Super-resolution Reconstruction for Infrared Images of Power Transmission Lines with N-Gram Transformer

In recent years, infrared imaging techniques have been applied to monitor transmission lines to ensure safety, but the images obtained are usually of low resolution. To solve this problem, we propose the NTSR model and introduce the Transformer framework, which can effectively extract global dependencies and obtain higher quality reconstruction results. The local self-attention mechanism is also used to calculate the self-attention of each window. However, it is prone to distortion when facing severely corrupted regions. Therefore we introduce N-Gram, using forward and backward N-Gram, which can help the model to better aggregate neighbourhood features and generate clearer images. Experimental results on real transmission line infrared images show that the proposed method achieves significant improvements compared to other methods, and generates images with better perceptibility and clarity.