Research On Wavefront Sensorless Adaptive Optics Based On Deep Learning

Due to atmospheric turbulence and other factors that cause the beam to be prone to wavefront aberrations during propagation,making the images captured by telescopes and other devices suffer from severe degradation,and adaptive optics is an important means of compensating for the aberrated wavefront.According to the different ways of measuring the wavefront information,an adaptive optics system can be divided into two categories:wavefront sensor adaptive optics(WFS AO)systems and wavefront sensorless adaptive optics(WFS-less AO)systems.WFS AO systems add a separate measurement optical path,making the optical system structure more complex.More importantly,the use of WFS AO systems have limitations as the application areas of adaptive optics continue to expand,such as optical communication,extended target imaging and laser intracavity aberration correction.The WFS-less AO technology abandons the traditional wavefront sensor,and uses the iterative optimization control algorithm to take the correction effect as the evaluation index,and the voltage information required by the wavefront corrector as the optimization parameter to realize the distorted wavefront correction through iterative optimization,which has the advantages of simple structure,simple operation and high accuracy,but still has slow convergence speed and difficult to meet the real-time requirements of aberration correction.In recent years,machine learning,especially data-driven deep learning,has been introduced into the field of AO technology.This kind of method trains the network model by constructing a network model,using the focal plane light intensity image as the input and the distorted wavefront informations as the outputs,and finally realizes the direct outpust from the light intensity image to the front end of the distorted wave,which can calculate the turbulence aberrations at one time without iteration.Therefore,the speed of aberration resolution can be greatly improved.However,the technology is still in the initial stage of basic research,and many practical core issues still need to be verified.Therefore,in view of the current problems of AO technology,this paper focuses on the research of WFS-less AO technology based on deep learning,and the main work of the paper is as follows:1)A WFS-less AO system based on deep learning is built,and study the numerical simulation method of atmospheric turbulence.Finally,the Kolmogorov turbulence theory is selected to simulate atmospheric turbulence with different intensities,on this basis,an accurate optical imaging system model was established,which was consistent with the actual physical imaging process.The system focal plane degradation images corresponding to the observed targets under different atmospheric turbulence intensities were obtained,and a large number of reliable available data sets including focal plane degradation images and corresponding turbulence aberrations were established.2)Different deep learning neural networks have been built,including ordinary CNN,Res Net networks and Efficient Net-B0 network,and the focal plane degradation image in the established data sets have been used as the input of the network,and the corresponding atmospheric turbulence aberrations as the outputs of the network.The structure of the deep neural network and the specific network training algorithm have been studied with the objective of aberration detection accuracy,detection range and solution speed.The networks models between the target degraded image and the turbulence aberration coefficients were accurately established.Then,a large number of focal plane degradation images are generated by the established simulated AO system,and several depth neural networks are tested to verify the accuracy,detection range and speed of the algorithm.The WFS-less AO technology based on deep learning is compared with the traditional WFS-less AO technology based on SPGD algorithm.3)The deconvolution strategy of degraded image based on the solution of wavefront aberration is studied.Based on the wavefront aberration predicted by the Efficient Net-B0 network,the point spread function of the imaging system is solved.On the basis of the system degradation model,the restoration calculation of the degraded image of the observation target is carried out through wiener filtering,and the high-resolution imaging of the observation target of the WFS-less AO system is obtained.