BP neural network-based frequency multiplication wavefront detection method
阅读说明：本技术 一种基于bp神经网络的倍频波前探测方法 (BP neural network-based frequency multiplication wavefront detection method ) 是由 徐志强 董理治 赵旺 何星 王帅 杨平 许冰 于 2019-09-30 设计创作，主要内容包括：本发明公开了一种基于BP神经网络的倍频波前探测方法,属于光学信息测量技术领域,利用泽尼克系数表征含畸变的基频光束和相应倍频光束的波前信息并组成训练集,再通过BP神经网络拟合两者之间的非线性关系,BP神经网络训练完成后,即可通过采集到的倍频光波前信息,直接复原出全口径的待测基频光束的波前信息。本发明将神经网络应用到基频、倍频光束相位关系的拟合上,简化了基于非线性倍频波前探测技术的测量过程,且模型预测过程迅速,足以满足实时波前探测的需求。(The invention discloses a frequency doubling wavefront detection method based on a BP neural network, which belongs to the technical field of optical information measurement, and is characterized in that wavefront information of a distorted fundamental frequency beam and a corresponding frequency doubling beam is represented by a Zernike coefficient to form a training set, and then a nonlinear relation between the fundamental frequency beam and the corresponding frequency doubling beam is fitted through the BP neural network, so that the wavefront information of the full-aperture fundamental frequency beam to be detected can be directly recovered through the acquired frequency doubling wavefront information after the BP neural network is trained. The invention applies the neural network to the fitting of the phase relation of the fundamental frequency and the frequency doubling light beams, simplifies the measuring process based on the nonlinear frequency doubling wavefront detection technology, has rapid model prediction process and can meet the requirement of real-time wavefront detection.)
1. A frequency multiplication wave front detection method based on a BP neural network is characterized in that: the method uses a BP neural network to construct a nonlinear relation between the phases of fundamental frequency light and frequency doubling light, realizes frequency doubling wavefront detection by using the relation, and specifically comprises the following steps:
s1, constructing a training set according to the fundamental frequency light and the frequency doubling light wave front Zernike coefficients;
s2, training a neural network model of the corresponding relation between the fundamental frequency light wave front zernike coefficient and the frequency doubling light wave front zernike coefficient based on the BP neural network;
s3, after the neural network model is built, frequency doubling is carried out on the base frequency light beam to be detected through the nonlinear crystal, the generated frequency-doubled light beam enters the wavefront sensor, and the wavefront sensor inputs the detected wavefront information into the computer;
s4, the computer calculates the Zernike coefficient of the frequency doubling wavefront according to the detected frequency doubling wavefront information and uses the Zernike coefficient as input, and the Zernike coefficient of the fundamental frequency wavefront to be detected is obtained by using the BP neural network model established in S2;
and S5, restoring the wavefront by the computer according to the Zernike coefficient of the fundamental frequency light wavefront to be detected, and completing the wavefront detection.
2. The frequency-doubling wavefront survey method based on the BP neural network according to claim 1, wherein the detailed construction process of the fundamental frequency light and frequency-doubling wavefront Zernike coefficient training set in step S1 is as follows:
(S11) the plane wave fundamental frequency light beam enters a wavefront sensor after frequency multiplication, and wavefront information of the frequency multiplication light beam at the moment is recorded as a calibration wavefront to counteract additional aberration generated by the beam shrinking and expanding lens and the spectroscope optical element;
(S12) generating random aberration by phase modulator, calculating its Zernike coefficient (Z)1，Z2，...，Zm) After the plane wave is reflected by the phase modulator, a fundamental frequency light beam containing phase distortion is generated and is emitted into the frequency doubling crystal;
(S13) the frequency-doubled light beam enters a wave front detector, and the wave front detector inputs the detected wave front information into a computer to calculate the Zernike coefficient (Z) of the wave front information1′，Z2′，...，Zn') record the corresponding fundamental frequency light and frequency doubling wave front Zernike coefficients { (Z)1，Z2，...，Zm)、(Z1′，Z2′，...，Zn') } as a set of training samples;
(S14) repeating the steps (S12) and (S13), randomly generating several groups of training samples to form a training set:
{(Z1，Z2，...，Zm)、(Z1′，Z2′，...，Zn′)}N。
3. the frequency-doubling wavefront detection method based on the BP neural network as claimed in claim 1, wherein the network model structure and training process of step S2 are as follows:
(S21) setting the number of input layer units as n, namely n-order Zernike coefficients before the wavefront of the frequency doubling light, and the number of output layer units as m, namely m-order Zernike coefficients before the wavefront of the fundamental frequency light;
(S22) number of hidden layer units:
(S23) excitation function:
(S24) error formula:
(S25) training the BP neural network model by adopting a gradient descent method, selecting N groups of training set samples for training, continuously updating the weight and the threshold value in the iterative network model until the total error of the BP neural network is less than a set value or the number of iterations is reached, and finishing the training.
Technical Field
The invention belongs to the technical field of optical information measurement, and particularly relates to a frequency-doubling wavefront detection method based on a BP neural network.
Background
The wavefront detection technology is a technology for measuring the wavefront distortion of light beams, and reconstructs the wavefront information of the light beams to be measured by sampling and modulating the light beams to be measured and measuring corresponding signals. Currently, common wavefront sensors include shack-hartmann wavefront sensors, shearing interferometers, curvature sensors, and the like. In view of the important application value of wavefront detection in the fields of optical detection, space target imaging, laser beam purification and the like, the development of a novel precise, sensitive, flexible and universal wavefront detection technology is always a hot point of research of people.
