隧道环境下基于深度学习的轻量级安全帽检测方法  

Method of lightweight helmet detection based on deep learning in tunnel environment

作　　者：高方玉 解玉文 张正刚 王道累 GAO Fangyu;JIE Yuwen;ZHANG Zhenggang;WANG Daolei(Science and Technology Innovation Department,China Electrical Equipment Group Co.,Ltd.,Shanghai 200436,China;Beijing Guowang Fuda Science&Technology Development Co.,Ltd.,Beijing 100070,China;College of Energy and Mechanical Engineering,Shanghai University of Electric Power,Shanghai 201306,China)

机构地区：[1]中国电气装备集团有限公司科技创新部,上海200436 [2]北京国网富达科技发展有限责任公司,北京100070 [3]上海电力大学能源与机械工程学院,上海201306

出　　处：《现代电子技术》2023年第14期147-151,共5页Modern Electronics Technique

基　　金：国家自然科学基金项目(61502297)。

摘　　要：隧道施工现场人员不按规定佩戴安全帽是事故发生的主要原因之一,使用安全帽检测算法能有效监督作业平台上所有人员安全帽佩戴的情况,及时作出风险预警,降低安全事故发生的可能。然而,工业上常用的安全帽检测算法计算复杂度较高,很难适用于隧道环境中的嵌入式移动设备,已有轻量级算法又很难在隧道光线差、背景复杂的条件下保持检测精确度。针对上述问题,文中提出一种基于改进YOLO_v3的轻量级安全帽检测算法,构建运算量较低的卷积模块LW_Conv,并以此改造主干网和特征金字塔。实验结果表明,改进算法的FLOPs约为YOLO_v3的10%,平均正确率(AP)比Tiny_YOLOv3高2%。One of the main reasons for accidents is that personnel on the tunnel construction site do not wear helmets according to regulations.The helmet detection algorithm can effectively supervise the wearing of helmet of all personnel on the working platform,and make timely risk warning to reduce the possibility of safety accidents.The commonly used helmet detection algorithms in industry have high computational complexity and are difficult to apply to embedded mobile devices in tunnel environments.Existing lightweight algorithms also find it difficult to maintain detection accuracy under conditions of poor tunnel lighting and complex backgrounds.To solve above problems,a lightweight helmet detection algorithm based on improved YOLO_v3 is proposed,and the convolution module LW_Conv with low computation is constructed,so that the backbone network and the feature pyramid are transformed.The experimental results show that the FLOPs of the improved algorithm is about 10%of YOLO_v3,and AP(average accuracy)is about 2%higher than Tiny_YOLOv3.

关 键 词：安全帽检测 轻量化卷积模块LW_Conv 隧道环境 改进YOLO_v3算法 深度学习 目标检测 

分 类 号：TN911.23-34[电子电信—通信与信息系统]