A method for detecting surface defects of strip steel under the background interference of spangles based on STI-YOLO

• 1:华北理工大学人工智能学院

• 2:华北理工大学河北省工业智能感知重点实验室

• 
  

摘要(Abstract)：

针对有花镀层钢板表面缺陷检测过程中由于锌花底纹干扰导致缺陷识别率低的问题，提出一种以目标检测算法YOLOv5s为基础并通过引入通道注意力机制和金字塔卷积网络的Spangles Texture Interference-YOLO(STI-YOLO)算法模型.根据缺陷数据集重新聚类，优化先验框；在特征融合网络PANet之前引入通道注意力机制SENet,抑制锌花背景的干扰；在预测网络之前添加金字塔卷积网络，取得更加丰富的上下文特征.实验结果表明，STI-YOLO模型提升了带钢表面缺陷的检测精度，平均精度均值mAP达到了95.79%,较YOLOv3、YOLOv4和YOLOv5s算法分别提高了13.13个百分点、14.59个百分点和2.07个百分点.检测速度为54.14 frame/s,满足实时性要求，可见STI-YOLO模型具有较好的检测性能.
Aiming at the problem of low defect recognition rate due to the interference of zinc flower shading in the surface defect detection of patterned coated steel plate, the Spangles Texture Interference-YOLO(STI-YOLO)algorithm model based on the target detection algorithm YOLOv5s and improved by introducing channel attention mechanism and pyramid convolution network is proposed. According to the defect data set, this paper reclusters & optimizes the anchor box. The SENet is introduced before the PANet to suppress the interference of the zinc flower background. The pyramid convolution network is added before the prediction network to obtain richer context features. The experimental results show that the STI-YOLO model improves the detection accuracy of strip surface defects, and the average accuracy mAP reaches 95.79%, which is 13.13 percentage points, 14.59 percentage points and 2.07 percentage points higher than that of YOLOv3, YOLOv4 and YOLOv5s algorithms respectively. The detection speed is 54.14 frame/s, meeting the real-time requirements. The STI-YOLO model has better detection performance.

关键词(KeyWords)：缺陷检测;特征融合;注意力机制;多尺度特征;锌花背景干扰
defect detection;feature fusion;attention mechanism;multiscale feature;zinc flower background interference

基金项目(Foundation):科技部重点研发项目(2017YFE0135700)

作者(Authors):魏明军;陈钊;纪占林;周太宇;闫旭文;刘铭;
Wei Mingjun;Chen Zhao;Ji Zhanlin;Zhou Taiyu;Yan Xuwen;Liu Ming;College of Artificial Intelligence,North China University of Science and Technology;Hebei Provincial Key Laboratory of Industrial Intelligent Perception,North China University of Science and Technology;

DOI:10.16366/j.cnki.1000-2367.2023.04.011

参考文献(References)：

• [1] ZHAO Z Q,ZHENG P,XU S T,et al.Object detection with deep learning:a review[J].IEEE Transactions on Neural Networks and Learning Systems,2019,30(11):3212-3232.

• [2] 韩芳芳.表面缺陷视觉在线检测关键技术研究[D].天津：天津大学，2012.HAN F F.Key techniques for surface defects online detection based on machine vision[D].Tianjin:Tianjin University,2012.

• [3] 罗志勇.用线列CCD测量钢板宽度和检测孔洞缺陷的信号处理[J].量子电子学，1991,8(1):9.LUO Z Y.Signal processing of measuring steel plate width and detecting hole defects with linear CCD[J].Chinese Journal of Quantum Electronics,1991,8(1):9.

• [4] 罗志勇，刘栋玉，江涛，等.新型冷轧带钢表面缺陷在线检测系统[J].华中理工大学学报，1996,24(1):75-78.LUO Z Y,LIU D Y,JIANG T,et al.A new on-line surface defect inspecting system for the cold-rolled steel strip[J].Journal of Huazhong University of Science and Technology,1996,24(1):75-78.

• [5] 杨永敏，樊继壮，赵杰.冷轧带钢表面缺陷视觉检测系统[J].吉林大学学报(理学版),2011,49(5):911-917.YANG Y M,FAN J Z,ZHAO J.Cold rolling strip surface defect vision detection system[J].Journal of Jilin University(Science Edition),2011,49(5):911-917.

• [6] 布申申，田怀文.基于卷积神经网络的带钢表面缺陷检测算法[J].机械设计与制造，2022(7):29-33.BU S S,TIAN H W.Strip surface defect detection algorithm based on convolutional neural network[J].Machinery Design & Manufacture,2022(7):29-33.

• [7] REDMON J,DIVVALA S,GIRSHICK R,et al.You only look once:unified,real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).[s.l.:s.n.],2016:779-788.

• [8] REDMON J,FARHADI A.YOLO9000:better,faster,stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).[s.l.:s.n.],2017:6517-6525.

• [9] REDMON J,FARHADI A.YOLOv3:an incremental improvement[EB/OL].[2022-06-10].https://arxiv.org/abs/1804.02767.

• [10] LECUN Y,BOTTOU L,BENGIO Y,et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.

• [11] HU J,SHEN L,SUN G.Squeeze-and-excitation networks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.[s.l.:s.n.],2018:7132-7141.

• [12] DUTA I C,LIU L,ZHU F,et al.Pyramidal convolution:rethinking convolutional neural networks for visual recognition[EB/OL].[2022-06-11].https://arxiv.org/abs/2006.11538.

• [13] LIN T Y,MAIRE M,BELONGIE S,et al.Microsoft COCO:common objects in context[C]//Proceedings of 2014 European Conference on Computer Vision.Berlin:Springer,2014:740-755.

一种基于STI-YOLO的锌花背景干扰下带钢表面缺陷检测方法.pdf