Explainable underwater target recognition models: principles, methods, and applications

With the increasing strategic importance of the ocean, underwater intelligent systems have become essential for signal processing, target recognition, and autonomous navigation. The widespread application of deep learning has significantly advanced underwater acoustic missions, but its “black box” nature has led to critical concerns about decision explainability, limiting its trustworthy application in high-risk scenarios. This paper provides a systematic review of explainable models for underwater target recognition, elaborating on the core concepts and main methods of explainability. It also reviews research progress and representative achievements in sonar imaging, signal analysis, and autonomous navigation. Finally, future directions, including causal reasoning, cross-modal collaboration, and physical knowledge integration, are identified to provide a reference for developing safe and reliable underwater intelligent systems.