基于改进YOLOv8n的钢材表面缺陷检测算法  

Steel Surface Defect Detection AlgorithmBased on Improved YOLOv8n

作　　者：姚若禹 郑世玲 史怡璇 张思启 张锌飞 高飞龙 张霞[1] YAO Ruoyu;ZHENG Shiling;SHI Yixuan;ZHANG Siqi;ZHANG Xinfei;GAO Feilong;ZHANG Xia(School of Physics Science and Information Technology,Liaocheng University,Liaocheng 252059,China)

机构地区：[1]聊城大学物理科学与信息工程学院,山东聊城252059

出　　处：《聊城大学学报(自然科学版)》2025年第2期177-189,共13页Journal of Liaocheng University:Natural Science Edition

基　　金：国家自然科学基金项目(62205136);山东省自然科学基金项目(ZR2022MF284);聊城大学博士科研启动基金项目(318050062);聊城大学横向课题(K23LD94)资助。

摘　　要：针对当前钢材缺陷检测中目标特征复杂导致检测精度较低的问题,提出了一种基于改进YOLOv8n的钢材缺陷检测算法GOS-YOLO。首先,使用由轻量级卷积GSConv构建的Slim-neck范式作为特征融合网络,在减小模型参数量的同时提高精度。其次,将骨干网络的部分C2f模块替换为与全维度动态卷积(ODConv)相结合的C2f_ODConv模块,以实现模型的多维特征关注进而提高模型的精度。最后,将结合了多分支结构与压缩和激励操作的SENetV2注意力机制嵌入颈部网络,增强模型对复杂特征的提取能力。实验结果表明,在NEU-DET数据集上,GOS-YOLO的R、mAP50、mAP50~95较YOLOv8n分别提高了3.3%、1.7%、2.3%。在VOC2007数据集上mAP50~95提高了1%,FLOPs降低了16%。Aiming at the current problem of low detection accuracy due to the complexity of target features in steel defect detection,a steel defect detection algorithm,GOS-YOLO,based on improved YOLOv8n is proposed.Firstly,a Slim-neck paradigm constructed by lightweight-convolution GSConv is used as the feature fusion network,which improves the accuracy while reducing the number of model parameters.Secondly,some of the C2f modules of the backbone network are replaced with C2f_ODConv modules combined with full-dimensional dynamic convolution(ODConv),to achieve multi-dimensional feature attention of the model and thus improve the accuracy of the model.Finally,the SENetV2 attention mechanism,which combines a multi-branch structure with squeeze and excitation operation,is embedded into the neck network to enhance the model's ability to extract complex features.The experimental results show that on the NEU-DET dataset,the R,mAP50,and mAP50~95 of GOS-YOLO are improved by 3.3%,1.7%,and 2.3%,respectively,compared with YOLOv8n.On the VOC2007 dataset mAP50~95 improved by 1%and FLOPs decreased by 16%.

关 键 词：深度学习 计算机视觉 YOLOv8 缺陷检测 

分 类 号：TG142.1[一般工业技术—材料科学与工程] TP391.41[金属学及工艺—金属材料] TP183[金属学及工艺—金属学]