Atmospheric turbulence phase space-time estimation algorithm

Abstract

The invention provides an atmospheric turbulence phase space-time estimation method. The method comprises the following steps: acquiring an atmospheric turbulence phase diagram introduced by atmospheric turbulence, and processing the atmospheric turbulence phase diagram into N parts of equal-length atmospheric turbulence phase diagram data, wherein each part of data comprises an atmospheric turbulence phase diagram sequence and an atmospheric turbulence phase diagram of the next time period; establishing a neural network model, taking the atmospheric turbulence phase diagram sequence as network input, taking the atmospheric turbulence phase diagram in the next time period as a network golden standard, calculating a loss function value of network output and the golden standard, and carrying out gradient back propagation of the loss function value to update parameters of a network model; and actually measuring a section of atmospheric turbulence phase diagram and inputting into the network model to obtain the atmospheric turbulence phase diagram of the next time period. According to the invention, the real-time performance of phase measurement of the adaptive optical system can be ensured.

Description

technical field [0001] The invention belongs to the field of adaptive optics, and relates to a space-time prediction method of atmospheric turbulence phase. Background technique [0002] Due to the high frequency of atmospheric turbulence transformation, the existing wavefront sensing hardware and phase extraction algorithms are relatively time-consuming, which cannot guarantee the real-time and accuracy of the phase measurement of the adaptive optics system. Therefore, deep learning is needed to learn from the subject The phase information of atmospheric turbulence is extracted from the distortion intensity information affected by atmospheric turbulence. [0003] Deep learning is a technology that learns the nonlinear relationship between the input image and the output image from a large number of data pairs by building a neural network. When the existing deep learning method is used to measure the phase of atmospheric turbulence, it only considers using the current distor...

AI Technical Summary

Problems solved by technology

[0004] 1. The existing deep learning methods for predicting follow-up situations all use a large number of cyclic and stacked network structures, which makes the calculation amount larger and cannot meet the real-time performance of the phase measurement of the adaptive optics system

[0005] 2. The existing cyclic neural network for estimating the follow-up situation only considers the dependence of time series, but ignores the dependence between the spatial structure and channels of two-dimensional images, so that the characteristics of many time series images cannot be well utilized. Unable to meet the real-time and accuracy of the phase measurement of the adaptive optics system

[0006] 3. The existing cyclic neural network for predicting the follow-up situation is very easy to train and overfit, so that it can only guarantee the accuracy of the training data, but cannot meet the accuracy of the actual adaptive optics system

Images