A texture guided transmission line image enhancement method

With the increasing demand of transmission line inspection, image detection technology plays a crucial role in foreign object detection on transmission lines. However, due to the influence of natural environment, the image quality is often interfered with by rain, fog, blur and other factors, which affects the accuracy of detection. In order to improve image quality and accuracy of foreign object detection, a texture guided transmission line image enhancement (TGTLIE) method is proposed. Firstly, the texture inference network (TINet) is used to extract texture information from input interference images. Then, the texture information is used as a priori knowledge to use the texture-based conditional generative adversarial network (TCGAN) to achieve adaptive image deraining, defogging and deblurring. In order to improve the detail quality and robustness of image generation, a neural gradient algorithm and dual path attention mechanism are proposed in the generative network of GAN. At the same time, in order to prevent local artifacts of spatial variation, an effective global-local discriminator structure is introduced into the discriminant network to perform global and local inspections on the generated images. In addition, the Charbonnier loss, SSIM loss and global-local generative adversarial loss are used in multiple stages to train the model to achieve the best performance. Both quantitative and qualitative results on different interference UAV datasets show that TGTLIE can effectively remove noise interference under different conditions and improve image quality, PSNR and SSIM reached 33.218dB,34.921 dB,30.725 dB and 0.956,0.962,0.926 respectively, and show excellent performance in foreign object detec-tion tasks. The method in this paper provides effective technical support for intelligent inspection and fault warning of transmission lines.