Abstract: In order to solve the problem that the operator has insufficient or excessive antifreeze spraying due to mistakes and fatigue in the process of coal mine coal transmission, resulting in the adhesion or slippage of coal and the conveyor belt, and then causing production accidents, an automatic antifreeze spraying system for coal mine conveyor belt based on visual monitoring was developed. Through the detection, recognition and classification of on-site conveyor belt images by edge algorithms, the system issues remote control instructions to the antifreeze valve to complete the automatic spraying of antifreeze. A u-net-based image data augmentation method (U-NHME) was introduced to augment the original dataset samples, and then YOLO-V7 was used as the network for target localization and recognition, so as to achieve accurate recognition of all-weather outdoor images. Map and other evaluation indexes were used to enhance, train and recognize images with different coal quantities. Experimental results show that compared with the original YOLO-V7 network, the recognition accuracy of the proposed algorithm is improved by 2 percentage points, and the recognition accuracy of conveyor belt coal quantity is improved. The system is highly reliable and scalable, bringing a more efficient, safe and environmentally friendly production method to the coal industry.