A method for predicting and correcting the distortion of vortex beams in advance under dynamic turbulence

Abstract

The present invention relates to a method for advanced prediction and correction of distorted vortex beams under dynamic turbulence. By using a pre-correction network model that combines a convolutional neural network and a time series network, the advanced prediction of the turbulence phase screen and the real-time correction of the distorted vortex light are realized, and a method for advanced prediction and correction of distorted vortex beams under dynamic turbulence is proposed. Different from the existing adaptive optics without a wavefront sensor and the adaptive optics correction method based on deep learning, this method extracts the turbulence perturbation characteristics in the intensity map of the distorted vortex light and uses the spatio-temporal coupling characteristics of atmospheric turbulence to establish a mapping relationship between the historical state and the future state of the turbulence phase screen. When the atmospheric evolution frequency is 30 frames / s, the pre-correction model can predict the future 1-6 frames of the turbulence phase screen in advance and perform real-time correction on the distorted vortex light. This method has achieved good results in terms of network prediction accuracy and wavefront correction timeliness compared with traditional correction methods, and is expected to be applied to an efficient and real-time closed-loop correction system for distorted vortex light.

Description

Technical Field

[0001] The present invention relates to a method for predicting and correcting distorted vortex beams in dynamic turbulence, and designs a pre-correction network model. By predicting the turbulence phase screen at future moments in advance, real-time correction of the distorted vortex light caused by dynamic atmospheric turbulence disturbance is carried out. When the frame rate of the turbulence phase screen is 30 frames / s, the model can accurately predict the turbulence phase screen of the next 1-6 frames in advance, and realize real-time correction of the vortex light closed-loop correction system based on deep learning. Technical Background

[0002] The vortex phenomenon in the optical field was first discovered by Boivin, Dow and Wolf near the focal plane of a lens group in 1967. In 1973, Bryngdahl first explored the experimental method for preparing vortex light. In 1979, Vaughan and Willets successfully prepared vortex light using a continuous laser. In 1990, Yu and Bazgenov V first completed the preparation of vortex light using the grating method. In 1992, L. Allen discovered that under paraxial conditions, the vortex beam with a phase factor has orbital angular momentum, where l is the topological charge number of the orbital angular momentum of the vortex light, is the azimuth angle; each photon carries of orbital angular momentum, is the reduced Planck constant. This angular phase factor indicates that during the propagation of the vortex light, if the light beam propagates one period, the wavefront just rotates one week around the optical axis, and the phase also changes by 2πl accordingly.

[0003] As a new type of structured light beam with a helical wavefront, vortex light has important application values in the fields of optical communication, particle micro-manipulation, motion detection, optical micro-measurement, etc. Laguerre-Gaussian light is a typical vortex light. The photons in the light beam not only have spin angular momentum (SAM), but also have orbital angular momentum (OAM). The topological charge number determines the size of OAM. The complete single-mode Laguerre-Gaussian beam has a circular ring intensity distribution and a hollow dark core. The region where the intensity at the beam center is zero is defined as the phase singularity. Vortex beams can be divided into two categories according to the type of phase singularity. One category is that the deflection direction of the optical field is the same and the phase of the singularity is uncertain, which is called the phase vortex light; the other category is that the polarization direction of the singularity is uncertain, which is called the vector vortex light. Laguerre-Gaussian light is a kind of phase vortex light. Superposition of multiple single-mode vortex lights can obtain a superposition state vortex light, which has different intensity and phase distributions from the single-state vortex light.

[0004] The generation of vortex light is the basis for the research on vortex light. Common generation methods include mode conversion method, computer-generated holography method, spatial light modulator method, Q-plate method, and matrix spiral phase plate method. Under laboratory conditions, the spatial light modulator method is a commonly used generation method. The spatial light modulator controls the electric field to cause changes in the spatial phase or amplitude image of the liquid crystal display, thereby writing certain information into the light wave and realizing the modulation of the light wave. By using the complex amplitude modulation technology to generate the holographic pattern of vortex light and loading it onto the spatial light modulator, and irradiating the spatial light modulator with a linearly polarized Gaussian beam, the output light is the vortex beam.

