深度学习下吊装作业工人防护装备及吊钩检测方法  

Protective equipment and hook testing methods for workers in lifting operations under deep learning

作　　者：李华[1] 薛曦澄 吴立舟 王岩彬 钟兴润[1] LI Hua;XUE Xicheng;WU Lizhou;WANG Yanbin;ZHONG Xingrun(College of Resources Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China;Shaanxi No.11 Construction Engineering Company,Xianyang 712000,Shaanxi,China)

机构地区：[1]西安建筑科技大学资源工程学院,西安710055 [2]陕西建工第十一建设集团有限公司,陕西咸阳712000

出　　处：《安全与环境学报》2024年第3期1027-1035,共9页Journal of Safety and Environment

基　　金：陕西省建设厅科技发展计划项目(2020-K32);西安建科大工程技术项目(XAJD-YF23N010)。

摘　　要：为解决危大工程中吊装作业安全管理的问题,基于深度学习构建目标检测算法(You Only Look Once version 5,YOLOv5)网络模型,针对进入吊装作业区域内人员的防护装备进行多目标融合检测,并对吊钩在施工过程中的状态进行检测。在原始的检测网络模型中引入4种注意力机制,并通过5种训练模型的结果对比分析,进而选择卷积块注意力模块(Convolutional Block Attention Module,CBAM)最优模型。优化后的检测模型对安全帽的平均识别精度达86.5%,对反光衣的平均识别精度达83.0%,对吊钩的状态识别精度达92.0%。将训练好的人员检测模型和吊钩检测模型打包成exe执行文件,应用到施工安全管理人员的中控平台,可帮助管理人员更好地判断吊装作业的工作情况,进而及时进行风险管控。To solve the problem of safety management of lifting operations in hazardous large projects,this paper constructs a YOLOv5(You Only Look Once version 5)network model based on deep learning.It performs multi-objective fusion detection of the protective equipment of persons entering the hoisting area and detection of the status of the hook during the construction process.Separate datasets are constructed for people and hooks,and a framework for two types of networks is designed.The algorithm design is optimized and applied using examples in four steps.The algorithm framework is packaged,optimized,and programmed for import into the monitoring platform in the crane operator's room and into the terminals of the safety management platform.Using the YOLOv5s model as the base framework,four attention mechanisms,namely SE(Squeeze-and-Excitation Module),CBAM(Convolutional Block Attention Module),CA(Coordinate Attention),and ECA(Efficient Channel Attention),are introduced.The attention mechanism with the greatest overall improvement is selected as the recognition model for this paper through a comparison of model performance data.The optimal CBAM model is selected by comparing and analyzing the results of five training models.The optimized detection model achieves an average recognition accuracy of 86.5% for helmets and 83.0% for reflective clothing.The state recognition accuracy of the hook is 92.0%.Furthermore,the YOLOv5 model and the added CBAM attention mechanism are packaged into an executable(exe)application.The trained personnel detection model and the hook detection model are packaged into an exe executable file,and it is demonstrated that the exe application presents good detection results.This approach proves to be an effective and fast way to detect the safety of personnel and the safety status of the hook.The exe package file does not depend on the hardware configuration of the computer and can be widely used for safety inspections of construction sites.When applied to the central control platform for constructi

关 键 词：安全工程 深度学习 注意力机制 exe文件打包 施工管理 

分 类 号：X947[环境科学与工程—安全科学]