High-resolution restoration of solar images degraded by atmospheric turbulence effect using improved CycleGAN

Abstract

Due to the atmospheric turbulence effect, the solar images received by solar telescopes are generally severely degraded. In order to restore and improve the solar images, we proposed an improved Cycle Generative Adversarial Network (CycleGAN). By incorporating the identity loss and the perceptual loss into the loss function of CycleGAN, the ability of the network to capture image pixels and high-frequency information is significantly improved. On the basis, the influence of the turbulence intensity and the image noise on the restoration effect of solar images is analyzed. Results show that the improved CycleGAN has the advantages of smaller error and higher clarity of the restored images. In addition, the proposed method does not require paired images as training datasets, providing a new processing method for optical system telescopes and other fields of image processing.