Abstract: In the process of underground operation safety monitoring in coal mine, it mainly relies on fixed sensing detection, and the monitoring data transmission is difficult, which leads to the decrease of abnormal operation safety monitoring and early warning frequency. Therefore, this paper proposes the design and analysis of underground operation safety monitoring method integrating pyramid structure and attention mechanism. The local coordinate system is established in the coal mine and monitoring nodes are deployed. The monitored space is determined through the coordinate system and the local monitoring node. Within this space, the local coordinate system is converted to the global coordinate system, and the image collected by the nodes in the global coordinate system is classified as the applicable conditions. By using the collected images, the spatial features are extracted, the spatial attention is clearly defined, and the key monitoring area is mapped with the guide of the features, and the mean square error of security monitoring between points in the global coordinate system is calculated in this area. Add dynamic warning rule, according to the change of monitoring mean square error, use the rule to determine whether the monitoring situation is abnormal, output the final monitoring result. The experimental results show that the monitoring and warning frequency of operation anomaly safety obtained by the proposed method can all reach more than 205 Hz, which indicates that the proposed method has a fast perception speed and strong superiority in the abnormal perception of operation reality.