一种改进YOLOv8的钢材表面缺陷检测算法  

A Steel Surface Defect Detection Algorithm Based on Improved YOLOv8

作　　者：黎源东 贺智明[1] LI Yuandong;HE Zhiming(School of Information Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,China)

机构地区：[1]江西理工大学信息工程学院,江西赣州341000

出　　处：《软件导刊》2024年第11期200-205,共6页Software Guide

摘　　要：由于制造工艺不完善,加上外部因素影响,钢表面常会存在一些缺陷,从而影响其寿命及可用性。表面缺陷检测是工业生产中的必要过程,而传统的表面缺陷检测算法存在精度低、速度慢的缺点。为此,在YOLOv8模型基础上加以改进,将原有的损失函数CIoU替换为SIoU函数,并在Backbone部分引入ShuffleAttention(SA)注意力机制,以提高对图像浅层和深层特征信息的提取能力,最后针对数据集特点在网络中增加一个小目标检测层,强化特征提取能力。实验表明,改进的YOLOv8-LSD算法,较原算法的mAP值提高了3.9%,降低了缺陷误检测和漏检率。Due to imperfect manufacturing processes and external factors,steel surfaces often have defects that can seriously affect their life and usability.Therefore,surface defect detection is a necessary process in industrial production.Traditional surface defect detection algorithms have the drawbacks of low accuracy and slow speed.Therefore,this article improves on the YOLOv8 model.Replace the original loss function CIoU with the SIoU function,and introduce the ShuffleAttention(SA)attention mechanism in the Backbone section to improve the extraction ability of shallow and deep feature information in the image.Finally,add a small object detection layer in the network based on the characteris⁃tics of the dataset to enhance the feature extraction ability.Experiments show that the improved YOLOv8-LSD algorithm,which improves the mAP value by 3.9%over the original algorithm,reduces the false detection and leakage rate of defects.

关 键 词：钢材表面缺陷 YOLOv8 SIoU ShuffleAttention(SA) 检测层 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]