[0005] In the application of the long-distance transmission of superposition-state vortex light, the random fluctuations of the atmospheric temperature and pressure cause random fluctuations in the atmospheric refractive index, resulting in the distortion of the light wavefront, amplitude fluctuations, beam jitter, intensity scintillation of the vortex light, crosstalk between adjacent orders of modes, and the beam quality is severely affected. The accuracy of the measurement of the target parameters of the vortex light will also be severely affected. Therefore, through certain wavefront correction technology compensation, the long-distance transmission of vortex light can maintain good beam quality, which is very important for the application of vortex light in detection, communication, etc.

[0006] Currently, the common method for vortex optical wavefront distortion correction is adaptive optics technology. The distorted vortex light caused by atmospheric turbulence is compensated through components such as wavefront sensors, controllers, and wavefront correctors. Although adaptive optics technology can well restore the transmitted vortex optical wavefront, the adaptive optics system often has a highly complex structure. For the correction of distorted vortex light under atmospheric turbulence, traditional adaptive optics technology without a wavefront sensor is generally used. One is to use the stochastic - parallel - gradient - descent (SPGD) optimization algorithm to control the wavefront corrector to correct the vortex light, forming a feedback loop through cyclic iteration; the other is the vortex light turbulence pre - compensation scheme based on the Gerchberg - Saxton (GS) algorithm. Although both algorithms can correct the vortex light distorted by atmospheric turbulence, these algorithms often require long - term iteration and are prone to falling into local optimal solutions, having defects such as long processing time and difficulty in convergence. With the increase in the number of vortex light modes and the intensity of turbulence, under the requirement of a certain compensation effect, the difficulty of implementing the algorithm continuously increases, and the real - time performance of distorted vortex light correction becomes worse. Even the adaptive optics (AO) technology based on deep learning, which has developed rapidly in recent years and has a faster correction speed than traditional methods, still has a computational delay for the wavefront. Therefore, the current AO system is not essentially a real - time correction system. The entire AO system always corrects the distorted wavefront of the vortex light at the past moment, rather than the distorted wavefront at the current state. Therefore, in fact, no matter how fast the hardware speed is, it is not enough to eliminate the theoretical correction error caused by the random variation of turbulence. In the process of actually implementing the closed - loop correction system, when facing such a non - immediate compensation system, this patent proposes a pre - correction model, which is a real - time correction model with high correction accuracy.

[0007] Vortex optical wavefront distortion correction is of great significance for expanding applications. In the actual application process of the vortex light long - distance target detection field, the random fluctuations of atmospheric temperature and pressure cause random fluctuations in the atmospheric refractive index, resulting in optical wavefront distortion, amplitude fluctuations, causing vortex light beam jitter, light intensity scintillation, and mode crosstalk between adjacent orders, and the beam quality is severely affected. The accuracy of vortex light target parameter measurement will also be severely affected. Therefore, compensating and correcting the wavefront distortion through certain wavefront correction technologies, so that the vortex light can maintain good beam quality during long - distance transmission, is very important for the applications of vortex light in detection, communication, etc. This patent proposes a pre - correction model that can accurately predict the future - moment turbulence phase screen based on the past and current turbulence distributions and use the predicted results for timely compensation of distorted vortex light. Summary of the Invention

[0008] The technical problem solved by the present invention is as follows: The current AO system is not essentially a real-time correction system. The entire AO system always corrects the distorted wavefront of the vortex light at a past moment, rather than the distorted wavefront in the current state. A method for advanced prediction and correction of distorted vortex beams under dynamic turbulence is proposed. By using the spatio-temporal coupling characteristics of the dynamic atmospheric turbulence of the perturbed beam at low wind speeds, a mapping relationship between the historical state and the future state of the turbulent phase screen is established, and the future turbulent phase screen is predicted in advance, effectively solving the dilemma of correction lag and achieving real-time correction with high correction accuracy. In the proof-of-concept experiment, it only takes 90 ms to accurately predict the future 1 to 6 frames of turbulent phase screens, and the distorted vortex light can be corrected in a timely manner with good correction effects. This method has good flexibility, rapidity, and robustness, is easy to operate, and has strong generalization ability.

[0009] The pre-correction model consists of two parts: a coefficient mapping module and an advanced prediction module. The coefficient mapping module learns the turbulence characteristics in a large number of intensity maps of distorted vortex light based on a convolutional network and predicts the turbulent phase screens corresponding to the first 16 frames of distorted vortex light images, which are used as historical data and input into the advanced prediction module that extracts time series information based on a recurrent neural network. In the pre-correction model, the 16-frame turbulent screens predicted by the coefficient mapping module are used as historical data to input into the advanced prediction module to predict the 17th-frame turbulence information, and then the 2nd to 17th frames are used as the historical data of the advanced prediction module to predict the 18th frame and so on. That is, a sliding window with a size of 1*16 and a step size of 1 is used to continuously update the training set input into the advanced prediction module to train the advanced prediction module, so as to establish a non-linear mapping relationship between the past state and the future state and achieve the function of predicting the future 1 to 6 frames of turbulent phase screens from the known turbulent phase time series.

