基于改进 YOLOv8的织物缺陷检测算法  

Fabric defect detection algorithm based on improved YOLOv8

作　　者：王川[1,2] 李晓龙 王公轲[3] 段德全[1,2] 常升龙 WANG Chuan;LI Xiaolong;WANG Gongke;DUAN Dequan;CHANG Shenglong(College of Computer and Information Engineering,Henan Normal University,Xinxiang,Henan 453000,China;College of Softwear,Henan Normal University,Xinxiang,Henan 453000,China;College of Materials Science and Engineering,Henan Normal University,Xinxiang,Henan 453000,China)

机构地区：[1]河南师范大学计算机与信息工程学院,河南新乡453000 [2]河南师范大学软件学院,河南新乡453000 [3]河南师范大学材料科学与工程学院,河南新乡453000

出　　处：《毛纺科技》2025年第4期133-141,共9页Wool Textile Journal

基　　金：国家自然科学基金项目(62072159);河南省科技攻关项目(232102211061,222102210011,252102111168)。

摘　　要：针对织物瑕疵检测中瑕疵种类多样、部分瑕疵极小并存在着极端长宽比等问题,提出一种基于改进YOLOv8的织物瑕疵检测算法模型RDF-YOLOv8n。首先,在YOLOv8基线模型中引入感受野注意力卷积RFAConv,构建C2f_RFAConv模块,增强模型对部分缺陷特征的提取能力;其次,引入可变性大核注意力(Deformable-LKA)加入C2f中,组成C2f_DLKA模块,提高模型细小缺陷类型的检测能力;最后,采用Focaler_CIoU损失函数替代原有的损失函数,显著加快模型的收敛速度。结果证明:RDF-YOLOv8n模型的平均精度值(mAP值)为60.1%,相较于原模型平均精度均值提升了7.7%,推理速度为69帧/s,模型大小为9.3 MB,满足低算力设备部署条件,达到在生产中对织物瑕疵检测标准的要求。In order to solve the problems of fabric defect detection,such as various defects,very small defects and extreme aspect ratio,a fabric defect detection algorithm model,RDF-YOLOv8n,based on improved YOLOv8,was proposed.Firstly,the receptive field attentional Convolution RFAConv was introduced into the YOLOv8 baseline model,and the C2f_RFAConv module was constructed to enhance the model′s ability to extract partial defect features.Secondly,Deformable LKA was added to C2f to form C2f_DLKA module,which improved the detection ability of fine defect types.Finally,Focaler_CIoU loss function was used to replace the original loss function,which significantly speeds up the convergence of the model.The results show that the average accuracy value(mAP value)of the RDF-YOLOv8n model is 60.1%,which is 7.7%higher than the average accuracy of the original model.The inference speed is 69 frames/s,and the model size is 9.3 MB,which meets the deployment conditions of low-computing equipment and meets the requirements of fabric defect detection standards in production.

关 键 词：YOLOv8 织物瑕疵检测 RFAConv 注意力机制 Focaler_CIoU 

分 类 号：TS106.4[轻工技术与工程—纺织工程] TP391.41[轻工技术与工程—纺织科学与工程]