Abstract: According to the characteristics of underground drilling scene in coal mine, the recognition methods based on the image features of coal-rock and the feedback of drilling parameters were selected. In order to improve the accuracy of coal-rock interface recognition, the deep learning model was used to realize the coupling recognition of two heterogeneous information. The coupling recognition model was constructed based on neural network mode.Based on CNN convolution neural network, the depth residual neural network model was extended, and its structure was mainly composed of backbone network, neck network and head network. Through the training of database model, the final recognition method was obtained. Finally, accuracy of the identification method was verified by in-situ drilling test in underground coal mine, and the recognition accuracy of coal-rock interface was up to 92%.