Open AccessUnsupervised defect detection based on the pseudo-defect generation
Author(s) -
Jiekang Feng
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/2010/1/012163
Subject(s) - computer science , inpainting , artificial intelligence , annotation , product (mathematics) , image (mathematics) , quality (philosophy) , pattern recognition (psychology) , computer vision , machine learning , mathematics , philosophy , geometry , epistemology
With the progress of industrial production, more and more products need to be tested to ensure product quality. Detecting and locating surface defects have become a challenging and practical problem. In previous researches, the supervised training needs a lot of manual annotation data and defective product data. And the unsupervised method based on image inpainting can’t reconstruct complex images with high precision. In this paper, we proposed an unsupervised defect detection method based on the pseudo-defect generation to solve the problem of insufficient defective samples and the detecting accuracy problem. We conducted experiments on the AITEX dataset which get 93.3% DR, 3.2% FAR, and 35.5% MIOU. And it also shows outstanding effects in real industrial scenes.