基于改进YOLOv8的轻量化焊缝表面缺陷检测方法  

Lightweight Weld Surface Defect Detection Method Based on Improved YOLOv8

作　　者：陈新 徐辉 张孜勉 熊铁军 陈曾雄 CHEN Xin;XU Hui;ZHANG Zimian;XIONG Tiejun;CHEN Zengxiong(Hunan Branch of China UNICOM,Changsha 410014,China)

机构地区：[1]中国联合网络通信有限公司湖南省分公司,湖南长沙410014

出　　处：《现代信息科技》2025年第3期44-49,共6页Modern Information Technology

摘　　要：针对钢材焊缝表面缺陷检测中因缺陷尺度小、形态多变及缺陷与背景对比度低等因素导致的误检率与漏检率偏高问题,提出一种基于改进YOLOv8模型的轻量化焊缝表面缺陷检测方法。首先,在模型主干网络引入空间金字塔分解(SPD)模块,以增强模型对小尺度缺陷的检测能力;其次,在特征融合网络嵌入SimAM注意力机制,强化模型对低对比度缺陷的特征表征能力;再次,采用Wise-IoU替代传统边界框回归损失函数,优化模型定位精度;最后,通过ADown模块改进下采样方法,有效保留焊缝缺陷的细节特征。实验结果表明:改进模型的检测精度、召回率与平均精度均值(mAP)分别提升了3.7%、1.6%和3.6%,其综合性能优于原始模型及其他主流目标检测模型,为工业场景下的焊缝缺陷检测系统部署提供了有效解决方案。Aiming at the problems of high false detection rate and missed detection rate in steel weld surface defect detection due to factors such as small defect scale,variable morphology and low contrast between defects and background,a lightweight weld surface defect detection method based on improved YOLOv8 model is proposed.Firstly,the Spatial Pyramid Decomposition(SPD)module is introduced into the model backbone network to enhance the model's ability to detect small-scale defects.Secondly,the SimAM is embedded in the feature fusion network to enhance the feature representation ability of the model for low contrast defects.Thirdly,Wise-IoU is used to replace the traditional bounding box regression loss function to optimize the localization accuracy of the model.Finally,the down-sampling method is improved by the ADown module to effectively retain the detailed features of the weld defects.The experimental results show that the detection accuracy,recall rate and mean Average Precision(mAP)of the improved model are increased by 3.7%,1.6%and 3.6%,respectively.Its comprehensive performance is better than the original model and other mainstream object detection models,which provides an effective solution for the deployment of weld defect detection systems in industrial scenarios.

关 键 词：焊缝缺陷检测 YOLOv8 SPD模块 SimAM注意力 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.4[自动化与计算机技术—控制科学与工程]