The existing researchers combine nonlinear frequency multiplication with wavefront detection to expand the measurement range of the traditional wavefront detection technology, but are affected by factors such as phase mismatch and walk-off effect, and the transmission result of aberration in the frequency multiplication process is not linearly superposed but nonlinear, so that direct detection is difficult. The method adopted by researchers at present is to measure the distribution of frequency-doubled light intensity through a fiber coupler with a movable tail end to calculate the phase mismatch amount at different positions in the caliber of a light beam, then to find the Wave vector direction at different positions according to the relation between the phase mismatch amount and the Wave vector of the light Wave, and then to fit the Wave front of the fundamental frequency light beam to be measured (see Roc i o Borrego-varias, Carolina Romero, et al, "Wave-front regenerative amplified microscopic beam by second-harmonic generation" Optics express.19(23),22851,2011). However, the method is only suitable for one-dimensional wavefront detection at present, and the measurement process is complex, so that the requirement of real-time wavefront detection cannot be met.
Disclosure of Invention
In view of the above problems, the present invention provides a frequency-doubling wavefront detection method based on a BP neural network, which is a feedforward neural network based on an error back propagation algorithm, and can effectively fit a nonlinear relationship between a fundamental frequency and a frequency-doubling wavefront, thereby realizing real-time full-aperture frequency-doubling wavefront detection.
In order to achieve the purpose, the invention provides the following technical scheme:
a frequency multiplication wave front detection method based on a BP neural network comprises the following steps:
s1, constructing a training set according to the fundamental frequency light and the frequency doubling light wave front Zernike coefficients;
s2, training a neural network model of the corresponding relation between the fundamental frequency light wave front zernike coefficient and the frequency doubling light wave front zernike coefficient based on the BP neural network;
s3, after the neural network model is built, frequency doubling is carried out on the base frequency light beam to be detected through the nonlinear crystal, the generated frequency-doubled light beam enters the wavefront sensor, and the wavefront sensor inputs the detected wavefront information into the computer;
s4, the computer calculates the Zernike coefficient of the frequency doubling wavefront according to the detected frequency doubling wavefront information and uses the Zernike coefficient as input, and the Zernike coefficient of the fundamental frequency wavefront to be detected is obtained by using the BP neural network model established in S2;
s5, the computer recovers the wavefront according to the Zernike coefficient of the fundamental frequency light wavefront to be detected, and the wavefront detection is completed.
As a further aspect of the present invention, the detailed construction process of the training set in step S1 is as follows:
s11, the plane wave base frequency light beam enters the wave front sensor after frequency multiplication, and the wave front information of the frequency multiplication light beam is recorded as the calibration wave front to offset the additional aberration generated by the optical elements such as the beam shrinking and expanding lens, the spectroscope and the like;
s12, generating random aberration by phase modulator, calculating its Zernike coefficient (Z)1,Z2,…,Zm) After the plane wave is reflected by the phase modulator, a fundamental frequency light beam containing phase distortion is generated and is emitted into the frequency doubling crystal;
s13, the generated frequency-doubled light beam enters a wave front detector, the wave front detector inputs the detected wave front information into a computer, and then the Zernike coefficient (Z) of the wave front information is calculated1′,Z2′,…,Zn') record the corresponding fundamental frequency, frequency multiplication Zernike coefficients { (Z)1,Z2,…,Zm)、(Z1′,Z2′,…,Zn') } as a set of training samples;
s14, repeating the steps S12 and S13, and randomly generating a plurality of groups of training samples to form a training set:
{(Z1,Z2,…,Zm)、(Z1′,Z2′,…,Zn′)}N。
as a further aspect of the present invention, the BP neural network in step S2 is constructed as follows:
s21, the number of input layer units is n, namely n-order Zernike coefficients before the frequency multiplication wave front, and the number of output layer units is m, namely m-order Zernike coefficients before the fundamental frequency wave front;
s22, number of hidden layer elements:
s23, excitation function:
s24, error equation:
And S25, selecting N groups of training set samples for training by adopting a gradient descent method for training the BP neural network model, and continuously updating the weight and the threshold value in the iterative network model until the total error of the BP neural network is less than a set value or the number of iterations is reached, thus finishing the training.
Compared with the prior art, the invention has the beneficial effects that:
(1) fitting a nonlinear relation between the fundamental frequency and the frequency doubling phase by adopting a BP neural network model, and directly detecting the frequency doubling wavefront, namely recovering the full-caliber fundamental frequency wavefront to be detected;
(2) the network model is trained by the mode of collecting training samples under the line, the actual online detection process can be used for rapidly restoring the wavefront to be detected, and the requirement of real-time wavefront detection is met.
Drawings
FIG. 1 is a schematic diagram of a training set construction process;
FIG. 2 is a schematic diagram of a BP neural network model;
FIG. 3 is a block diagram of a frequency-doubled wavefront sensing process;
FIG. 4 is a schematic diagram of a wavefront sensing process according to the present invention.
In the figure: 1. the laser beam splitter comprises a first beam-reducing lens, a second beam-reducing lens, a KDP frequency doubling crystal, a
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to specific embodiments and the accompanying drawings. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.