测控技术2025，Vol.44Issue(9)：1-8,30,9.DOI:10.19708/j.ckjs.2025.08.244
基于无监督学习的表面缺陷检测算法研究综述
Review of Surface Defect Detection Algorithms Based on Unsupervised Learning
摘要
Abstract
With the development of intelligent manufacturing,the demand for industrial surface defect detection is becoming increasingly urgent.Traditional manual visual inspection and image processing methods based on manual features have shortcomings such as low efficiency,high false detection rate,and poor universality.Al-though deep learning technologies have significantly improved detection performance,supervised learning meth-ods still face two major challenges:insufficient annotated data due to the scarcity of defect samples and sensi-tivity to class distribution.Therefore,unsupervised defect detection technology has become a research hotspot.Two types of unsupervised defect detection algorithms based on representation and reconstruction are systemati-cally sorted out,and compared and analyzed from the dimensions of algorithm principles,technological evolu-tion,and application effects.Especially for the detection difficulties faced by high-end industrial fields such as aviation manufacturing,the application prospects of solutions such as lightweight modeling and multi-scale fea-ture fusion are explored,providing theoretical reference and technical guidance for the intelligent upgrading of industrial quality inspection.关键词
表面缺陷检测/无监督学习/无损检测/图像处理Key words
surface defect detection/unsupervised learning/nondestructive testing/image processing分类
信息技术与安全科学引用本文复制引用
赵艳涛,杨春宝,徐云山,修杰辰..基于无监督学习的表面缺陷检测算法研究综述[J].测控技术,2025,44(9):1-8,30,9.基金项目
深圳市科技计划重点项目(KJZD20230923113801003) （KJZD20230923113801003）