[0010] The technical solution of the present invention is:

[0011] The present invention relates to a method for advanced prediction and correction of distorted vortex beams under dynamic turbulence, which mainly includes the following steps:

[0012] (1) Build an experimental platform as Figure 3 shown, encode the vortex light hologram and load it onto the spatial light modulator (SLM-1 (UPOLabs HDSLM80R)), and project the expanded and collimated horizontal linearly polarized light onto the SLM-1 to prepare an undisturbed vortex light.

[0013] (2) Incorporate low wind speeds using the coefficients of the first 15 Zernike polynomials, and numerically simulate the continuous dynamic turbulence phase screen in combination with the spatio-temporal coupling characteristics of the dynamic atmosphere. Load the continuous dynamic turbulence phase screen onto SLM-2 (UPOLabs HDSLM80R) to simulate the phase perturbation of the vortex light by the turbulence. After the light beam passes through SLM-2 to obtain continuously distorted vortex light, use a CCD camera to detect and collect the light beam intensity distribution on the propagation path of the light beam. Make the collected data into a suitable size and input it into the network.

[0014] (3) The pre-correction model first passes through a coefficient mapping module improved based on the Convmixer model, and uses the convolutional network to learn the turbulence characteristics in a large number of distorted vortex light intensity maps and predict the turbulence phase screen corresponding to the first 16 frames of distorted vortex light images.

[0015] (4) The advanced prediction module part takes the turbulence information predicted by the coefficient mapping module as historical data input, predicts the wavefront aberration (Zernike polynomial coefficients) caused by the turbulence perturbation at future moments, and loads the advanced predicted turbulence phase screen onto SLM-2 before the next frame arrives at the turbulence evolution frequency, realizing high-precision real-time correction of the distorted vortex light.

[0016] The principle of the present invention is:

[0017] Laguerre-Gaussian light is a typical vortex light and is a set of solutions to the paraxial wave equation in the cylindrical coordinate system. When the propagation distance z = 0, its complex amplitude can be expressed in the cylindrical coordinate system as:

[0018]

[0019] Among them, U is the wave vector of the Laguerre-Gaussian light, is the cylindrical coordinate, r is the radial distance, is the polar angle, m is the topological charge number, p is the radial node number, ω0 is the beam waist radius of the fundamental mode Gaussian light, is the Laguerre polynomial, i is the imaginary unit, and π is the circumference ratio. The mode parameters of the vortex light include the topological charge number and the radial node number.

[0020] For the sake of concise expression and retaining the characteristics of the vortex light, Equation (1) can be abbreviated as:

[0021] E1 = Aexp(imφ) (2)

[0022] E2 = Aexp(-imφ) (3)

[0023] E = E1 + E2 (4)

[0024] Among them, E1 and E2 respectively represent the optical intensity vectors of two singlet vortex lights, E represents the optical intensity vector of the superposition state vortex light, A represents the intensity distribution, i is the imaginary unit, m is the topological charge number, and φ is the phase.

[0025] Using Richard D. Richmond and Stephen C. Cain's combination of the Zernike mode method and the frequency domain method, the dynamic simulation of the atmospheric turbulence phase screen is realized by the time-domain evolution of the Zernike polynomial. As follows:

[0026]

[0027] Among them, Θ atm (w, s, t) characterizes the atmospheric turbulence phase screen varying with time t, characterizes the Zernike polynomial within the circular domain, where the Zernike polynomial coefficient α j (t) can be obtained as follows:

[0028]

[0029] Among them is the defined time-correlated vector, which is a random vector with a value of 0 and a variance of , characterizes the square root of the covariance.

