Generative adversarial network-assisted image classification for imbalanced tire X-ray defect detection

A high-performance tire X-ray defect image classification method plays a key role in enhancing the automation level of tire defect detection. In industrial practice, however, a typical challenge is that the collected datasets of diverse tire defects are often imbalanced. To address this issue, a Wasserstein generative adversarial network (WGAN)-assisted image classification method is proposed for imbalanced tire X-ray defect detection. To expand the minority classes in original datasets, a WGAN model is established to generate high-quality X-ray defect images. Considering the feature similarity of different defect grades in the same type, the WGAN is trained based on a pre-trained model to extract deep features. An improved deep convolutional neural network model is restructured for performance improvement. Finally, the augmented balanced datasets are used to train the improved network for image classification of tire X-ray defects. The experiments validate that the proposed method is effective for type and grade classification of imbalanced tire X-ray defect detection, and shows better classification performance than existing popular models.