Research of Surface Defect Detection Method of Hot Rolled Strip Steel Based on Generative Adversarial Network

摘要

In order to solve the problems of various types of defects and few image samples in the process of surface defect recognition of hot rolled strip steel, and to improve the accuracy of defect recognition and classification, a new method based on Generative Adversarial Network (GAN) for surface defect recognition of strip steel is proposed in this paper. What’s more, this paper improves the structure of GAN aiming the unstable training and simple structure of the model. Conditional label vector is introduced into the input of generator and discriminator, and multiple classification branch is added to classify surface defects. The improved GAN is verified by the surface defect images of hot rolled strip steel collected in the industrial field. The simulation results show that this method can effectively identify and classify 6 kinds of surface defects such as patches, crazing, and pitted surface, with an average classification accuracy of 88%.