Transmission Line Vibration Damper Detection Using Multi-Granularity Conditional Generative Adversarial Nets Based on UAV Inspection Images

Abstract

The vibration dampers can eliminate the galloping phenomenon of transmission lines caused by the wind. The detection of vibration dampers based on visual technology is an important issue. Current CNN-based methods struggle to meet the requirements of real-time detection. Therefore, the current vibration damper detection work has mainly been carried out manually. In view of the above situation, we propose a vibration damper detection-image generation model called DamperGAN based on multi-granularity Conditional Generative Adversarial Nets. DamperGAN first generates a low-resolution detection result image based on a coarse-grained module, then uses Monte Carlo search to mine the latent information in the low-resolution image, and finally injects this information into a fine-grained module through an attention mechanism to output high-resolution images and penalize poor intermediate information. At the same time, we propose a multi-level discriminator based on the multi-task learning mechanism to improve the discriminator’s discriminative ability and promote the generator to output better images. Finally, experiments on the self-built DamperGenSet dataset show that the images generated by our model are superior to the current mainstream baselines in both resolution and quality.

Keywords

• power transmission lines
• vibration dampers detection
• unmanned aerial vehicle (UAV)
• conditional generative adversarial nets (CGAN)
• Monte Carlo search (MCS)