• 首页
  • 期刊简介
  • 编委会
  • 投稿指南
  • 收录情况
  • 杂志订阅
  • 联系我们
引用本文:谷长江,高法钦.改进YOLOv5s的钢材表面缺陷检测[J].软件工程,2023,26(8):31-34.【点击复制】
【打印本页】   【下载PDF全文】   【查看/发表评论】  【下载PDF阅读器】  
←前一篇|后一篇→ 过刊浏览
分享到: 微信 更多
改进YOLOv5s的钢材表面缺陷检测
谷长江, 高法钦
(浙江理工大学计算机科学与技术学院, 浙江 杭州 310018)
siesuta@outlook.com; gaofaqin@zstu.edu.cn
摘 要: 缺陷检测是生产中重要的环节,基于钢板表面缺陷特征不明显和难以提取导致的检测精度不足问题,文章在YOLOv5s检测网络的基础上进行改进,首先基于DO-Conv过参数化模块改进网络特征提取模块,然后使用ULSAM注意力机制改进网络的颈部(Neck),提出改进的YOLOv5s缺陷检测网络。基于NEU-DET数据集的实验结果表明,改进的YOLOv5s缺陷检测网络检测平均准确率达76.6%,较YOLOv5s和YOLOv4分别提升了7.8%和6.3%,有效提高了钢材表面缺陷检测精度。
关键词: YOLOv5s;缺陷检测;注意力机制;过参数化
中图分类号: TP399    文献标识码: A
Steel Surface Defect Detection Based on Improved YOLOv5s
GU Changjiang, GAO Faqin
(Department of Computer Science and Technology, Zhejiang Sci-tech University, Hangzhou 310018, China)
siesuta@outlook.com; gaofaqin@zstu.edu.cn
Abstract: As defect detection is a crucial part of production, this paper proposes an improved YOLOv5s detection network to solve the problem of insufficient detection accuracy caused by unclear and difficult extraction of surface defect features on steel plates. Firstly, the network feature extraction module is improved based on the DO-Conv over-parameterized module, and then the neck of the network is improved by using the ULSAM attention mechanism. Finally, an improved YOLOv5s defect detection network is proposed. The experimental results based on the NEU-DET dataset show that the improved YOLOv5s defect detection network has mean average precision of 76.6% , which is 7.8% and 6.3% higher than YOLOv5s and YOLOv4, respectively. The improved network effectively improves the accuracy of steel surface defect detection.
Keywords: YOLOv5s; defect detection; attention mechanisms; over-parameterization


版权所有:软件工程杂志社
地址:辽宁省沈阳市浑南区新秀街2号 邮政编码:110179
电话:0411-84767887 传真:0411-84835089 Email:semagazine@neusoft.edu.cn
备案号:辽ICP备17007376号-1
技术支持:北京勤云科技发展有限公司

用微信扫一扫

用微信扫一扫