改进的前端轻量级网络工业手套缺陷检测研究  

Research on Defect Detection of Industrial Gloves based on Front-end Lightweight Network

作　　者：王犇 江灏[1,2] 陈静[1,2] WANG Ben;JIANG Hao;CHEN Jing(College of Electrical Engineering and Automation,Fuzhou University,Fuzhou,China,350108;Research Institute of Power System and Power Equipment,Fuzhou University,Fuzhou,China 350108)

机构地区：[1]福州大学电气工程与自动化学院,福州350108 [2]福州大学电力系统与装置产业研究院,福州350108

出　　处：《福建电脑》2023年第5期16-20,共5页Journal of Fujian Computer

基　　金：融合视觉感知与交互的架空输电线路无人机边云协同巡检关键技术研究(No.2022H6020)资助。

摘　　要：纺织业是我国的重要产业。近年来,人们对纺织品的需求与日俱增。手套作为纺织品的一种,在实际生产中难免出现一些缺陷样品。为了实现更有效率的生产和管理,本文提出了一种改进的轻量型YOLOv5算法来实现纺织手套的缺陷检测,以YOLOv5s网络作为教师模型,先对其进行剪枝,再将剪枝后的模型作为学生模型,在不损失精度的前提下,训练出一个符合前端检测需求的轻量化模型。实验结果表明,压缩后的模型精度能达到0.93,参数量仅为原教师模型的32%,计算量仅为原模型的36%,更加有利于工业上对纺织手套的精益生产和管理,符合嵌入前端的需求。Textile industry is an important industry in our country.In recent years,the demand for textile is increasing.Gloves are a kind of textile,some defect samples are inevitable in the actual production.In order to achieve more efficient production and management,an improved lightweight YOLOv5 algorithm is proposed in this paper to realize the defect detection of textile gloves.The YOLOv5s network is used as the teacher model,pruning is carried out first,and then the pruning model is used as the student model.On the premise of not losing accuracy,a lightweight model is trained to meet the requirements of front-end detection.The experimental results show that the precision of the compressed model can reach 0.93,the number of parameters is only 32%of the original teacher model,and the calculation amount is only 36%of the original model,which is more conducive to the lean production and management of textile gloves in industry,and meets the needs of the embedded front end.

关 键 词：深度学习 轻量化 手套缺陷检测 精益生产与管理 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]