基于改进YOLOv8的热轧带钢表面缺陷检测方法  

Surface Defect Detection Method for Hot-rolled Strip Steel Based on Improved YOLOv8

作　　者：肖科[1] 杨昕宇 韩彦峰[1] 宋斌 XIAO Ke;YANG Xinyu;HAN Yanfeng;SONG Bin(College of Mechanical and Vehicle Engineering,Chongqing University,Chongqing 400030,China;ROKAE(Shandong)Robot Group Co.,Ltd.,Jining 275312,China)

机构地区：[1]重庆大学机械与运载工程学院,重庆400030 [2]珞石(山东)机器人集团有限公司,山东济宁275312

出　　处：《湖南大学学报（自然科学版）》2024年第12期67-77,共11页Journal of Hunan University:Natural Sciences

基　　金：国家重点研发计划资助项目(2022YFB4702201);国家自然科学基金资助项目(52375039)。

摘　　要：针对目前热轧带钢表面缺陷检测精度低和效率低的问题,提出了一种基于改进YOLOv8s的目标检测算法.首先,提出了一种基于特征图二次拼接并融入GAM的SPPD模块,提升了模型多尺度信息融合能力.其次,提出了一种融合可变形卷积的特征提取模块DCNblock,以增大模型的感受野,提取完整的缺陷信息.最后,将特征融合网络中的C2f模块替换为BoT(bottleneck transformer)结构,将Transformer中的多头自注意力机制与卷积融合,提升模型的全局位置信息感知能力.实验结果表明,本文提出的算法在NEU-DET数据集上的平均精度均值(mAP)达到了80.5%,较原有的YOLOv8算法提升了5个百分点,同时检测速度达到了83帧/s,满足实时检测的需求.A object detection algorithm based on improved YOLOv8s is proposed to address the issues of low accuracy and low efficiency in surface defect detection of hot-rolled strip steel.Firstly,an SPPD module based on feature map secondary stitching and incorporating GAM is proposed,which enhances the model’s multi-scale information fusion ability.Secondly,a feature extraction module DCN-block that integrates deformable convolution is proposed to increase the receptive field of the model and extract complete defect information.Finally,the C2f module in the feature fusion network is replaced with a BoT(bottleneck transformer)structure,and the multi-head self-attention mechanism in the Transformer is fused with convolution to enhance the model’s global position information perception ability.The experimental results show that the proposed algorithm achieves mean average precision(mAP)of 80.5%on the NEU-DET dataset,which is five percentage points higher than the original YOLOv8 algorithm.At the same time,the detection speed reaches 83 frames per second,meeting the requirements of real-time detection.

关 键 词：热轧带钢 表面缺陷 目标检测 深度学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]