Research on Partial Discharge Image Recognition of Onboard Cable Terminal Based on Generative Adversarial Network

This paper addresses the challenges of difficult data acquisition and limited samples of partial discharge (PD) data from onboard cable terminals, which affect the accuracy of insulation defect classification. We propose an enhancement method for PD data from onboard cable terminals based on Generative Adversarial Network (GAN). Using Boundary Equilibrium GAN (BEGAN), Deep Convolutional GAN (DCGAN), and Wasserstein GAN with Gradient Penalty (WGAN-GP) to learn from PD image samples of onboard cable terminals, we evaluate the image generation quality of the three networks. The results show that all three networks have good generation effects on the PD images. We utilize BEGAN, DCGAN, and WGAN-GP to enhance the few-shot PD image dataset of onboard cable terminals and conduct classification training on the enhanced dataset. The results indicate that the enhanced datasets from all three generative networks can improve the accuracy of image classification, with DCGAN demonstrating the best enhancement effect.