基于改进YOLOv5算法的纸袋缺陷检测  

Research on Paper Bag Defects Detection Based on Improved YOLOv5 Algorithm

作　　者：杨萌 张爱军[1] 潘文松 YANG Meng;ZHANG Aijun;PAN Wensong(School of Mechanical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China;Hangzhou KDx Biotechnology Co.,Ltd.,Hangzhou 311199,China)

机构地区：[1]南京理工大学机械工程学院,南京210094 [2]杭州康代思锐生物科技有限公司,杭州311199

出　　处：《组合机床与自动化加工技术》2024年第3期105-108,113,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：江苏省研究生科研与实践创新计划项目(SJCX22_0103)。

摘　　要：为了提高纸袋生产企业在制造过程中对纸袋手把或底部缺陷的检测精度,提出了一种基于改进YOLOv5算法的纸袋缺陷检测方法。改进算法为了提高网络定位能力,增强网络的特征学习表达能力,引入了坐标注意力机制,接着引入EIoU损失函数对原始损失函数进行改进,以此来改善原始网络损失函数纵横比的合理性,提升回归精度,最后引入一种具有类似跨阶段局部结构的简化空间金字塔池化结构,减少冗余信息处理,提升网络检测性能。实验结果表明,改进算法的平均精度平均值mAP@.5为87.3%,mAP@.5∶.95为56.8%,与YOLOv5算法相比mAP@.5提升了1.6%,mAP@.5∶.95提升了0.9%,在纸袋缺陷检测上有更优越的表现。In order to improve the detection precision of paper bag manufacturers on the handle or bottom defects,a paper bag defects detection method based on the improved YOLOv5 algorithm was proposed.In order to improve the network positioning ability and enhance the feature learning and expression ability of the network,the Coordinate Attention was introduced,and then the EIoU loss function was introduced to improve the loss function,so as to improve the rationality of the aspect ratio of the original network loss function to improve the regression precision,and a simplified spatial pyramid pooling structure with CSP-like structure was introduced to reduce redundant information processing and improve the detection performance.Experimental results show that mAP@.5 and mAP@.5∶.95 of the improved algorithm are 87.3%and 56.8%,respectively.Compared with the YOLOv5 algorithm,mAP@.5 and mAP@.5∶.95 of the improved algorithm are increased by 1.6%and 0.9%respectively,showing better performance in the detection of paper bag defects.

关 键 词：纸袋缺陷 注意力 损失函数 空间金字塔池化 改进算法 

分 类 号：TH16[机械工程—机械制造及自动化] TG66[金属学及工艺—金属切削加工及机床]