Damage detection method for mine conveyor belt based on deep learning

In order to solve the problem that the current conveyor belt damage detection methods lack research on damage types other than conveyor belt tear, a damage detection method for mine conveyor belt based on deep learning is proposed. And the conveyor belt damage types are classified by the Yolov4-tiny target detection network. The Yolov4-tiny target detection network uses CSPDarknet53-tiny as the backbone feature extraction network, draws on the Resnet residual idea, uses residual blocks to prevent the loss of high-level semantic features in the deep network. At the same time, the method uses feature pyramid network to obtain the fusion of high-level and low-level semantic information to achieve the purpose of improving detection precision. The two effective feature layers in CSPDarknet53-tiny are input into the prediction network Yolo Head, and the prediction frames are filtered by the score ranking and non-maximum suppression algorithm to predict the types of conveyor belt damage. The experimental results show that the average precision of the Yolov4-tiny target detection network on the conveyor belt damage data set for the four damage types of surface scratches, tears, surface damage and breakdown is 99.36%, 94.85%, 89.30%, and 86.76% respectively, and the mean average precision is 92.57%. Compared with Faster-RCNN, RFBnet, M2det, SSD, Yolov3, EfficientDet and Yolov4 target detection networks, the Yolov4-tiny target detection network achieves the fastest detection speed on the data set with a frame rate of 101 frames/s. The network achieves better balance between speed and precision, and occupies relatively less computing resources. The detection of fresh samples outside the data set verifies that the method in this paper has good generalization ability.