基于改进YOLOv5的铝型材料缺陷检测  

Aluminum Profile Material Defect Detection Based on Improved YOLOv5

作　　者：李欢 周敏[1,2] 陈晨[1,2] 张美洲 赵松怀[1,2] 金凯 LI Huan;ZHOU Min;CHEN Chen;ZHANG Meizhou;ZHAO Songhuai;JIN Kai(Key Loratory of Metallurgical Equipment and Its Control,Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081,China;Hubei Key Loratory of Mechanical Transmission and Control Engineering,Wuhan University of Science and Technology,Wuhan 430081,China)

机构地区：[1]武汉科技大学冶金装备及其控制教育部重点实验室,武汉430081 [2]武汉科技大学机械传动与控制工程湖北省重点实验室,武汉430081

出　　处：《组合机床与自动化加工技术》2025年第2期194-199,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金项目(51975431)。

摘　　要：针对铝型材生产过程中由于多种工艺因素影响产生表面缺陷、尺度差异大、小目标容易漏检等问题,提出了一种基于YOLOv5改进的铝表面缺陷检测模型。引入线性上下变换模块(LCT),提高模型对缺陷特征的提取能力;将Neck最近邻插值上采样方式换为轻量级上采样算子CARAFE,充分保留上采样特征图的小目标信息;用C3Faster模块替换Backbone的C3模块,轻量化模型的同时保留边缘信息。实验结果表明:改进模型的均值平均精度为87.8%,相比YOLOv5提高了3.8%。改进模型的参数量减少了8.4%,模型大小减小了8.5%。改进模型在轻量化的同时提升了目标检测的精度。Aiming at the problems of surface defects due to the influence of multiple process factors in the production process of aluminum profiles,large scale differences,and easy missed detection of small targets,this paper proposes an improved aluminum surface defect detection model based on YOLOv5.Linear context transform(LCT) is introduced to improve the extraction ability of the model for defect features;the Neck nearest neighbor interpolation up-sampling method is replaced by the lightweight up-sampling operator CARAFE,which fully retains the small target information of the up-sampled feature maps;and the C3Faster module is used to replace the Backbone′s C3 module,which retains the edge information while lightweighting the model.The experimental results show that the mean average accuracy of the improved model is 87.8%,which is 3.8% higher than that of YOLOv5.The amount of parameters of the improved model is reduced by 8.4%,and model size has been reduced by 8.5%.The improved model improves the accuracy of target detection while lightweighting.

关 键 词：机器视觉 铝型材料 YOLOv5 

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]