Research On Astronomical Telescope Image Restoration Algorithms Based On Multi-frame Blind Deconvolution

Ground-based astronomical telescopes are crucial for observation,but atmospheric turbulence can degrade image quality,as reflected in the point spread function(PSF).Adaptive optics can repair wavefront distortion,but hardware limitations mean that resolution is below the diffraction limit and post-processing techniques are required.This thesis explores the multi-frame blind deconvolution algorithm for postprocessing of images from ground-based astronomical telescopes.The aim is to overcome the effects of atmospheric turbulence and noise,enhance the resolution and quality of images,and restore images close to the diffraction limit.The main contributions of this thesis are two-fold: noise suppression and model optimization.The specific contents are as follows:Firstly,in the iterative process of image and PSF in the multi-frame blind deconvolution algorithm,PSF may be affected by noise and cannot converge.This thesis introduces an Lq-norm sparse constraint robust to PSF noise and uses a semi-quadratic splitting method for optimization to effectively suppress noise during PSF iteration and estimate a compact PSF structure.Secondly,to improve optimization efficiency and effect of the image sub-problem,this thesis proposes a multi-frame correction blind deconvolution algorithm,abbreviated as MFCBD.MFCBD decomposes the image sub-problem into four steps: solving fidelity terms,denoiser,corrector,and solving prior terms.The denoiser adopts a plug-and-play algorithm and can choose a suitable denoising algorithm.The corrector compares the results of denoiser and solving prior terms using the objective function as a benchmark and selects results that ensure convergence for the next iteration.For actual images,compared to the best results among other advanced algorithms,the MFCBD algorithm improves the average PSNR increased by 0.63 dB,the average SSIM increased by 0.0429.On the second dataset,the contrast in the granule area is increased by 82.24%.Thirdly,multi-frame blind deconvolution algorithms often fail to estimate finestructured PSFs,resulting in loss of texture details in reconstructed images and edge blurring.This thesis proposes a hybrid iterative multi-frame blind deconvolution and phase difference algorithm,abbreviated as HIMBPD.HIMBPD combines multi-frame blind deconvolution methods and phase difference methods through PSF to calculate PSFs with high-frequency fine structures using wavefront phase coefficients to reconstruct images approaching diffraction limits.For simulated images,compared to the best results among other advanced algorithms,the HIMBPD algorithm improves the average PSNR by 3.02 dB and the average SSIM by 0.1728,and reduces the average RMSE of the wavefront phase by 0.0805ζ under four different noise and turbulence intensities.Finally,experiments on real AO-corrected images from astronomical telescopes and simulated images show that the algorithms proposed in this thesis have good recovery effects and robustness and can provide reference for research on such algorithms.