Abstract: In order to prevent potential hazards and ensure coal mine safety, it has become an important way to improve the level of mine safety management to detect the behavior of mine operators. In view of the low accuracy of the commonly used intelligent detection methods, a detection algorithm for miner behavior based on S3DD-YOLOv8n was proposed. Firstly, in order to extract the time information of the video data and maintain the time continuity, 3D dilated convolution was introduced into the backbone network of YOLOv8n to improve the data enhancement algorithm. Secondly, the SE (Squeeze & Excitation) attention mechanism was introduced to improve the network's attention on key information. Finally, the deformable convolution was introduced to improve the fitting degree of the model to the miner behavior. According to the experimental verification of DsLMF+ data set, the average accuracy mAP50 of the algorithm reaches 97.0%, which is 4.0% higher than that of YOLOv8n. At the same time, the accuracy rate P and the regression rate R increased by 12.9% and 7.0% respectively, reaching 92.5% and 90.4%. The algorithm can detect the miner behavior efficiently and accurately.