Neural Network Design for the Restoration of Blurry Images Taken by the Line Inspection Robot

Transmission line inspection robots have played an important role in transmission line inspection. However, the inspection robot occasionally oscillates when capturing images, resulting in blurry inspection images. This article proposes a neural network model based on the improved SRNdeblur algorithm. Firstly, we use residual dense modules (RDB) instead of residual modules to reduce lost information in the network. Secondly, the multi-loss fusion function is used to focus on the pixel differences, overall structure differences and frequency space differences between image pairs. Finally, a fuzzy image dataset PLID for power inspection robots was created. And the test results show that our proposed algorithm performs better on PSNR and SSIM compared to SRNdeblur.