改进YOLOv8的带钢表面缺陷检测技术  

Improved YOLOv8 Strip Surface Defect Detection Technology

作　　者：杜娟[1] 南晓林 晋美娟 刘宇航 宋文辉 DU Juan;NAN Xiaolin;JIN Meijuan;LIU Yuhang;SONG Wenhui(School of Mechanical Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China;Shanxi Pingyang Industry Machinery Co.,Ltd.,Linfen 043000,China)

机构地区：[1]太原科技大学机械工程学院,太原030024 [2]山西平阳重工机械有限责任公司,临汾043000

出　　处：《组合机床与自动化加工技术》2025年第4期167-172,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：山西省重点研发计划项目(202102150401009)。

摘　　要：针对工业生产环境中对热轧带钢的表面缺陷检测存在精度低、误检和漏检的问题,提出了一种基于YOLOv8n的改进算法。首先在特征提取部分将GhostConv代替部分普通卷积,降低模型的参数量,同时嵌入CPCA模块,提高模型对重要信息的提取能力;在特征融合部分将SPDConv代替普通卷积,最大限度保留特征信息;增加3个辅助检测头,提高模型的检测能力;在预测部分将NWD损失函数代替CIoU损失函数,提高模型对小目标的检测性能。实验结果表明改进算法相较于原算法,mAP提高了3.3%。在保证检测精度和速度的条件下,该技术能更好地应用于带钢表面的缺陷检测。An improved algorithm based on YOLOv8n was proposed to solve the problems of low precision,false detection and missing detection of surface defects of hot rolled strip steel in industrial production environment.Firstly,GhostConv replaces some common convolution in feature extraction to reduce the number of parameters in the model.At the same time,CPCA module is embedded to improve the ability of the model to extract important information.In the feature fusion part,SPDConv replaces common convolution to preserve the feature information to the maximum extent.Three auxiliary detection heads are added to improve the detection ability of the model.In the prediction part,the NWD loss function is replaced by the CIoU loss function to improve the detection performance of the model on small targets.The experimental results show that the mAP of the improved algorithm is improved by 3.3%compared with the original algorithm.Under the condition of ensuring the detection accuracy and speed,the technology can be better applied to the surface defect detection of strip steel.

关 键 词：YOLOv8 缺陷检测 GhostNet 注意力机制 NWD 

分 类 号：TH165[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]