Research On Foreign Objects Detection Of Mine Belt Conveyor Based On Deep Learning

As the main production equipment in coal mine,belt conveyor is widely used in underground working place,transportation lane,loading station,ground coal building and other production scenes.Foreign objects such as gangue or anchor bolts in raw coal are also easy to cause coal transport equipment clogging,scratch the belt or cause belt tearing,even lead to coal mine safety accidents.At present,most of the above problems are solved by manual inspection,which has the disadvantage of high labor intensity,low efficiency and high detection rate.The thesis researches the foreign objects detection methods of mining belt conveyor based on deep learning to improve the accuracy and speed of detection,accomplish the real-time detection of foreign objects in conveyor,and reduce the loss caused by belt failure.The research in this thesis is as follows:(1)The thesis adopts adaptive median filter denoising and contrast limited adaptive histogram equalization for image enhancement to solve the problems of poor sharpness and low brightness which caused by the dark of downhole environment to provide experimental basis for the follow-up research,and studies the two-stage target detection algorithm Faster R-CNN and the one-stage target detection algorithm YOLOv5 based on deep learning.Experimental results show that YOLOv5 has advantages in detection speed while considering detection effect,so the thesis selects it as the detection algorithm for subsequent improvement.(2)The loss function CIOU in the process of network training only considers the difference between the width and height ratio of the prediction frame,it ignores the real width and height relationship between the prediction frame and the real target frame,which affects the convergence effect of the model.At the same time,the network uses a large number of convolution operations to increase the number of parameters and the amount of computation.To resolve the problems above,the thesis designs an attention mechanism into the network to strengthen the model’s attention to locally important feature information;uses the lightweight Ghost convolution to replace part of the original network of the general convolution to reduce the number of parameters and computation and quicken the rate of reasoning;Focal-EIOU was used to improve CIOU loss function.EIOU carries out regression on the width and height of the prediction frame,minimizes the width and height difference between the detection frame and the real frame,obtains better positioning effect and balances positive and negative samples in the training process.According to results of the multiple experiments,the detection accuracy and speed of the improved YOLOv5 model have been improved.(3)The foreign object detection system of mine belt conveyor based on Py Qt5 is designed to accomplish real-time monitoring of foreign objects in the belt during operation,and record and save the information of foreign objects for subsequent inspection and analysis.