Abstract: Existing coal safety supervision mostly adopts the mode of "people-oriented", which has problems such as strong subjectivity and high energy consumption. It can easily lead to negligent behaviors such as playing cell phones, smoking, sleeping on duty, and it is difficult to ensure that the coal mine safety production is carried out in a stable and orderly manner. In view of this, this paper utilizes the target detection technology, based on deep learning method, to study the intelligent recognition method of the working status of the personnel in the coal mine dispatching room. First, a YOLOv5-based method for recognizing the working status of personnel in coal mine dispatching room is constructed; Then, a convolutional attention module is introduced into the feature fusion sub-network in order to enhance the information extraction ability of the model in different channels and different spatial locations; Finally, the stepped residual convolution module with multi-scale feature fusion capability is used to enhance the model's ability to extract subtle features and fuse global features of personnel in the dispatching room. The experimental results show that the average accuracy rate of the full-state recognition of the method proposed in this article reaches 91.7%, and the average accuracy rate of small-scale action state recognition such as playing with mobile phones and smoking is 94.4% and 86.0% respectively, which verifies the effectiveness of the proposed model in identifying the working status of dispatchers.