Research On Mining Personnel Abnormal Behavior Detection Technology Based On Edge Vision

Coal energy serves as a crucial pillar for driving the economic development of our country.The abnormal behaviors of mining personnel can easily trigger production accidents,thus affecting the safety operation of the mining site.The traditional video surveillance mode that relies on human labor is inefficient,while video detection methods based on high-performance servers suffer from issues such as excessive computation and network transmission traffic,which are not suitable for meeting the actual needs of mining operations.Therefore,this thesis conducts an in-depth analysis of the characteristics of abnormal behaviors of mining personnel,and established an intelligent mining personnel safety detection platform based on edge devices.This platform aims to achieve rapid perception and real-time detection of abnormal behaviors of mining personnel.The main research contents are as follows:(1)In response to the issues of high computational complexity and limited detection types in existing behavior recognition methods,this thesis proposes a cascaded analysis algorithm based on spatial features and time series for detecting abnormal behaviors of mine personnel.Firstly,a spatial feature-based abnormal behavior detection method is employed to classify the behaviors of mining personnel in surveillance images.Then,the behavior classification results are input into a time series queue for analyzing the long-term behaviors of mine personnel.Finally,the abnormality level of mining personnel’s behaviors is determined based on the degree of abnormality.Experimental results on real-world video data from mining enterprises validate that the proposed cascade analysis algorithm can effectively improve the accuracy of abnormal behavior detection.(2)In response to the issues of large model parameters,high computational complexity,and slow inference speed of YOLO-v7-tiny model in the task of abnormal behavior detection of mine personnel,this thesis proposes a more suitable RE-YOLO model for edge devices.Firstly,Rep Block is introduced to replace the efficient layer aggregation networks module,effectively reducing the model’s parameters and computational complexity,and improving the detection speed of the model.Then,efficient channel attention module is embed at key positions in the backbone network to improve the detection accuracy of the model.Finally,experiments conducted on edge devices including Jetson Xavier NX and Huawei Atlas 500 validate that the proposed RE-YOLO model outperforms YOLO-v7-tiny in terms of computational complexity,model parameters,and inference speed,achieving fast detection of abnormal behaviors of mining personnel.Finally,based on the aforementioned algorithm and model optimization strategies,this thesis designs and develops a mine personnel abnormal behavior detection system based on edge vision,which is deployed on Huawei Atlas 500.The system mainly includes modules for monitoring device management,real-time video preview,mine personnel abnormal behavior alarm,and abnormal behavior information statistics,providing effective guarantees for safety production in mining enterprises.This thesis has 57 figures,18 tables,and 85 references.