Methods, Data, and Applications of Artificial Intelligence in Underwater Image Enhancement and Restoration

Underwater vision technology serves as the core foundation for enabling “perception and interaction” in intelligent underwater robotic systems; however, its performance is severely constrained by image degradation. This paper provides a narrative review of research achievements in the field of underwater vision over the past decade, covering research approaches in mechanisms, methods, data, and applications. First, we analyze the physical models of underwater image degradation and elucidate its causes. Building on this foundation, the paper focuses on reviewing the algorithmic contributions of artificial intelligence to underwater image enhancement and restoration, summarizing multiple mainstream datasets such as UIEB and EUVP, as well as image quality evaluation metrics. Furthermore, through multi-benchmark experiments, we compare representative methods in terms of image quality, robustness to degradation, computational efficiency, and downstream visual tasks. The experimental results indicate that the performance of different methods varies depending on the dataset, evaluation metrics, and degradation conditions, and that the performance rankings differ across various evaluation metrics. Finally, this paper discusses key challenges, such as the discrepancy between objective evaluation and subjective perception, as well as model robustness and environmental adaptability, and proposes future research directions to provide guidance for the development of underwater image enhancement and restoration technologies and task-oriented evaluation.