基于YOLOv5s-CBC的钢材表面缺陷检测  

Steel Surface Defect Detection Based on YOLOv5s-CBC

作　　者：刘翰林 刘凌云[1] 李超凡 仝保国 LIU Hanlin;LIU Lingyun;LI Chaofan;TONG Baoguo(School of Electrical and Information Engineering,Hubei University of Automotive Technology,Shiyan 442002,China)

机构地区：[1]湖北汽车工业学院电气与信息工程学院,十堰442002

出　　处：《组合机床与自动化加工技术》2024年第10期155-159,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：湖北省自然科学基金项目(2016CFB401)。

摘　　要：为了解决工业上钢材表面缺陷检测精度低的问题,提出了一种基于YOLOv5s目标检测网络的钢材表面缺陷检测方法YOLOv5s-CBC。首先,在主干网络中引入CBAM注意力机制,加强对钢材缺陷图像的特征提取,同时弱化钢材表面背景对检测结果的影响;其次,用轻量级上采样算子CARAFE替换颈部的最近邻插值上采样算子,增强感受野,提高目标检测能力并保持轻量化;最后,采用加权双向特征金字塔网络BIFPN修改原始模型中的PANet,增强模型的特征融合能力。实验结果表明,YOLOv5s-CBC在NEU-DET数据集上的平均精度均值(mAP)达到了80.1%,比原有YOLOv5s提高了3.6%。表明该方法具有良好的高效性和检测精度,为工件识别领域带来了一种有效的解决方案。In order to solve the problem of low detection accuracy of steel surface defects in industry,a steel surface defect detection method YOLOv5-CBC based on the improved YOLOv5s target detection network is proposed.Firstly,the CBAM attention mechanism is introduced in the backbone network to strengthen the feature extraction of steel defect images,and at the same time weaken the influence of the steel surface background on the detection results.Secondly,the nearest neighbor interpolation upsampling operator in the neck is replaced with the lightweight upsampling operator CAERAFE,which enhances the receptive field,improves the target detection ability and maintains light weight.Finally,the weighted bidirectional feature pyramid network BIFPN is used to modify the PANet in the original model to enhance the feature fusion ability of the model.The experimental results show that the average precision(mAP)of YOLOv5s-CBC on the NEU-DET dataset has reached 80.1%,which is 3.6%higher than the original YOLOv5s.This indicates that the method has good efficiency and detection accuracy,bringing an effective solution for the field of workpiece recognition.

关 键 词：缺陷检测 YOLOv5s 注意力机制 CARAFE 加权特征金字塔 

分 类 号：TH164[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]