基于YOLOv4的铜带表面缺陷识别研究  

Research on surface defect recognition of copper strip based on YOLOv4

作　　者：王紫玉 张果[1] 杨奇 尹丽琼 WANG Ziyu;ZHANG Guo;YANG Qi;YIN Liqiong(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming,Yunnan 650500,China;WISCO Group Kunming Iron&Steel Co,Ltd,Anning Company,Kunming,Yunnan 650302,China)

机构地区：[1]昆明理工大学自动化学院,云南昆明650500 [2]武钢集团昆明钢铁股份有限公司安宁公司,云南昆明650302

出　　处：《光电子．激光》2022年第2期163-170,共8页Journal of Optoelectronics·Laser

基　　金：国家重点研发计划(2017YFB0306405);国家自然科学基金(61364008);云南省重点研发计划(2018BA070)资助项目。

摘　　要：本文提出一种基于YOLOv4铜板带材表面缺陷检测模型,针对铜金属板带材生产过程中产生的表面缺陷形式多样、位置随机而导致难以快速定位和识别的问题,采用大数据驱动的深度学习策略,以铜带表面缺陷图像为训练样本,对YOLOv4目标检测模型进行训练,实验结果表明,改进的模型识别铜带表面缺陷的全类别平均精度均值(mean average precision,mAP)为93.37%,高于原始YOLOv4模型的全类别平均精度91.46%,检测速度达到49帧/秒,与双阶段的检测模型更快地R-CNN(faster region-based convolutional neural network,Faster R-CNN)相比,在保证检测精度的同时提升检测速度,能够满足在线检测需要,适合完成铜带工业生产过程中缺陷检测任务。This paper proposes a surface defect detection model on a copper plate and strip based on YOLOv4.Aiming at the problem of surface defects in the production process of copper metal plate and strip that are difficult to locate and identify due to their various forms and random positions,a big datadriven deep learning strategy is adopted.Using the copper strip surface defect image as the training sample,the YOLOv4target detection model is trained.The experimental results show that the improved model recognizes the copper strip surface defect with a full-category mean average precision(mAP)of 93.37%,which is higher than the original YOLOv4.The model has an average accuracy of 91.46%for all categories and a detection speed of 49frames per second.Compared with the two-stage detection model faster region-based convolutional neural network(Faster R-CNN),it can improve the detection speed while ensuring the detection accuracy,which can meet the needs of online detection.Defect detection task in industrial production process is suitable for completing copper strips.

关 键 词：深度学习 缺陷识别 YOLOv4 模式识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]