基于改进YOLOv5s的矿工防护装备检测模型研究  

Research on Detection Model of Miner Protective Equipment Based on Improved YOLOv5s

作　　者：张翥 杨玉中[2] ZHANG Zhu;YANG Yuzhong(School of Business Administration,Henan Polytechnic University,Jiaozuo 454003,China;School of Energy Scienceand Engineering,Henan Polytechnic University,Jiaozuo 454003,China)

机构地区：[1]河南理工大学工商管理学院,河南焦作454003 [2]河南理工大学能源科学与工程学院,河南焦作454003

出　　处：《煤炭技术》2024年第10期242-246,共5页Coal Technology

摘　　要：为解决井下矿工由于安全意识淡薄而不正确使用个人防护装备(PPE)的问题,提出一种实时检测矿工是否佩戴PPE的目标检测网络模型。首先,采用现场监控视频采集和数据爬取的方式收集数据,使用半监督学习方法标注图像并利用图像增强技术扩充数据,构建井下矿工PPE数据集;其次,基于深度学习的方法建立PPE检测模型,将YOLOv5s的主干网络替换为MobileOne来降低网络参数,提升检测速度;最后,在特征融合层中嵌入三维注意力机制SimAM,实现对复杂环境下的PPE检测。实验结果表明:改进的模型平均检测精度达到94.4%,在CPU端的检测速度达到67帧/s,可以满足井下实时检测的要求。In order to solve the problem of incorrect use of personal protective equipment(PPE)by underground miners due to their weak safety awareness,a target detection network model for real-time detection of PPE is proposed.Firstly,the PPE data set of underground miners is constructed by using on-site surveillance video acquisition and data crawling to collect data,using semi-supervised learning method to label images and image enhancement technology to expand data.Secondly,the PPE detection model was established based on deep learning,and the backbone network of YOLOv5s was replaced with MobileOne to reduce network parameters and improve detection speed.Finally,three-dimensional attention mechanism SimAM is embedded in the feature fusion layer to realize PPE detection in complex environments.The experimental results show that the average detection precision mean Average Precision of the improved model reaches 94.4%,and the detection speed of the CPU side reaches 67 f/s,which can meet the requirements of downhole real-time detection.

关 键 词：深度学习 目标检测 个人防护装备(PPE) 井下矿工 轻量化 YOLOv5s 

分 类 号：TD79[矿业工程—矿井通风与安全]