摘要
钢材表面缺陷检测是工业生产中至关重要的一项检测工作,针对钢材表面缺陷检测中漏检以及对于细小缺陷检测精度不佳等问题,提出了一种改进Faster R-CNN算法。在FPN(Feature pyramid networks)与RPN(Region proposal network)之间引入特征融合模块与轻量化通道注意力模块,增加模型对精细特征的捕捉能力。改进模型在NEU-DET数据集上的实验结果显示,最终mAP(Mean average precision,记为m_(AP))值为80.2%,比原始模型提高了12.6%,FPS提高了40.9%。该算法能够有效提升钢材表面缺陷的检测精度,为钢材表面缺陷自动检测提供参考。
Steel surface defect detection is a crucial inspection work in industrial production,and an improved Faster R-CNN algorithm is proposed to address the leakage detection and poor accuracy of steel surface defect.A feature fusion module and a lightweight channel attention module are introduced between FPN(Feature pyramid networks)and RPN(Region proposal network)to increase the model′s ability to capture fine features.The experimental results of the improved model on the NEU-DET dataset show that the m_(AP)(Mean average precision)value is 80.2%,which is 12.6%higher than the original model,and the detection speed is improved by 40.9%.The algorithm can effectively improve the detection accuracy of steel surface defects and provide a reference for the automatic detection of steel surface defects.
作者
冷岳峰
刘正
徐宝祎
李志轩
LENG Yuefeng;LIU Zheng;XU Baoyi;LI Zhixuan(School of Mechanical Engineering,Liaoning Technical University,Fuxin 123000,Liaoning,China)
出处
《机械科学与技术》
北大核心
2025年第1期75-83,共9页
Mechanical Science and Technology for Aerospace Engineering
基金
辽宁省教育厅理工类项目(22-1142)。
