Wear State Detection of Conveyor Belt in Underground Mine Based on Retinex- YOLOv8-EfficientNet-NAM
Lijie Yang,Guangyu Chen,Jiehui Liu,Jinxi Guo
DOI: https://doi.org/10.1109/access.2024.3363834
IF: 3.9
2024-02-23
IEEE Access
Abstract:The belt surface of the mine belt conveyor can cause serious wear under the condition of long-term high-load operation, which can have a negative impact on production, bring economic losses, even endanger personal safety, and cause serious production accidents. Manual detection requires a lot of manpower and material resources, and is highly dependent on empirical judgment, which is with low efficiency and security risks. Therefore, in this study, we introduce a new conveyor belt wear detection algorithm Retinex-YOLOv8-EfficientNet-NAM (RYEN algorithm) based on deep learning and machine vision technology to replace manual detection, improving detection efficiency and recognition accuracy. The wear degree of belt is reclassified and defined according to the mechanical properties and wear texture characteristics of belt with different wear degrees, and a new special data set for belt wear detection is established. Aiming at the low brightness, high noise and complex working conditions of the underground mine, Gaussian filtering and bilateral filtering are used as the central surround function of the improved Retinex algorithm, and then channel fusion is performed with the image after histogram equalization and adaptive brightness adjustment. The improved Retinex multi-image fusion algorithm is used to preprocess the collected image. EfficientNet has the performance of reasonably allocating the input resolution, network depth, and channel width, and can maximize the performance of the network with limited resources. EfficientNet is used to replace Darknet53 of YOLOv8 as the backbone of the feature extraction network, which improves the detection accuracy under limited computing resources. A lightweight attention module NAM is added to the improved network, which improves the detection speed without reducing the detection accuracy. Experimental results show that RYEN algorithm effectively maintains the smoothness of the image during the image preprocessing stage, improves the brightness and contrast of the image, and better preserves the edge information of the image. RYEN algorithm achieves a detection speed of 66FPS and an average accuracy of 98.57%. Compared with the original YOLOv8 algorithm, the accuracy of RYEN algorithm is increased by 6.4% and the speed is increased by 13.2%. In comparison experiments with similar methods, RYEN algorithm occupies less hardware resources, has strong generalization ability, good performance, and has high detection speed and accuracy.
computer science, information systems,telecommunications,engineering, electrical & electronic