Solar Image Restoration Based On Deep Learning

The resolution of astronomical images obtained by ground-based telescopes are seriously affected by the atmospheric turbulence,tracking error and aberrations of the telescope.Although adaptive optics system can correct the atmospheric turbulent,but due to the residual error,measurement error and finite correction speed,there are still non-negligible residual errors,which would limit observation image quality.The image reconstruction algorithm is important to increase image quality.Texture is one of the most obvious characteristics in solar images and it is normally described by texture features.According to our experience,textures from solar images of the same wavelength are similar.We assume these texture features are multi-fractals.Based on this assumption,we propose a pure data-based image restoration method: with several high-resolution solar images as references,we use the Cycle-Consistent Adversarial Network to restore blurred images which have the same physical process.We test our method with simulated and real observation images and find that our method can improve the spatial resolution of solar images,without loss of any frames.Because our method does not need paired images as training set or additional instruments,it can be used as a post-processing method for solar images obtained by either seeing-limited telescopes or telescopes with wide field adaptive optic systems.In order to achieve better recovery results,we use neural network to model the multi-fractal properties of feature textures,and design the image perception evaluation algorithm to select reference images.We find that our images selected from the high-resolution database by the perception evaluation algorithm can further increase the efficiency of our algorithm.At the same time,the perception algorithm can also be used to evaluate the image quality.taking high resolution images in the same band as references,the perception evaluation can be used to evaluate blur effects.