基于深度学习的铝型材表面缺陷检测研究  

Research on Aluminum Profile Surface Defect Detection Based on Deep Learning

作　　者：刘国民 温秀兰[1] 高华卿 林玮轩 曾昱宁 LIU Guomin;WEN Xiulan;GAO Huaqing;LIN Weixuan;ZENG Yuning

机构地区：[1]南京工程学院自动化学院 [2]南京工程学院经济与管理学院

出　　处：《计量与测试技术》2024年第10期23-25,共3页Metrology & Measurement Technique

基　　金：江苏省产学研合作项目(项目名称:工业物联网设备健康度和工艺改善算法研究,项目编号:BY20230901);江苏省大学生实践创新训练计划项目(项目名称:基于深度学习的铝型材表面缺陷检测研究,项目编号:202311276083Y)。

摘　　要：针对铝型材表面缺陷检测的精度和效率较低等问题,本文提出一种基于Yolox深度学习的铝型材表面缺陷检测方法。该方法采用Yolox-s深度学习模型进行训练,并对已有的缺陷数据集通过旋转、调整亮度、对比度等方法进行扩充,将扩充后的70%、20%、10%的数据集分别用于训练集、验证集、测试集,进行实验验证。结果表明:与经典Yolov3、Yolov5及Yolov7算法相比,该方法的缺陷检测准确率高达90%,适用于铝型材行业。Aiming at the low accuracy and efficiency of aluminum profile surface defect detection,this paper propo-ses a method of aluminum profile surface defect detection based on Yolox deep learning.Yolox-s deep learning model was adopted for training,and the existing defect data set was expanded by rotating,adjusting brightness,contrast and other methods.70%,20%and 10%of the expanded data sets were used in the training set,verifica-tion set and test set respectively for experimental verification.The results show that compared with the classical Yolov3,Yolov5 and Yolov7 algorithms,the defect detection accuracy of this method is as high as 90%,which is suitable for the aluminum profile industry.

关 键 词：深度学习 Yolox-s 铝型材 表面缺陷检测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]