Research on the Expansion Algorithm of Leakage Magnetic Flux Defect Image Dataset for Long-Distance Oil and Gas Pipelines Based on DCGAN-Res-CBAM

The deep convolutional generative adversarial network (DCGAN) faces issues such as poor image quality, unstable model framework, and slow convergence during data augmentation. To address these challenges, an improved DCGAN model for pipeline leakage magnetic flux defect image generation is proposed. This model incorporates residual networks and attention mechanisms into the DCGAN framework to enhance the model's feature extraction capabilities. Experimental results demonstrate that the quality of pipeline leakage magnetic flux defect images generated by the given model is superior to those generated by DCGAN, WGAN, CGAN, and SAGAN. The average Frechet Inception Distance (FID) score achieved is 11 6.17, with the minimum FID score reaching 84.94. The proposed model can stably generate higher quality pipeline leakage magnetic flux defect images, providing an effective approach for expanding the tire defect sample dataset. This contributes to effectively solving the small sample problem faced by the development of deep learning in the field of defect detection.