A Lightweight Algorithm Based On YOLOv5 For Relative Position Detection Of Hydraulic Support At Coal Mining Faces

Hydraulic support is one of the core equipment in the comprehensive mining face of coal mine.The orderly movement of hydraulic support group in the mining process is a necessary condition to ensure the safe and orderly mining of coal mine.For the possible abnormal position and posture of the hydraulic support,the current production relies on manual inspection to find and eliminate.At the same time,in view of the problems of the existing detection method of the hydraulic support moving position in the colliery comprehensive mining face,which is mainly based on the contact method,which depends on the complex detection equipment,and has poor flexibility and maintainability,this paper proposes a detection method of the hydraulic support moving in the colliery comprehensive mining face,LG-YOLO,which is based on the target detection,and uses the deep learning network model,The base part of the hydraulic support in the field of machine vision is extracted,and its position and posture are judged according to the coordinate information of each hydraulic support in the two-dimensional plane.The main work of this paper is as follows:(1)A displacement detection method of hydraulic support in fully mechanized mining face based on target detection is proposed.First,we do image enhancement processing based on double gamma function on the original image data collected from the fully mechanized coal mining face.Then the convolution neural network is used for feature extraction and target detection of the base of the hydraulic support.According to the obtained coordinate information of each hydraulic support,the relative position relationship values such as the relative deflection angle between adjacent hydraulic supports are calculated,so as to judge the position and posture of hydraulic support.(2)In order to facilitate the deployment of the model in the end equipment of the coal mine,this paper introduces the GhostNet convolution module and the Ghostresidual module in the YOLOv5s network to reduce the amount of parameters of the network model and the occupation of computing resources,and introduces the PRe LU activation function to achieve faster reasoning speed.In the post-processing stage,DIo U NMS is used to improve the accuracy of detection of closely aligned targets.(3)In order to further compress the model,this paper also prunes the convolutional neural network channel based on sparse regularization for the optimized model,conducts sparse regularization training for the parameters of the convolution layer and BN layer of the model,and obtains the neural network model with sparse weights,and then evaluates the importance of the two according to the sparsity of the filter and the characteristic scaling coefficient of BN layer,Finally,a structured pruning method is used to prune the sparse filter and its corresponding connections.The experiment shows that the algorithm in this paper can effectively detect the moving state of the hydraulic support,and at the same time,compared with YOLOv5 s,the model volume is reduced by 73%,and the calculation amount is reduced by 69%,which meets the requirements of end-side equipment deployment and real-time detection,and has great significance for the construction of less human and unmanned coal mining face operation.