Abstract: This paper designs a coal mine safety hazard warning method based on intelligent video analysis, develops a coal mine safety hazard warning platform, and realizes the real-time detection and active warning of personnel unsafe behavior.The platform adopts advanced video coding and decoding technologies and streaming media transmission protocols to ensure real-time and efficient transmission of underground monitoring video, and integrates an underground low-quality image enhancement component, which improves the image quality through contrast enhancement, de-fogging, noise reduction and other algorithms.Secondly, a miner detection model based on multispectral multi-scale feature fusion with depth saliency detection is used, which is combined with multispectral image fusion technology to improve the model recognition accuracy and robustness.Finally, the violation event warning component is developed to realize the sound and light warning of multiple violations and other safety hazards.The effect of on-site application shows that the platform builds a comprehensive safety warning chain of “before-the-fact-after-the-fact” and realizes intelligent judgment, dynamic assessment, and monitoring of human unsafe behaviors through the three-end fusion of “WEB-end-PC-client-APP-end” and “WEB-end-PC-client-APP-end” warning mode.It realizes intelligent judgment, dynamic assessment and real-time warning of human unsafe behaviors, enhances the awareness of personnel's safety and strengthens the ability of coal mine safety supervision.