[0030]

[0031] Among them, D characterizes the beam receiving aperture, r0 is the atmospheric coherence length, characterizes the intensity of the beam perturbed by turbulence. According to the event correlation, the accurately correlated continuous phase screen coefficient vector can be obtained:

[0032]

[0033] Among them, A is the transmittance function of the telescope aperture, which is 1 when the aperture is open and 0 when the aperture is closed; using Bayes' law, the conditional mean and conditional variance of the time-correlated vector are respectively:

[0034]

[0035]

[0036] Among them is the correlation function of the time-correlated vector, as follows:

[0037]

[0038] From formula (6) and formula (8), can be further expressed as:

[0039]

[0040] After substitution and simplification, we get:

[0041]

[0042] where τ x = w2 - w1, τ y = s2 - s1, δτ x = δw2, δτ y = δs2, R θ is the phase correlation function; the first row of the numerator serves as an independent multiplier of the Zernike polynomial within the aperture. Let Define the inner product double integral operation using the property relationship of the Fourier transform Then formula (13) is further simplified to:

[0043]

[0044] The correlation of the atmospheric turbulence phase screen R θ cannot be directly calculated. Therefore, utilize the relationship between the phase structure and the correlation of the phase screen:

[0045]

[0046] Thus,

[0047]

[0048] where D θ is the phase structure function;

[0049] Solve to obtain For the given N-term Zernike polynomial, D, R0, and wind speed v x , v y The calculated According to equations (9) and (10), any can be accurately updated because for any given parameter conditions of atmospheric turbulence, only one calculation is required and it is pre-stored. When needed later, it can be directly called to quickly simulate the atmospheric turbulence phase screen Θ atm .

[0050] The present invention selects the first 15 terms of the Zernike polynomial to simulate the dynamic turbulence phase screen, adds a transverse wind speed of 90 mm / s, and sets the frame rate between the turbulence phase screens to 30 frames / s to simulate the perturbation of the actual vortex light by dynamic turbulence within one cycle time.

[0051] Under the background of using the rotational Doppler effect of vortex light in the superposition state with a large topological charge number for long-distance target detection, the ±20th order superposition state vortex light is selected as the research object (i.e., m = 20). A method for predicting and correcting the distortion of vortex beams in advance under dynamic turbulence is designed, and a pre-correction network model combining a convolutional neural network and a time series network is used to realize the real-time correction of adaptive optics. When the frame frequency of dynamic turbulence is 30 frames / s, the model can accurately predict the turbulence screens of the next 6 frames and correct the vortex light. After correction, the mode purity of the vortex light is generally increased to more than 90%. The invention proposes a pre-correction model, such as Figure 2As shown in the figure, the model consists of two parts, one is the coefficient mapping module and the other is the advanced prediction module. The pre-correction model first predicts the corresponding turbulence information in each distorted vortex beam image through a coefficient mapping module improved based on the Convmixer model; in the advanced prediction module, we use the prediction module to input the turbulence information predicted by the coefficient mapping module as historical data to predict the wavefront aberration caused by the turbulence perturbation at future moments. The advanced prediction module is an improvement based on LSTM. LSTM is an excellent variant model of the Recurrent Neural Network (RNN), inheriting most of the characteristics of the RNN model. It uses memory cells and gate mechanisms to control the transmission of sequence information, fully extracts the correlation information of time series, and solves the problems of gradient disappearance, gradient explosion, and "long term dependencies" in the training process of traditional RNN for long sequences to a certain extent. Compared with ordinary RNN, LSTM can perform better in longer sequences and is very suitable for dealing with problems highly related to time series. Therefore, it is selected to implement the advanced prediction function of the distorted vortex beam information under the dynamic turbulence model. In this model, we first use the coefficient mapping module to learn the turbulence characteristics of the corresponding frames from a large number of distorted vortex beam images and predict the turbulence screens corresponding to the first 16 frames of distorted vortex beam. Then, the predicted 16-frame turbulence screens are used as historical data to input into the advanced prediction module to predict the turbulence information of the 17th frame. Then, the 2nd to 17th frames are used as the historical data of the advanced prediction module to predict the 18th frame and so on. That is, a sliding window with a size of 1*16 and a step size of 1 is used to continuously update the training set input into the advanced prediction module to train the advanced prediction module, so as to establish a non-linear mapping relationship between the past state and the future state, and finally realize the function of predicting the phase change of 1 to 6 frames in the future from the known phase time series. Among them, in the coefficient mapping module, the image passes through a patch embedding layer to convert the original 2D image into a series of 1D patch embeddings, passes through a Gaussian Error Linear units (GELU) activation layer, and then passes through a Convmixer layer. The Convmixer block itself consists of a depthwise separable convolution (i.e., a grouped convolution with the number of groups equal to the number of channels h) and a pointwise convolution (i.e., a convolution with a kernel size of 1×1). After each convolution, there is an activation function GELU and batch normalization (BatchNorm) after activation. After multiple applications of the Convmixer block, global pooling is performed to obtain a feature vector of size h, which is then passed to the softmax classification.Finally, the trained coefficient mapping module can accurately predict the corresponding turbulent phase information of the distorted vortex light intensity map as the historical data of the lead prediction module. In our pre-correction model, the first 16 frames of turbulent phase screens are selected as the historical data and input into the lead prediction module improved based on LSTM. The LSTM network introduces a gate mechanism to control the flow and loss of features, namely the forget gate, input gate, and output gate. Among them, the forget gate is used to forget or discard some information and receives a long-term memory C. t-1 (the output passed from the previous unit module) and decides which part of C to retain and forget t-1 ; the input gate decides which new information to store, and finally, the output part of the cell state is determined by the output gate Sigmiod function, and the cell state is processed through the tanh layer.

