Conveyor belt damage detection based on machine learning

Abstract

In view of the damage problems of the conveyor belt, which is prone to scratches, tears, and ruptures during long-term operation, machine vision technology can be used for efficient detection, and t-SNE dimension reduction processing is introduced into the visual extraction feature network to solve the clutter of visual feature data, so as to facilitate the efficient use of data. According to the damage condition of conveyor belt, an improved Faster R-CNN method for surface damage detection of conveyor belt is proposed. Based on the Faster R-CNN neural network, the detection method preferred MobileNet network for image lightweight feature extraction, and then introduced the background classification of the fusion of anchor original features and convolution into the RPN module to enhance the damage feature information of the conveyor belt. Finally, data sets of conveyor belt surface damage were constructed for data test, and VGG-19 and ResNet-18 backbone networks were used for test comparison, respectively. The results showed that the improved Faster R-CNN algorithm could effectively identify the damage states of conveyor belt scratches, tears and ruptures.