leading performance in both self-made sonar datasets and public sonar datasets, and the model can achieve competitive underwater contour tracking tasks for sonar images at a fast speed.