Abstract: Due to the high speed of the conveyor belt,the large variations in the shape and color of the coal flow,and the complex lighting conditions,traditional methods for detecting the coal flow on conveyor belts often suffer from low accuracy and are prone to interference. Therefore,a coal flow detection algorithm based on super-resolution depth image restoration is proposed. This algorithm uses YOLOv3 as the basic framework and combines a super-resolution depth image restoration model to process blurred and noisy coal flow images. The image restoration model,through an encoder-decoder structure,extracts and repairs the features of damaged images while preserving shallow texture information and passing it to deeper layers. The processed clear coal flow images are then detected using the YOLOv3-based object detection algorithm. Experimental results from a coal processing enterprise in Northern China show that when the image damage rate is 50%,the proposed image restoration model improves the structural similarity by 10% compared to the mutual encoder-based image restoration model. Compared to YOLOv4-tiny,the proposed coal flow detection algorithm increases the processing speed by 56 fps,demonstrating that this algorithm can effectively improve the efficiency of coal flow detection on conveyor belts. 

Key words: object detection,conveyor belt,coal flow,super-resolution,image restoration,deep learning