测控技术2025，Vol.44Issue(6)：25-31,7.DOI:10.19708/j.ckjs.2025.03.216
基于改进YOLOv5s的轻量化钢材表面缺陷检测模型
Lightweight Model for Steel Surface Defect Detection Based on Improved YOLOv5s
摘要
Abstract
A lightweight model based on improved YOLOv5s is proposed to address the problems of complex structures,large parameter quantity,poor real-time performance,and low detection accuracy in the current steel surface defect detection models.Firstly,the backbone network of YOLOv5 is replaced with MobileNetV3 to achieve model lightweight and improve detection speed.Secondly,the online convolutional re-parameteriza-tion(OREPA)technology is introducted to further reduce training costs.The K-means++algorithm is applied to improve the accuracy and convergence rate of prior box clustering.Finally,the CIoU(Complete Intersection over Union)loss function is replaced with EIoU(Extended Intersection over Union),which accelerates conver-gence and improves regression accuracy.Experimental results show that the proposed model increases mean av-erage precision by 2.8％,reduces the parameter count by 84.0％,decreases the model size by 81.4％,and improves detection speed by 60.8％.This model effectively balances lightweight design and detection accura-cy,is easy to deploy,which is suitable for real-time defect detection in steel production.关键词
钢材表面缺陷检测/YOLOv5s/轻量化模型/MobileNetV3Key words
steel surface defect detection/YOLOv5s/lightweight model/MobileNetV3分类
计算机与自动化引用本文复制引用
王同康,高天,乔文涛,香超,张隆..基于改进YOLOv5s的轻量化钢材表面缺陷检测模型[J].测控技术,2025,44(6):25-31,7.基金项目
河北省杰出青年科学基金项目(E2022210084) （E2022210084）
