SRGAN based super-resolution reconstruction of power inspection images

Abstract

Abstract Ensuring the operational safety of the electric power system critically depends on effective power inspections. However, traditional methods face challenges in detecting minor faults such as cracks and corrosion in electrical equipment. In this thesis, the Super Resolution Generative Adversarial Network (SRGAN) is introduced into the field of power inspection for the first time. Additionally, the dedicated dataset (BDZ dataset) was developed. This includes a large number of high-resolution images for power line inspection. The primary objective is to enhance the resolution of inspection images, thereby significantly improving the accuracy and reliability of defect detection in the power system. Numerous experiments have demonstrated that the SRGAN model outperforms traditional models in the super-resolution reconstruction of power inspection images, particularly in recovering image texture details. Using the BDZ dataset significantly enhances image resolution. When employing the same SRGAN model, PSNR increased by 2.47 dB and SSIM by 4.10% compared to the standard dataset. This research introduces new methodologies for advancing electric power inspection technologies, providing a more robust assurance for the safe and reliable operation of electric power systems.

Keywords

• Super-resolution reconstruction
• Power inspections
• Generative adversarial networks
• BDZ dataset