Nonparametric Point Spread Function Model Based On Deep Neural Networks For Optical Telescopes

Optical telescope is a kind of photoelectric instruments that are used for astronomical observations.With the development of modern astronomy,requirements of improved performance of astronomical telescopes are constantly increasing.It would be a possible way to improve resolution and sensitivity of telescopes by increasing its aperture size,but ground-based astronomical optical telescopes are affected by many factors.For example,small changes of temperatures,or the structure may cause image quality degradation.The variation of the imaging quality of optical system can be represented by the point spread function model(PSF).PSF represents the impulse response of the imaging system.By studying properties of PSFs,it is possible to improve the data quality.In this paper,we first review the two main methods for PSF modelling.In general,a lot of research has been done in modelling of ground-based telescope PSFs,but applications mainly focus in long exposure PSFs and the modelling methods are mainly parametric models.This is because the shape of long exposures PSFs is symmetrical and suitable for parametric models.The shape of short-exposure PSF is complex,which is difficult to be represented by parametric model.However,for astronomical observations such as rapid sky survey,lucky imaging and solar observations,the short exposure is necessary,so modelling of short-exposure PSFs also needs to be studied.In this paper,a data-driven approach is developed for PSF modelling of the Wide Field Small Aperture Telescopes(WFSAT)with arbitrary states and exposure time.In this paper,the effect brought by atmospheric turbulence and that brought by static aberrations are discussed separately.Because the PSF of the whole optical system represents the overall effect at each stage of the imaging process,PSF can also be disassembled and applied at different stages.Overall effects are the same.The advantage of separation of PSF modelling is that they can be carried out step by step in a more detailed way.After a detailed analysis of the degradation process for WFSAT imaging,we will further discuss modelling of PSFs.The deep neural networks(DNN)are used to build PSF models,PSF-NET and TEL-NET.The PSF-NET is used to model atmospheric turbulence induced PSFs and the TEL-NET is used to model PSFs brought by static aberrations.According to test results,PSF models presented in this paper can represent the PSF manifold space induced by any turbulence profiles and estimate PSFs indifferent field of views.This would help scientists in designing new methods to increase quality of astronomical images.