Research On Restoration Method Of Astronomical Image Based On U-Net

Astronomical research has increasingly higher requirements for astronomical images captured by telescopes.However,in the process of shooting astronomical images,due to the interference of the telescope’s own noise and external environment,the quality of astronomical images is mostly unsatisfactory.Therefore,it is necessary to improve the quality of astronomical images.Effective astronomical image processing methods can accelerate the progress of astronomical research.Although the traditional image processing method solves this problem to a certain extent,the method is simple,slow,inefficient,and the model is simple,and the restoration effect is not good enough in the performance of PSRN and SSIM.Traditional methods are not enough to deal with the development of current scientific research.Therefore,this paper adopts deep learning method to process astronomical images.This paper firstly introduces the causes and characteristics of astronomical image noise,and on this basis,expounds the influence of different degradation models on astronomical images.Secondly,the traditional image algorithms and the restoration algorithms based on deep learning are analyzed,and the advantages and disadvantages of each method are pointed out.Secondly,this paper builds a mathematical model of image generation based on the problem that the astronomical telescope takes a long time and is disturbed by noise during the shooting process.It is proposed to use U-Net to restore ground-exposure astronomical images,and add an attention mechanism to improve the network on the basis of U-Net.The low-exposure image is used as the input of the network,and various loss functions and activation functions are used to optimize the network.The experimental results show that the final network can restore astronomical images well,and the PSNR index is improved by nearly 5%,and the SSIM index is improved by about 0.5.Finally,in view of the problem of different kinds of defocusing degradation in astronomical images,considering the types and distribution of degradation modes,four kinds of point spread functions for image restoration and restoration are constructed by means of blind restoration based on deep learning.And proposed Res Net and U-Net to build a deep learning model Res UNet.Experiments show that the point spread function proposed in this paper can effectively restore most astronomical defocused images.Compared with traditional restoration algorithms and some classical deep learning restoration algorithms,the Res UNet-based astronomical image restoration network has better performance.