Research On Foreign Object Monitoring System In Coal Flow Of Belt Conveyor Based On Deep Learning

Belt conveyor plays an important role in coal transportation,so the smooth operation of the belt conveyor is the key to ensure the safe and efficient production of coal.In the process of coal mining,gangue from the interlayer of rock and iron from machine vibration will enter the belt conveyor system along with the coal flow.This will cause coal piling,belt scratches,tears and other safety accidents,affecting the safety of coal mine production.At present,the detection of foreign objects in coal flow of belt conveyor mainly relies on manual detection or special sensor detection,which is time-consuming and laborious,and the detection effect is difficult to meet the accuracy requirements of harsh working conditions in coal mines.In view of the above problems,this thesis studies the monitoring technology based on deep learning for foreign objects in coal flow of belt conveyor,and uses machine vision to realize highprecision and real-time detection of foreign bodies,so as to ensure the safety of coal transportation.The main works of this thesis are as follows:Firstly,the image enhancement method of low-light coal flow foreign object based on self-calibrated illumination learning is studied.Due to the dim lighting in the coal mine and the influence of dust,water mist and other factors,the collected images of foreign objects in the coal flow of belt conveyor have low illumination,and the features of foreign bodies are not obvious.Therefore,it is necessary to preprocess the images to enhance the features of foreign objects.In contrast to the shortcomings of traditional low light image enhancement methods,the self-calibrated light learning algorithm is introduced to enhance the low light coal flow foreign object image.The effectiveness of the low light coal flow foreign object image enhancement method based on selfcalibrated light learning is verified by designing comparative tests on the self-built low light foreign object image enhancement data set.Then,the method of coal flow foreign object identification of belt conveyor based on improved YOLOv5 is studied.By analyzing the performance of three current mainstream target detection algorithms in terms of detection accuracy and running speed,the detection model of foreign object in coal flow of belt conveyor based on YOLOv5 is constructed.Aiming at the deficiency of YOLOv5 in detection accuracy,an optimization strategy combining prior box estimation optimization,fusion of attention mechanism and improvement of loss function was constructed.Firstly,the binary K-means algorithm is used to re-estimate the prior box,and the new prior box matches the target box to a higher degree.Secondly,coordinate attention mechanism is introduced into the feature extraction network of YOLOv5 to enhance the feature extraction ability of the model.Finally,the classification loss function is replaced by the zoom loss function to improve the convergence rate of the model training loss.Experimental results show that the optimized algorithm can improve the detection accuracy of foreign object identification data set in coal flow of belt conveyor,and reduce the occurrence of foreign body leakage and false detection while maintaining the running speed.Finally,the foreign object monitoring system of coal flow in belt conveyor is designed.Through the hardware selection,the construction of the coal flow simulation environment of the belt conveyor is completed.The software platform of the coal flow foreign object monitoring system of the belt conveyor is designed by using Python language combined with Py Qt5 and Open CV library,and Tensor RT is used for reasoning acceleration and deployment of the model.In the simulated environment of coal flow,the effect test of coal flow foreign object monitoring system of belt conveyor is carried out.The experimental results show that the system designed in this thesis can realize the high-precision and real-time detection of foreign body,which verifies the feasibility of the system.There are 71 figures,18 tables and 100 references in this thesis.