Research on super-resolution reconstruction of transmission lines based on SinGAN using a single image

Abstract

The purpose of this study is to address the problem of blurry transmission line images caused by foggy and cloudy weather, which severely hinders defect detection and intelligent recognition. To improve the quality and usability of inspection images, a SinGAN-based single-image super-resolution reconstruction method is proposed. The methodology exploits SinGAN’s advantage of requiring no paired datasets and adopts a self-supervised framework with an image pyramid structure. By learning texture features at multiple scales, the model progressively reconstructs high-resolution images and restores structural and textural details from a single blurry input. The findings demonstrate that the proposed method significantly enhances image clarity and improves structural fidelity. Quantitative evaluation is performed using PSNR and SSIM metrics for image quality and YOLOv7 mAP for detection performance. Results show clear improvements in detail restoration and detection accuracy compared with conventional super-resolution approaches, especially in scenarios with scarce samples. The practical implications of this study highlight that the SinGAN-based approach provides a scalable and efficient solution for real-world transmission line inspections. By eliminating the need for paired training data and enabling rapid deployment, this method enhances smart grid inspection capabilities under challenging conditions and offers strong potential for practical engineering applications.