DP-GAN: A Transmission Line Bolt Defects Generation Network Based on Dual Discriminator Architecture and Pseudo-Enhancement Strategy

To solve the problem of scarcity of bolt defect samples in transmission lines, we propose a bolt defect image generation method based on dual discriminator architecture and pseudo-enhancement strategy (DP-GAN). First, we propose a residual discriminator network structure, coupled with a dual discriminator GAN architecture, to enhance the diversity of generation while preserving image feature information. Then, a generated image fidelity assessment method is designed to evaluate the fidelity of generated images by fitting the real dataset and screening out high-quality fake samples. Finally, a new pseudo-enhanced training strategy is proposed, which uses pseudo-samples to augment the few-shot dataset, which solves the problem of poor generation quality due to too few images of bolt defects. We construct a few-shot bolt defect dataset and conduct experiments on this dataset. Experimental results demonstrate that the bolt defect images generated by our proposed method have better quality and richer diversity than other image generation methods. Additionally, the proposed method significantly improves the performance of bolt defect classification. The classification accuracy shows a significant improvement over the CNN-only baseline.