In view of the harsh working environment of coal mine and the difficulty of realizing real-time risk identification and early warning by traditional manual supervision methods, this paper develops a set of security risk early warning system driven by video intelligent analysis technology. The system integrates key technologies such as downhole video quality optimization, multi-spectral and multi-scale feature fusion detection, and cloud edge-end collaborative computing. By constructing a full-chain management and control mechanism from information perception, intelligent judgment to hierarchical early warning, the automatic identification and immediate intervention of unsafe behaviors of operators are realized. The system adopts the architecture of WEB client, PC client and mobile APP client, and relies on Java distributed microservice framework, mixed data storage strategy and message queue technology to ensure the high efficiency of a large number of video data in the processing process and the stability of transmission. The field application results show that the platform can significantly improve the automation level and response time of coal mine safety supervision, and provide a feasible technical path for mine intelligent safety management.