[0052] Select the weighted linear combination of the mean square error (MSE) of the predicted value and the actual value of the Zernike coefficient as the objective loss function of the coefficient mapping module. Its expression is:

[0053]

[0054] where a i represents the true value of the i-th order Zernike coefficient, represents the predicted value of the i-th order Zernike coefficient, and t is the coefficient of the Zernike polynomial. The network optimizes the parameters of each layer with MSE as the objective function until the best values of each layer structure are reached when MSE reaches the lowest. Compare the mode purity of the optical field mode before and after correction, and the mode purity has been greatly improved, and the compensation effect is obvious.

[0055] The mode purity of the vortex light mode is defined as the mode proportion of any optical field Ψ and can be calculated as:

[0056]

[0057] where,

[0058]

[0059]

[0060] where, represents the Laguerre-Gaussian mode. In the present invention, the proportion of the required vortex light mode in the total mode is defined as the mode purity. Formula (18) is the orbital angular momentum spectrum expanded in the angular space. In the measurement of the mode purity of the present invention, the topological charge number l ranges from -30th order to 30th order.

[0061] Compared with the existing scheme, the main advantages of the scheme of the present invention are:

[0062] (1) The model can extract turbulence characteristic information through learning and establish a nonlinear mapping relationship between the past and future states of the turbulence phase screen. It can predict the turbulence phase screen of the sixth frame after the input distorted vortex light intensity image and make advance corrections to the distorted vortex light.

[0063] (2) It can buy 5 frames of turbulence screen evolution time for the hardware and software system response (the turbulence frame rate is 30 frames / s), thus improving the real-time performance of the correction system.

[0064] (3) The mode purity of the vortex light after the future frame turbulence phase screen correction predicted by the pre-correction model can generally reach more than 90%, which significantly improves the quality and stability of the vortex light beam, and provides an effective implementation plan model for the problem of the inability to truly correct the distorted vortex light in real time due to delays in the software and hardware systems during the application of the turbulence correction system. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of the advance prediction correction method;

[0066] Figure 2 This is the principle framework diagram of the pre-correction network;

[0067] Figure 3 This is the diagram of the experimental device and experimental platform;

[0068] Figure 4 This is a comparison diagram between distorted vortex light and undisturbed vortex light;

[0069] Figure 5 Diagram of the iterative training process of the Convmixer network

[0070] Figure 6 Diagram of the iterative training process for the LSTM network

[0071] Figure 7 Analysis of the compensation effect of the pre-correction model on the distorted vortex light Specific implementation plan

[0072] The present invention uses symmetrical superposition state vortex light as the experimental object and the implementation object is a spatial light modulator. The specific implementation steps are as follows:

[0073] First, the vortex phase hologram was loaded on SLM-1, and the perturbation phase was not loaded on SLM-2, so that SLM-2 only acted as a reflector in the optical path, and the original light vortex intensity distribution without turbulence was recorded. The experimental intensity distribution, phase distribution and mode purity of the superimposed OVs without atmospheric turbulence disturbance were collected, such as Figure 2As shown in the first line, it is obvious that the intensity distribution and phase distribution of the superposition state vortex light without disturbance are uniform. After measurement, the mode of the superposition state vortex light without disturbance can reach 94.99%. Then, a dynamic atmospheric turbulence phase screen based on the Kolmogorov turbulence theory is loaded on the SLM-2 screen. The atmospheric turbulence screen can be characterized by the atmospheric coherence length r0, and the atmospheric turbulence intensity can be expressed as where D is the beam diameter. The intensity distribution, phase distribution, and mode purity of the collected distorted OVs were experimentally measured, as shown in the 2nd, 3rd, and 4th of Figure 2 respectively. As shown in Figure 3 , the intensity distribution map of the sequence of distorted OVs with time correlation received by the CCD is used as the input of the pre-correction model. When inputting 16 frames of historical turbulence phase screens, for each input frame of turbulence information, the coefficients of the first 15 Zernike polynomials characterizing atmospheric turbulence in the subsequent 1 - 6 frames can be predicted. Thus, a turbulence phase screen is generated and its conjugate is taken to generate a compensation screen.

