Generative Adversarial Network-Based Data Augmentation for Tyre Surface Defect Detection

Rubber tyre surface inspection is highly relying on the experienced inspectors, which is inefficient and unstable. Therefore, machine vision-based inspection of tyre surface defects draws increasing attention from manufacturers to ensure product quality. In recent years, deep learning has become a dominant approach for surface defect detection. However, a huge amount of data is required to train a robust deep learning model. In order to overcome the problem of limited training datasets in real production environment, this paper investigates the effectiveness of Generative Adversarial Networks (GAN) on data augmentation of tyre defect detection. Firstly, artificial defect images are generated using the Pix2Pix method. The expanded datasets are used to train a Convolutional neural network (CNN) as well as a Support Vector Machine (SVM) model. Compared with the original data set, the expanded data set can evidently improve the accuracy of the classifiers. With the increase of the artificial image ratio, the improvement tends to slow down. The accuracy of the CNN classification in this study is higher than that of the SVM. Furthermore, the ratio of artificial data in the whole datasets influences the classification performance.