摘要
加工而成的钢材表面可能会存在一定的缺陷,对钢材的外观和质量造成严重的影响,这些缺陷可以通过多种方法来完成分类和分割。传统检测方法精度不高且效率低下,采用基于深度学习的钢表面缺陷检测方法可有效提高检测性能。文章总结了近年来诸多学者提出的基于深度学习的缺陷分类和分割方法,介绍了这些算法的特点以及基于这些算法得到的改进算法,并对各类算法进行了比较,得出各种算法的优缺点。最后,总结了现阶段基于深度学习的缺陷检测技术存在的问题,并对未来的发展进行了展望。
The surface of processed steel may have certain defects that seriously affect the appearance and quality of the steel.These defects can be classified and segmented through various methods.Traditional detection methods have low accuracy and efficiency,and using deep learning-based steel surface defect detection methods can effectively improve detection performance.This paper summarizes the defect classification and segmentation methods based on deep learning proposed by many scholars in recent years,introduces the characteristics of these algorithms and the improved algorithms based on these algorithms,and compares various algorithms to identify their advantages and disadvantages.Finally,the existing problems of defect detection technology based on deep learning at the current stage are summarized,and the future development is prospected.
作者
费佳杰
李宏胜
任飞
吴敏宁
王光荣
FEI Jiajie;LI Hongsheng;REN Fei;WU Minning;WANG Guangrong(School of Automation,Nanjing Institute of Technology,Nanjing 211167,China;Digital Intelligent Mine Research Institute/Artificial Intelligence Research Institute,Sinoma(Nanjing)Mining Research Institute Co.,Ltd.,Nanjing 210000,China;School of Information Engineering,Yulin University,Yulin 719000,China)
出处
《现代信息科技》
2023年第19期107-112,共6页
Modern Information Technology
基金
南京工程学院大学生科技创新基金项目(TB202317004)
榆林市2021产学研项目(CXY-2021-102-02)。
关键词
缺陷检测
深度学习
目标检测
卷积神经网络
缺陷分类
defect detection
deep learning
object detection
convolutional neural network
defect classification
