AC-SNGAN: Multi-class data augmentation for damage detection of conveyor belt surface using improved ACGAN

Abstract

The current method for detecting surface damage on conveyor belts requires a lengthy training and deployment period, and its performance is limited by the size of the data set. Here, a novel visual detection method called Auxiliary Classifier Spectrally Normalized GAN (AC-SNGAN) is proposed, which incorporates a data augmentation module based on generative adversarial networks to augment the feature database of conveyor belt surface damage samples. The deployment time of the detection method is reduced by applying the Wasserstein distance and spectral normalized strategy. The experimental results indicate that the proposed method can efficiently and stably generate multiple classes of high-quality conveyor belt surface defect samples. The detection method's mAP reaches 99.31 % when trained with 200 samples per class of tears, scratches, and healthy features. The proposed detection method contributes to the highly reliable monitoring of conveyor belt safety operations in cases of limited sample resources.