Abstract: ‍ ‍Underwater observations are one of the most intuitive means of ocean exploration. Images acquired in underwater observations are not always sufficient for the observation needs due to the influence of underwater optical and acoustic properties as well as clutter and aquatic creatures. The effective processing， analysis， and application of underwater images is a challenging topic. Although image processing and computer vision techniques have been extensively studied in atmospheric environments， given the differences in imaging principles and application contexts， the algorithms proposed for atmospheric natural images cannot be directly transferred to underwater tasks. And the visual applications proposed for underwater scenes often suffer from inadequate consideration of the task context and poor generalization. This paper addresses two main tools of underwater observation： optical images and acoustic images， starting with underwater image characteristics， provides a more complete summary and analysis of the working ideas of underwater observation solutions by sorting out successful underwater image processing and image quality evaluation cases at home and abroad in a task-oriented and demand-based approach for the first time. In addition， this paper discusses the development process of underwater machine vision applications， explored the development prospects and optimization directions of related fields in detail， and brings new ideas for breaking the bottleneck of marine vision applications and building the Smart Ocean system.