Abstract: In order to implement the national deployment requirements for the prevention and control of major risks in non-coal mines，and to solve the problem that the traditional video monitoring system of metal underground mines only has the function of post-event replay，lacking real-time warning and intelli⁃ gent recognition capabilities，based on the engineering background of Huaibei Dongxin Mining，an AI vid⁃ eo intelligent auxiliary supervision system based on deep learning was designed and constructed. The sys⁃ tem covers key areas underground and on the surface，and deploys 20 AI intelligent cameras，integrating 12 types of recognition algorithms，achieving real-time monitoring and warning of personnel behavior， equipment status，and environmental risks. The overall architecture，functional modules，key technolo⁃ gies，and engineering deployment requirements of the system were studied and analyzed，and the system was comprehensively tested in combination with the technical guidance document of Anhui Province. The trial operation results show that the recognition accuracy of the system for violations is above 95%，the alarm response is timely，and the operation is stable. This system，based on the integration of deep learn⁃ ing and edge computing，realizes closed-loop management of intelligent video recognition in multiple un⁃ derground scenarios，provides a replicable technical solution and engineering example for the intelligent supervision of non-coal mines.

Key words: AI video recognition, intelligent mine, edge computing, behavior recognition, risk warning