[1]李思思,葛华勇.改进YOLOv7的钢材表面缺陷检测模型[J].计算机技术与发展,2024,34(08):78-85.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0133]
　LI Si-si,GE Hua-yong.Detection Model of Steel Surface Defect Based on Improved YOLOv7[J].,2024,34(08):78-85.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0133]
点击复制

改进YOLOv7的钢材表面缺陷检测模型

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
34
期数:
2024年08期
页码:
78-85
栏目:
人工智能
出版日期:
2024-08-10

文章信息/Info

Title:
Detection Model of Steel Surface Defect Based on Improved YOLOv7
文章编号:
1673-629X(2024)08-0078-08
作者:
李思思; 葛华勇
东华大学 信息科学与技术学院,上海 201620
Author(s):
LI Si-si; GE Hua-yong
School of Information Science and Technology,Donghua University,Shanghai 201620,China
关键词:
钢材; 缺陷检测; YOLOv7; 解耦头; 特征融合
Keywords:
steel; defect detection; YOLOv7; decoupling head; feature fusion
分类号:
TP391.4
DOI:
10.20165/j.cnki.ISSN1673-629X.2024.0133
摘要:
在钢材生产过程中会出现各种的瑕疵缺陷,影响其使用寿命。 针对在钢材生产过程中检测表面缺陷效率低下的问题,提出了一种基于 YOLOv7 改进的钢材表面缺陷检测模型。 该模型改进耦合检测头为非对称多级通道压缩解耦头,解决不同检测任务之间的冲突,减少前向传播过程中目标置信度任务的特征丢失;设计轻量化 FCSP block 增强主干网络特征提取能力和颈部网络特征融合能力,提高模型对缺陷的定位能力的同时显著提升了检测速度;为进一步丰富小目标浅层特征增强其表达能力,加入可学习参数以促进动态特征融合,促进网络学习多样化特征。 在 NEU-DET 数据集上的实验结果表明,相较于原 YOLOv7 模型,改进后的模型 mAP 提高了 8. 8 百分点,FPS 提高了 11. 6,验证了该模型在检测精度和检测速度上都有所提升;在光照减弱后的 NEU-DET 数据集和 GC10-DET 数据集分别做通用性对比实验,结果表明该模型能有效地应用于工业中的钢材表面缺陷检测任务。
Abstract:
In the process of steel production,there will be various defects and flaws,which will affect its service life. To address the issue of low efficiency in detecting surface defects during the steel production process,a steel surface defect detection model based on improved YOLOv7 is proposed. The model introduces an improved coupled detection head called the asymmetric multi-level channel compression decoupling head to resolve conflicts between different detection tasks and reduce feature loss in the target confidence task during forward propagation. A lightweight FCSP block is designed to enhance the feature extraction capability of the backbone network and the feature fusion capability of the neck network,improving the model's defect localization ability while significantly increasing detection speed. To further enrich the expression capability of shallow features for small targets, learnable parameters are introduced to promote dynamic feature fusion,facilitating the network in learning diverse features. Experimental results on the NEU-DET dataset demonstrate that the improved model achieves an 8. 8 percentage point increase in mAP and an 11. 6 FPS improvement compared to the original YOLOv7 model,validating enhancements in both detection accuracy and speed. Generalization experiments on the NEU - DET dataset under reduced lighting conditions and the GC10 -DET dataset confirm the model's effectiveness in industrial steel surface defect detection tasks.

相似文献/References:

[1]张燕 刘春.基于圆心定位的瓶口三圆周快速缺陷检测算法[J].计算机技术与发展,2009,(06):243.
　ZHANG Yan,LIU Chun.A Rapid Defect Detecting Algorithm of Bottle＇s Rim Based on Central Location and Three- Circle Method[J].,2009,(08):243.
[2]舒坚 胡茂林.基于Markov随机场模型的纹理图像的缺陷检测[J].计算机技术与发展,2006,(05):65.
　SHU Jian,HU Mao-lin.Detection of Defect Based on Texture of Markov Random Field Model[J].,2006,(08):65.
[3]邓忠华 刘飞.印刷品缺陷检测的初步方案[J].计算机技术与发展,2006,(10):171.
　DENG Zhong-hua,LIU Fei.Preliminary Scheme for Defect Inspection of Printed Material[J].,2006,(08):171.
[4]李雪晨 汪仁煌 艾星芳.改进的弯曲度算法在阶梯修剪检测中的应用[J].计算机技术与发展,2012,(03):166.
　LI Xue-chen,WANG Ren-huang,AI Xing-fang.Use of Improved Tortuosity Algorithm in Step Form Disfigurement Test[J].,2012,(08):166.
[5]杨杰,卢盛林,赵晓芳.机器视觉在钢化玻璃缺陷检测中的应用研究[J].计算机技术与发展,2013,(03):211.
　YANG Jie,LU Sheng-lin,ZHAO Xiao-fang.Application and Research of Machine Vision in Tempered Glass Defect Inspection[J].,2013,(08):211.
[6]陈星[],徐迎晖[],肖青海[][]. QR码印刷品缺陷检测[J].计算机技术与发展,2015,25(10):191.
　CHEN Xing[],XU Ying-hui[],XIAO Qing-hai[][]. Defect Detection of Printed QR Code Image[J].,2015,25(08):191.
[7]杜晓辉,刘霖,张静,等.火花塞端面缺陷自动检测算法设计[J].计算机技术与发展,2019,29(02):172.[doi:10．3969/j．issn．1673－629X．2019．02．036]
　DU Xiaohui,LIU Lin,ZHANG Jing,et al.Optoelectronic Inspection of Defects for Surface of Spark Plug[J].,2019,29(08):172.[doi:10．3969/j．issn．1673－629X．2019．02．036]
[8]刘钱超,董超群,张垚.基于容器技术的软件测试优化研究[J].计算机技术与发展,2019,29(04):13.[doi:10. 3969 / j. issn. 1673-629X. 2019. 04. 003]
　LIU Qian-chao,DONG Chao-qun,ZHANG Yao.Research on Optimization of Software Testing Based on Container Technology[J].,2019,29(08):13.[doi:10. 3969 / j. issn. 1673-629X. 2019. 04. 003]
[9]张洪波,隋文涛,袁摇 林,等.基于深度自编码器网络的压盖缺陷检测[J].计算机技术与发展,2022,32(02):143.[doi:10. 3969 / j. issn. 1673-629X. 2022. 02. 023]
　ZHANG Hong-bo,SUI Wen-tao,YUAN Lin,et al.Capping Defect Detection Based on Deep Autoencoder Network[J].,2022,32(08):143.[doi:10. 3969 / j. issn. 1673-629X. 2022. 02. 023]
[10]徐奎奎,董兆伟,孙立辉,等.基于深度学习的 GH4169 合金组织表面缺陷检测[J].计算机技术与发展,2022,32(09):208.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 032]
　XU Kui-kui,DONG Zhao-wei,SUN Li-hui,et al.Surface Defect Detection of GH4169 Alloy Structure Based on Deep Learning[J].,2022,32(08):208.[doi:10. 3969 / j. issn. 1673-629X. 2022. 09. 032]

更新日期/Last Update: 2024-08-10