Reconstruction of Single-Frame Solar Speckle Image with Cycle Consistency Loss and Perceptual Loss

Abstract

High resolution reconstruction of solar speckle image is an important research subject in astronomy and solar physics. The common method uses the statistical information of a group of low-resolution images to reconstruct a high-resolution image through a specific algorithm. When this method is applied to the reconstruction of solar speckle image, the calculation process is complex and the reconstruction time is long because of the large amount of short-term obtained data. In this paper, based on the idea of image to image conversion, the single frame solar speckle image reconstruction method is proposed by using the generative adversarial networks (GAN), and the mapping from solar image to high-resolution image is established by using the cycle generative adversarial networks (CycleGAN). Due to the lack of high-frequency information in single frame speckle image, perceptual loss is introduced in CycleGAN to solve the problem of too smooth image reconstruction and lack of details. Combined with cycle consistent loss and perceptual loss, the quality of reconstructed image is improved. The experimental results show that the reconstruction accuracy of the proposed method is not less than that of the traditional algorithm, but in the time efficiency of the algorithm, it has more advantages than the traditional algorithm.