Abstract: As the coal washing industry’s demand for intelligent dry coal preparation technology and coal gangue image recognition methods grows, it is becoming more and more important to study the recognition methods of coal gangue mixed feature images under complex coal washing conditions. Based on theories of deep learning, image recognition and wireless communication, a coal gangue recognition and location system was designed based on convolutional neural networks in this paper. According to the complex conditions of the coal mine washing process, the five state categories of the surface characteristics of coal gangue were analyzed to construct a gangue data set. The improved AlexNet network and RPN network based on transfer learning obtain the classification information and pixel coordinates of the gangue mixed feature image samples, and obtain the position coordinates of the pixel coordinates in the camera coordinate system through the camera calibration method. The local area network of the coal gangue sorting robot distributed control system was constructed to realize the real-time coal gangue detection information interaction between the identification and positioning system and the main control system. The detection model of coal gangue image wass tested based on the coal gangue recognition and positioning system. The test results show that the detection accuracy of the recognition model of the gangue recognition and positioning system can reach 90.17%, the maximum positioning error of the gangue target is 9.45 mm, and the system response time is less than 350 ms, which meets the basic requirements of the complex washing of coal mines. The coal gangue recognition model has good detection results on the mixed feature images of coal gangue, which provides a research foundation for the application of coal gangue image recognition method to the development of intelligent coal preparation.