SemiCon: A Semi-supervised Learning for Industrial Image Inspection

Abstract

Automatic visual inspection system through computer vision has been studied for decades. Traditional methods are often based on image processing techniques, which require individual design for each product and show unsatisfying performance. Recently, deep learning algorithms have significantly promoted the capability of computer vision in various tasks and provided new prospect in automatic inspection system. Numerous studies applied supervised deep learning to inspect industrial images and reported promising results. However, the methods employed in these studies are often in supervised fashion where heavy manual annotation are required. This may not be realistic in many manufacturing scenarios because products are constantly updating. Data collection, annotation and algorithm training can only be proceeded after the completion of the manufacturing process, causing a significant delay to train the models to establish the inspection system. This research aims to tackle the problem using semi-supervised manner such that these computer vision-based machine learning approaches can be rapidly deployed in real-life scenarios. Specifically, this paper proposed SemiCon: a semi-supervised contrastive learning method to identify defective products from industrial images. The proposed method was tested on two open-source industrial images inspection datasets. Results demonstrated that SemiCon achieved better results in several occasions than state-of-the-art approaches.