CN-119376095-B - Wavefront correction method of wavefront-free detection self-adaptive optical system based on deep learning

CN119376095BCN 119376095 BCN119376095 BCN 119376095BCN-119376095-B

Abstract

The invention discloses a wavefront-free detection self-adaptive optical system wavefront correction method based on deep learning, which comprises the steps of S110, constructing a neural network, S120, constructing a data set, S130, processing the loss of back propagation by using a loss function, reversely propagating a smaller loss, enabling the network to learn a certain mapping relation, S140, designing a wavefront-free detection self-adaptive optical system wavefront correction algorithm for the neural network, and realizing closed loop correction of the wavefront-free self-adaptive optical system. According to the technical scheme of the invention, the wavefront correction speed and the wavefront correction precision are greatly improved, and the rapid and high-precision wavefront-free detection self-adaptive optical system wavefront correction is realized.

Inventors

ZHONG LIBO
ZHANG LINGXIAO
RAO CHANGHUI

Assignees

中国科学院光电技术研究所

Dates

Publication Date: 20260512
Application Date: 20241101

Claims (2)

1. The wavefront correction method of the wavefront-free detection self-adaptive optical system based on the deep learning is characterized by comprising the following steps of: s110, constructing a neural network for training a mapping relation from far-field focal plane light intensity to a group of complex conjugated wavefront data corresponding to the far-field focal plane light intensity; S120, constructing a data set, taking far-field focal plane light intensity as input of a neural network, and taking a pair of complex conjugate wave-front data corresponding to the far-field focal plane light intensity as standard output of the neural network, wherein the data set comprises a plurality of groups of data sets, one group of data sets comprises a far-field focal plane light intensity image and a pair of wave-front data, and the pair of wave-front data has the following relation , And Is a pair of complex conjugate wave fronts, The system is pupil plane coordinates, wavefront data is composed of Zernike coefficients conforming to Kolmogorov distribution, a data set is divided into a training set, a verification set and a test set, the training set and the verification set are used for a neural network to learn a mapping relation from far-field focal plane light intensity to one wavefront, and the test set is used for verifying the accuracy of the neural network; S130, processing the back propagation loss by using a loss function, wherein the back propagation loss is smaller, so that the network learns one of the determined mapping relations, and the loss function expression is as follows: , Wherein, the For the Zernike coefficients of the neural network output, And Is a pair of complex conjugate Zernike coefficients corresponding to far-field focal plane light intensity, namely standard output, And Is the root mean square error of a pair of standard outputs and neural network outputs, Is the final back propagation loss of the network, The root mean square error is expressed as: , Wherein, the Output as network The order Zernike coefficients are chosen to be, Is the standard output of the first The order Zernike is such that, Is the total order of the Zernike coefficients; The wavefront correction method comprises the steps of S140, designing a wavefront correction algorithm of a wavefront-free detection self-adaptive optical system for a neural network to realize closed loop correction of the wavefront-free self-adaptive optical system, performing wavefront correction on the self-adaptive optical system according to output of the neural network by the wavefront correction algorithm of the wavefront-free detection self-adaptive optical system, firstly determining an evaluation function of far-field focal plane light intensity, wherein the evaluation function is a Style ratio or girth energy, calculating a far-field focal plane light intensity evaluation function value at the beginning of each round of iteration, outputting a Zernike coefficient by the neural network according to the current far-field focal plane light intensity, obtaining a pair of Zernike coefficients which are complex conjugates by the wavefront-free detection self-adaptive optical system according to the Zernike coefficient output by the neural network, respectively applying two control signals according to the Zernike coefficients, calculating to obtain two evaluation function values, taking a control signal corresponding to the higher evaluation function value as a control signal finally applied by the round of iteration, and stopping algorithm iteration if the two evaluation function values are smaller than the evaluation function value at the beginning of iteration, and completing the self-adaptive optical system correction.
2. The method for wavefront correction of a non-wavefront-sensing adaptive optics system based on deep learning of claim 1, wherein the neural network is a convolutional neural network.

Description

Wavefront correction method of wavefront-free detection self-adaptive optical system based on deep learning Technical Field The invention relates to the technical field of wavefront correction of wavefront-free detection self-adaptive optical systems, in particular to a wavefront correction method of the wavefront-free detection self-adaptive optical system based on deep learning. Background The self-adaptive optical technology can compensate wavefront aberration and has wide application in the fields of laser transmission, beam purification and astronomical observation. Conventional adaptive optics systems utilize a wavefront sensor to detect the wavefront and control a wavefront corrector to compensate for the wavefront. However, the adaptive optics system based on the wavefront sensor is difficult to function in scenes such as strong aberration level, long horizontal transmission distance, dark and weak targets, and the like, and the system complexity and the system cost are high. Wave front detection-free adaptive optics technology does not need a wave front sensor to detect wave front, and far-field light intensity information is directly utilized to conduct iterative optimization to compensate wave front aberration, so that possibility is provided for application of the scenes. The wavefront-free detection adaptive optical system has lower system complexity and system cost and is not influenced by the aberration of the non-public optical path because a wavefront sensor is not needed. The wave front detection-free self-adaptive optical system based on the model-free optimization algorithm takes the performance index of the far-field image as a function of the control parameter, and the performance index is improved by iterative optimization of the optimization algorithm. 1997 m.a. Vorontsov et al proposed a random parallel gradient descent algorithm (SPGD) that directly uses far field information to perform closed loop correction of aberrations. Subsequent researchers have used other optimization algorithms in wavefront-free detection adaptive optics techniques as well. Closed loop correction of wavefront-free detection adaptive optics is achieved by using a simulated annealing algorithm (SA) as in 2006 S. Zommer et al, 2024 et al, huanhuan Yu, using an asymptotic endpoint Algorithm (APP). However, when the aberration is large, the optimization algorithm is easy to fall into the optimal solution, and the algorithm needs multiple iterations to realize aberration compensation, so that the real-time performance is poor. In recent years, the deep learning algorithm is widely applied in the field of adaptive optics due to the advantages of high real-time performance, high learning capacity, high generalization capacity and the like, and some researchers use the deep learning algorithm in the wavefront-free detection adaptive optics technology. 2019 Huimin Ma et al used AlexNet to estimate Zernike coefficients from the focal plane and out-of-focus images for wavefront correction without multiple iterations for wavefront-sensing adaptive optics, but additional out-of-focus cameras added to the system complexity and system cost. 2019 Nishizaki et al uses Xception to reconstruct the wavefront directly from a single light intensity image, but because the far field focal plane image corresponds to a pair of wavefronts that are rotated 180 ° from each other in complex conjugate relationship, the accuracy of the wavefront reconstruction of the single focal plane light intensity is poor due to the influence of this double solution problem. Although the depth learning-based wavefront-free detection adaptive optics technique greatly reduces aberration correction time, the existing depth learning-based wavefront-free detection adaptive optics technique is difficult to exhibit good correction effects when only focal plane information is available. The existing method still has a certain problem in the closed loop correction of the actual wavefront-free detection self-adaptive optical system, seriously influences the wavefront correction speed and correction precision of the wavefront-free detection self-adaptive optical system, and how to realize the rapid and high-precision wavefront correction of the wavefront-free detection self-adaptive optical system under the condition of using only far-field focal plane images is the problem which still needs to be solved at present. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a wavefront correction method of a wavefront detection-free adaptive optical system of an adaptive optical system based on deep learning, by training a neural network, the trained neural network can estimate one group of wavefront according to far-field focal plane light intensity, and by combining a wavefront correction algorithm of the wavefront detection-free adaptive optical system for the neural network, the rapid high-precision wavefront correction of the wav