| With the continuous growth of my country’s natural gas consumption demand,ensuring the safe and efficient production of natural gas mining enterprises is of great significance to alleviate my country’s energy demand.Domestic natural gas mining and construction operation areas are usually harsh and complex and changeable.Whether workers at heights wear safe belts is potentially dangerous to safety production.Therefore,how to efficiently detect whether workers at heights wear safe belts in real time is an urgent need for safe production in gas mines.solved problem.However,at present,safety supervisors generally use manual patrols to supervise workers at heights who do not wear safe belts in construction work areas of gas mines,and it is impossible to obtain the real-time status of each worker at heights wearing safe belts in time.In view of the above problems,this paper designs and develops a real-time detection system based on deep learning for the wearing of safe belts for high-altitude operators in gas mine operation areas to ensure the life safety of high-altitude operators in gas mine construction and operation areas.The main work and innovations of this paper are summarized as follows:(1)Safe belt dataset production: Due to the complex background of the gas mine construction site,the general object detection algorithm lacks scene adaptability,and its performance is easily affected by the environment.In response to the above problems,field inspection and data collection were carried out on the gas mine construction site.The image data collected on site and some image data collected through the Internet were marked and converted into standard VOC2007 format data sets using the Labelme tool.Perform data augmentation and preprocessing to meet algorithm training requirements.(2)Aiming at the problems that the detection results of traditional object detection algorithms are prone to a large number of repeated detection frames and low detection accuracy for small targets,a safe belt wearing detection algorithm YOLOSafebelt is proposed.First,YOLOSafebelt integrates ECA(Efficient Channel Attention)attention mechanism and DO-Conv convolution in the backbone network,so that the network can improve the object positioning ability of the network while filtering useless feature information;then,Pixel Shuffle upsampling is introduced on the PANET feature fusion network strategy,and design three dilated parallel feature enhancement modules(DPM＿S,DPM＿M,DPM＿L)to improve the YOLOSafebelt network’s ability to detect small objects.Experiments show that the proposed YOLOSafebelt detection network has strong feature extraction ability for safe belts,and has higher object detection accuracy and robustness.(3)There will be some ground operators on the monitoring screen of the gas mine operation area,which will lead to the problem that the algorithm will cause false detection of the ground personnel.This chapter proposes a method of using the "marking and fixing method " to distinguish high-altitude operators and ground operators.First,the well-designed "marks" that are easy to be detected by the algorithm are placed on the construction site,then the YOLOSafebelt network is used to identify the "marks",and finally a horizontal contour line is drawn according to the markers to determine whether the operator is working at a height.The "marking and fixing method" stipulates that the center of the object frame of the operator is higher than the horizontal contour line,that is,the operator at high altitude,and vice versa,it is the operator on the ground.(4)Based on the YOLOSafebelt network and the "marking and fixing method",a B/S mode real-time detection system for safe belt wearing is built.The system includes a video stream module,an image processing module,a system alarm module and a data storage module.Experiments show that the system can stably and efficiently perform real-time detection and alarm on workers not wearing safe belts at the gas mine construction site. |
PDF Full Text Request