| Title |
Augmentation of Defect Image Data using GAN |
| Authors |
조은희(Eunhee Cho) ; 진소미(Somi Jin) ; 전병환(Byeonghwan Jeon) ; 박인규(In Kyu Park) |
| DOI |
https://doi.org/10.5573/ieie.2022.59.10.98 |
| Keywords |
Deep learning; GAN augmentation; Transfer learning; Classification |
| Abstract |
Most manufacturers automate the defect inspection process to increase production efficiency by utilizing deep learning-based algorithms. However, an imbalanced situation often occurs unavoidably between normal and defect samples. Building a sufficient dataset is essential to training deep models, but can be time-consuming and costly. Critically, it decreases the performance and generalization of models. In this paper, we propose a GAN-based defect image data augmentation method to solve the imbalanced dataset problem in a real-world manufacturing. Through our proposed method, we provide one of the approaches to handle the imbalanced dataset by generating realistic and diverse defect data. Experimental results show that the proposed method is superior to other generative models, and effectiveness of our proposed method by improving the classification model through the augmented dataset. |