| Adaptive optics can keep the optical system in the best working state.According to whether the wavefront information is measured or not,the adaptive optics system can be divided into two categories: the adaptive optics system based on wavefront detection and the adaptive optics system without wavefront detection.The adaptive optics system without wavefront detection mainly adopts iterative optimization control algorithm,and its structure is simple.Due to the convergence of iterative optimization algorithm,the time-domain control bandwidth of the adaptive optics system is low,and it is difficult to meet the requirements of real-time correction.Deep learning has good function fitting ability,and is expected to establish the nonlinear mapping between far-field spot and incident wavefront.Therefore,the control strategy based on deep learning is studied in this paper to solve the problem of low correction bandwidth,which has important application value.Firstly,the numerical simulation model is established.Compared with SPGD algorithm,the neural network control algorithm has lower correction accuracy,but it does not need iteration,and can recover the incoming and outgoing wavefront well only by one frame of far-field spot.On this basis,the characteristics of neural network control algorithm are further analyzed.The simulation results show that the convolution neural network is better than the common depth neural network model in extracting image features.The restoration accuracy of convolution neural network for low-order Zernike aberrations is better than that for higher-order Zernike aberrations.Finally,a 39 unit without wavefront AO experiment device detection is built,the feasibility of the control method is preliminarily verified.And by Compared with the traditional iterative optimization control method,the neural network control algorithm can quickly get the control signal based on the far-field spot image,and has obvious advantages in real-time.Figure 46;Table 8;Reference 53... |
PDF Full Text Request