Abstract—Automated surface defect detection systems have received much attention for quality control in industrial production. Deep learning techniques are proving their capability in these systems, due to the complexity of defects and inspection requirements. However, in fact, the availability of defective data is a major challenge. It is thus difficult to build an efficient model for a high-accuracy inspection system. In this paper, we present a method to deal with this lack of defective data by using self-contrastive learning to enhance image representations and the margin loss to improve the discriminativeness of defect features. Experiments were performed on the NEU dataset and MixedWWM38 dataset for several data size settings and for the few-shot learning task. The obtained results demonstrate the effectiveness of our proposed method. Particularly, the method achieves an accuracy of 98.83% and 92.27% on NEU dataset and MixedWM38 dataset, respectively, with only 20 training samples per class.
 
Keywords—image classification, contrastive learning, representative learning, surface defect recognition


Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.