Industrial Product Surface Defects Detection Based on Weakly Supervised Learning

During the production of industrial products, appearance defects may lead to serious quality problems. Therefore, industrial product surface defect detection is critical for guaranteeing the quality of products. Supervised and unsupervised methods in deep learning theory have been used to address this problem, but they have limitations and may not fully leverage the datasets obtained from industrial scenarios. Therefore, this paper designs a deep learning architecture based on weakly supervised learning, which fully utilizes the collected industrial data set and is more in line with the actual industrial scenarios. Specifically, this paper presents an end-to-end network model consisting of two parts, a segmentation network and a decision network. The segmentation network is for defect segmentation, and the decision network is for defect classification. The model achieved a 100% detection rate on the KolektorSDD dataset with just 20 samples and showed promising results on the KolektorSDD2 dataset. The proposed architecture provides a more effective solution for defect detection in industrial settings.