Classification of Steel Surface Defect Using Convolutional Neural Network with Few Images

Abstract

Classification of steel surface defects in steel making industry is essential for the detection of defects through the classification of defects and for the analysis of causes that make defects. This makes it possible to reduce the defect rate of the product, and drastically reduces the mass defect in the steel making process. Recently, Deep Learning has been used for defect detection using Convolutional Neural Network(CNN). Compared to the existing rule-based method, the defect classification using CNN achieved high performance. However, learning CNN requires hundreds or thousands of images. In the case of a defective image, it is difficult to obtain images of thousands or hundreds of images. To overcome these shortcomings, few-shot learning which need few images to train network can be used. Siamese Neural Network using CNN is used for few-shot learning. In this paper, we use few-shot learning with Siamese Neural Network using CNN structure with contrastive loss to classify defects with a small number of steel surface defect images.