Research On Dynamic Variable Array Method And Image Restoration Algorithm Of Sparse Optical Synthetic Aperture

Optical synthetic aperture imaging technology is an important and effective way to achieve ultra-large aperture optical system,and it is also one of the main development directions for future ground-based telescopes and space-based optical remote sensors.It uses multiple small apertures to form an array to synthesize a large-aperture system to achieve highresolution imaging.Compared with traditional large-aperture imaging,the OSAI system has the advantages of low processing difficulty,low cost,light weight,small size,and flexible assembly design.However,because the OSAI system is filled with sub-apertures and has low spectral response,it is still difficult to meet the requirements when oriented to deep space exploration missions.In order to solve this problem,the existing methods usually adopt the method of optimizing the sub-aperture arrangement of individual array or sparse aperture variation to achieve high spectral coverage,with the former having the problems of weight and manufacturing difficulties and the latter having operational difficulties.Aiming at the problems of complex variable array method and difficult operation of existing sparse aperture variation methods,this thesis proposes a rotational and scaling variation method for sparse array with fixed rotation angle and baseline length variation.In addition,because the optical synthetic aperture system obtains a set of fuzzy degraded interference images,it needs to be restored to obtain high-resolution images.However,the existing algorithm models have strict assumptions and many restrictions,this thesis also conducts indepth research on image restoration algorithms,and proposes a restoration algorithm based on deep learning that can break through these restrictions.1.An asymmetric sparse aperture arrangement structure and variable array strategy are designed,and a rotational and scaling variation method of sparse array with fixed rotation angle and baseline length changes is proposed.This thesis proposes an improved sparse array variation method based on an in-depth analysis of the difficulty in the operation of the existing sparse aperture variation method caused by the variety of variation methods,which achieves high spectral coverage by a fixed variation method,that is,fixed rotation angle and baseline length variation.The simulation experiments and the experimental results of the built optical verification system both show the effectiveness of the proposed variable array strategy.2.The effects of phase error and noise on the restoration quality of interference images are analyzed and simulated.Phase errors are often unavoidable due to system design or hardware limitations.In this thesis,we first establish the piston error and tilt error models and analyze the effects on the system transfer function.Then,by adding noise to the variable array system and conducting imaging simulation experiments,the experimental results show that the sparse variable array system is more sensitive to noise and requires higher recovery algorithms.This thesis further analyzes the effects of phase error and noise on the imaging quality of the conventional OSAI system and the sparse variable array system,and gives the error tolerance requirements through data simulation.3.A restoration algorithm based on deep learning is proposed.The deep learning network model designed in this thesis is generally based on the U-Net structure,and the encoder and decoder modules are improved into transformer modules to solve the long-distance dependence problem of convolution,and at the same time,the loss function is improved by adding GANloss and perceptual loss on the basis of MSE loss,and the effectiveness of the method is proved by ablation experiments.Based on this improved network,this thesis also performs network restoration for the variable array system and the traditional Annular,Y and Golay types and the simulation results show that the PSNR and SSIM of this thesis’ s algorithm are significantly improved compared with U-Net,Wiener filtering,APEX and LR,which verifies the effectiveness of the sparse variable array system designed in this thesis.In addition,by adding different phase errors and noises to the system,the network models proposed in this thesis have achieved good results,which can effectively reduce the sensitivity of the variable array system to errors and noise.In order to further verify the effectiveness of the proposed sparse array system and algorithm,a synthetic aperture imaging demonstration verification system is also built.The system selects two apertures as the initial array,and uses deep learning network for image restoration after rotating and scaling the array.The experimental results show that the proposed method can eliminate blur to a greater extent and restore higher resolution images than the existing common methods.