Navigation safety is an important guarantee for the efficiency of water transportation. According to statistics, the proportion of water traffic accidents caused by human factors is as high as 80%. Among them, the crew members do not abide by the safety regulations, smoking, playing mobile phones while on duty and other behaviors seriously affect the safe navigation of the ship. At present, the traditional video surveillance system on board has certain limitations. It only uses the video terminal to monitor the fixed area, and completely relies on the background management personnel to supervise the surveillance video. This not only wastes a lot of personnel, but also fails to realize the front-end identification and alarm functions, which makes it difficult to effectively control the unsafe behavior in the first time. Based on this, based on the classical distributed system architecture, the project proposed a distributed ship supervision architecture based on front-end intelligence through the division of ship cabin functions, and embedded the improved YOLOv5s algorithm to greatly improve the identification speed and achieve front-end real-time alarm function. The designed distributed data storage node matches the terminal video equipment, which can not only realize the interactive communication between adjacent databases, but also trigger the buzzer alarm to remind the crew to stop the unsafe behavior after monitoring and identifying the unsafe behavior, and record and collect evidence. Finally, this project selects a scene inside a ship and takes the crew playing mobile phones during the bridge watch as an example to verify the proposed distributed identification system. The results show that the distributed intelligent ship supervision system based on front-end intelligence can effectively monitor the unsafe behaviors of the crew, and the identification accuracy is higher than 93%. Compared with the traditional depth recognition architecture, the volume of the improved algorithm is reduced by about 22%.