Design And Research Of Personnel Violation Recognition System In Mining Rock Burst Hazardous Area Based On Edge Intelligence

With the continuous increase of coal mining intensity,especially after entering the deep mining stage,rock burst disasters occur frequently.In order to reduce disaster losses,the state has issued relevant regulations,requiring people in the rock burst hazardous area to wear personal protective equipment,establishing and strictly implementing the regional personnel limit system.However,in recent years,there are still many rock burst disasters in which the personnel anti-burst measures are lacking and the number of miners is over the limit.The lack of supervision of violations has led to increased casualties among miners.In this thesis,static violations(no＿helmet and no＿clothes)and dynamic violation(personnel over the limit)recognition model is proposed,and the personnel violation recognition system based on edge intelligence is designed.The main research work of this thesis is as follows:First of all,Impact Area Training Dataset is built up.The monitoring video data of coal mine enterprises rock burst hazardous area is collected.After analyzing and processing these video data frame by frame,head-shoulder targets and static violations are annotated.Since there are very few static violations,the personal dataset suffers from the problem of class imbalance.Therefore,this thesis adopts a variety of data enhancement methods and collects similar coal mine image data and safety helmet image data for dataset construction and expansion.The experiment results verify the availability of data enhancement and expansion for alleviating the problem of class imbalance.Secondly,the object detection algorithm is used to recognize the static violation.In order to improve the recognition accuracy,this thesis proposes a YOLOv5＿CA＿FL model for static violations recognition,which is improved from two aspects:backbone network and loss function.On the one hand,Coordinate Attention Module is introduced into the backbone network to improve the model’s ability to capture channel and location information.On the other hand,the focal loss function is used to improve the model’s attention for few samples and difficult samples.The experimental results show that YOLOv5＿CA＿FL proposed in this thesis improves the detection accuracy of static violations and head＿shoulder targets,and the AP of difficult samples such as no＿clothes targets is greatly improved.At the same time,because the lightweight model is selected as the benchmark model,it can meet the real-time detection requirements in coal mine.Then,the object tracking algorithm is used to count the number of people in and out of the rock burst hazardous area,so as to recognize dynamic violation that personnel over the limit.In order to meet the real-time requirements of edge devices,a lightweight re-identification feature extraction network OSNet is used to optimize the Strong SORT tracker.The experimental results show that the people counting method of tracking-by-detection paradigm proposed in this thesis has high recognition accuracy for the low density scene of the rock burst hazardous area.And the result of people counting can be used to recognize that personnel over the limit.Finally,the personnel violation recognition system in rock burst hazardous area is designed.Edge computing and artificial intelligence are combined to complete the edge intelligent analysis and processing of monitoring video in mining rock burst hazardous area.In this system,the underground part completes the analysis and processing of the video streaming,so as to realize the display,control,warning and other functions according to the edge device processing results.The dispatcher realizes supervision and control through the ground visual platform of the system.The personnel violation recognition system based on edge intelligence can effectively improve the intelligent management level of rock burst hazardous area and reduce the loss after the disaster.