Research On Longitudinal Tear Protection Method Of Coal Mine Belt Based On Machine Vision

Belt conveyors are important transportation equipment in coal mines.Large gangue,anchor rods,wooden sticks and other foreign objects on the conveyor belt will not only cause belt tears,but also affect the quality of coal.In view of the problem of belt longitudinal tear protection,traditional research methods only focus on the detection of belt longitudinal tear,and ignore the cause of the accident.The accident cannot be prevented from the root cause,resulting in waste of production costs.This paper takes the precise detection of foreign matter on the coal flow surface and the longitudinal tear of the belt as the research object,and uses machine vision and deep learning algorithms to achieve dual protection of the longitudinal tear of the belt.Aiming at the problem of foreign object detection on coal flow surface,a foreign object detection algorithm based on unsupervised anomaly detection and attention mechanism is proposed.First of all,in order to realize the enhancement of downhole low-illuminance images,this paper improves the classic bright channel prior algorithm,and proposes a method of combining the bright and dark channels and adding suppression factors to the original transmission image;then,in order to achieve abnormal coal flow images detection,this paper improves the Skip-GANomaly algorithm,and proposes an unsupervised anomaly detection algorithm fused with attention mechanism;the experimental results show that the improved algorithm has improved comprehensive performance on the belt coal flow data set compared to the Skip-GANomaly algorithm The rate is 4.2%,and the detection speed is 20 ms per sheet,which meets the requirements of real-time accurate detection.Aiming at the problem of belt longitudinal tear detection,a longitudinal tear detection method based on improved Unet is studied.First,the image labeler tool and Augmentor algorithm are used to preprocess the data;then,the Unet-based deep semantic segmentation algorithm is used to detect the belt longitudinal tear.In order to further improve the detection speed of the algorithm,it is proposed to use deep separable convolution and reduce number of network layers and channels improves the algorithm structure;the experimental results show that the average intersection ratio of the proposed algorithm in this paper is 0.968,and the detection speed is comparable to that of the classic compared with the Unet algorithm,an increase of 38.5%.The coal mine underground belt longitudinal tear protection system designed in this paper can intelligently detect foreign matter on the coal surface and the belt longitudinal tear,and has a positive impact on the realization of the intelligent and informatization of the belt longitudinal tear protection.This topic contains 64 figures,9 tables and 94 reference.