GAN-based Data Augmentation for Metal Surface Defect Detection Using Convolutional Neural Networks

Abstract

Artificial Intelligence-based Automated Optical Inspection (AI-AOI) using Convolutional Neural Networks (CNNs) is commonly used for defect detection, including metal defect detection, in modern manufacturing. However, in most AOI applications, the occurrence of defects is much less than the normal ones. CNN-based defection models perform poorly due to the imbalanced and less divergent training data. This study presents the performance of CNN-based AOI for metal defect detection with the techniques of generative AI for data augmentation. The Wasserstein Generative Adversarial Network (WGAN) is employed for generating negative training data and increasing the divergence when training AOI models. The similarity of data generated by WGAN to the original ones is evaluated by the Structural Similarity Index Measure (SSIM). The performance of ten CNN models trained with data before and after being augmented by WGAN are compared. Three metal defect datasets are used for evaluating the performance of CNN-based AOI with WGAN. The experimental results show that the performance of defect classification can be improved by 1%-12% with data augmented by WGAN.