Research on Infrared Image Super-Resolution Based on Enhanced Generative Adversarial Network

Abstract

Infrared image super-resolution reconstruction is an effective way to improve the quality of infrared images. Aiming at the problem that the reconstructed image of the enhanced generative adversarial network Esrgan is prone to artifacts, This paper proposes a multi-branch enhanced generative adversarial network model, which adds an attention mechanism module and a residual module to the backbone of the original enhanced generative adversarial network model generator. The attention mechanism module strengthens the learning of useful channel information and suppresses useless channel information through the learning of global channel information. The residual module is a stack of the original Esrgan model block, which is the normalization of the channel dimension and retains the high-frequency features of the image. Experiments show that, compared with the original Esrgan model, the image texture details generated by the improved model are clearer and the resolution is higher, and the PSNR and SSIM of the image are significantly improved.