Research On Cophasing Method Of Space Segmented Telescope Based On Focal Plane Image Information

The telescope's light-collecting ability and resolving power are directly related to the size of its main mirror aperture,so both ground-based telescopes and space-based telescopes have continued to develop in the direction of large apertures.However,the increasing size of the primary mirror has brought unprecedented challenges to the design,processing,manufacturing,and testing of telescopes.The emergence of segmented telescopes has greatly reduced the quality of primary mirrors,processing costs,manufacturing cycles,and transportation launch costs and difficulties.At present,the LAMOST telescope in China,the Keck telescope in the United States,and the JWST telescope and Thirty Meter Telescope under construction all use segmented primary mirror mode.In order to make the overall imaging quality of the segmented telescope close to the level of the diffraction limit,generally it is required to ensure high coplanar accuracy between the sub-mirrors,to achieve optical confocal and cophase,and the RMS of cophase error value between the sub-mirrors is required to be less than 1/40 wavelength.Such high precision cannot be achieved by relying solely on traditional mechanical adjustment.Therefore,it is necessary to use active optical technology to adjust the position of each sub-mirror actively and accurately in real time to maintain a good shape of the main mirror surface of the telescope.The active optical system completes the sub-mirror confocal and cophase operations.At this stage,the sub-mirror point confocal implementation is relatively simple,and the sub-mirror cophase has always been a problem that puzzles researchers.According to the development requirements of segmented space large-aperture telescopes,this paper focuses on two phases: coarse cophase and fine cophase.The main contents are as follows:Firstly,an optical model of a space segmented large-aperture telescope is established.To solve the cophase difficulty caused by the large piston error in the coarse cophase of the segmented main mirror of the space large-aperture telescope,this paper will introduce the convolutional neural network.The convolutional neural network is traind by the feature dataset independent of imaging target which is composed of multi-band focal plane image and defocus image,thereby determining the piston error range of each submirror,and guiding the segmented mirror into a fine cophase range that can be solved by the phase diversity algorithm.The method has the advantages of fast calculation speed,large measurement range,strong robustness and low cost.Since the training dataset is only related to the cophase error of the sub-mirrors and has nothing to do with the imaging content,it is rid of the dependence on the imaging target.In addition,statistical methods were used in the paper to screen out sensitive data points for the pistion error of each submirror in the training sample,thereby reducing the difficulty of training the convolutional neural network used and improves the recognition success rate of piston error range.In order to reduce the number of iterations of the phase diversity algorithm used in the fine cophase of segmented mirrors and reduce the difficulty of convergence,this paper proposes to introduce the cuckoo optimization algorithm.This optimization algorithm has a simple model,few parameters and is easy to implement.By improving the step size control factor and local search position update formula in the original cuckoo search algorithm,it has a faster optimization speed.Compared with other population-based algorithms,especially under the condition that the phase diversity algorithm has a large optimization range and many parameters to be solved,the improved variable step size adaptive gradient cuckoo selection algorithm of the phase diversity algorithm has the higher resolution accuracy,faster calculation speed and higher success rate,benefits from its local search ability and global search ability controlled by the discovery probability.In order to improve the robustness of the phase diversity algorithm used in the fine cophase of segmented mirrors to environmental micro-vibration,a blind demotion blurring technique based on conditional adversarial network(DeblurGAN)is introduced.In order to prevent the network from removing the image vibration blur,the image blur caused by the cophase error of the sub-mirrors was also incorrectly processed.DeblurGAN was retrained using image vibration dataset with different motion blur levels containing wavefront aberration.After the training of this algorithm is completed,it has the ability of blind deblurring in the training motion blur level,without the need to predict the image blur kernel.Secondly,in order to improve the robustness of the phase diversity algorithm used in the fine phase of the segmented mirror in the actual noise environment,a deep denoising convolutional neural network(DnCNN)is introduced into the preprocessing process of the phase diversity algorithm,which improves the noise robustness of the phase diversity algorithm.This network is trained using dataset with different noise levels.The successfully trained DnCNN has blind noise reduction capabilities in the noise training range,and no noise evaluation is required before image processing.The above-mentioned composite phase diversity method runs fast and does not increase the time of the fine cophase of the segmented primary mirror.