[0074] For example, first collect a large dataset of the intensity distribution of distorted superposition state vortex light and the corresponding Zernike polynomial coefficient labels. The turbulence intensity of this dataset is D / r0 = 5, and the dataset is divided into a training set and a test set in a ratio of 4:1. According to the numerical simulation theory of the dynamic turbulence phase screen, the MATLAB software is used to randomly generate the first 15 Zernike polynomial coefficient labels with added wind speed and time correlation.

[0075] The distribution curves of the loss function Loss of the coefficient mapping module and the prediction module during the model training process are shown in Figure 5 and Figure 6 respectively. As the number of iterations increases, the value of the loss function gradually decreases and approaches 0. The four points A, B, C, and D in the two figures intuitively reflect the trend of the target loss function decreasing as the number of iterations increases. Figure 5 The curve representing the loss function of the coefficient mapping module decreasing with the number of iterations reaches a flat stage and approaches 0 after about 160 iterations. The coefficient mapping module can accurately predict the corresponding turbulence characteristic information based on the input intensity distribution of the distorted vortex light. As can be seen from Figure 6 , after about 500 iterations, the loss function curve of the forward prediction module reaches a flat stage and there is no further downward trend. The turbulence phase screen predicted by this module is very close to the actual value.

[0076] Analyze and verify the effect of compensating and correcting vortex light with a network for predicting turbulent phase screens in advance. Use a workstation server (AMD Ryzen Threadripper 3960×24-core processor@3.79GHz) to process the trained network, which can cyclically predict the turbulent phase screens of the next 1 to 6 frames within 90 ms to correct the distorted vortex light. From Figure 7 It can be seen from the first three columns that after being disturbed by different turbulent phase screens, the intensity distributions of the true distorted vortex light in the future frames have shown turbulent disturbance characteristics such as beam deformation and light intensity flicker to varying degrees, and the mode purity of the vortex light after being disturbed has decreased to varying degrees. After loading the pre-compensation model to predict the turbulent compensation screen in advance, when the mode purity of the vortex light without turbulent disturbance is 94.99%, the beam quality of the compensated 1 to 6 consecutive frames of distorted vortex light has been improved to over 92%, the light intensity distribution is more uniform, and the beam shape is a relatively standard circle. Therefore, the experiment proves that the pre-compensation network has the function of quickly and accurately predicting the turbulent compensation screens of the next 6 frames in advance, providing an effective implementation model for the problem that the distorted vortex light cannot be truly corrected in real time due to the delay of the software and hardware system during the implementation process of the turbulent correction system.

[0077] In addition, the spatial light modulator has certain limitations on the incident angle and power of the beam, so the specific optical path design also needs to be carried out according to the actual situation of the laboratory.

[0078] The content not described in detail in this invention book belongs to the prior art well-known to those skilled in the art.

Claims

1. A method for predicting and correcting the distortion of vortex beams in advance under dynamic turbulence, characterized in that: This method proposes a pre-correction model consisting of a coefficient mapping module and a lead prediction module by establishing a pre-correction model to learn the mapping relationship between the historical state and the future state of turbulence. It can lead-predict the future 1-6 frame turbulence phase screens and perform real-time correction on the distorted vortex light at the atmospheric turbulence evolution frequency of 30 frames / s. The specific steps are as follows: (1) Build an experimental platform, encode the vortex light hologram and load it onto the spatial light modulator SLM-1. Shine the horizontally polarized light that has been expanded and collimated onto SLM-1 to prepare undisturbed vortex light; (2) Add low wind speed using the first 15 Zernike polynomial coefficients, and numerically simulate the continuous dynamic turbulence phase screen in combination with the dynamic atmospheric spatio-temporal coupling characteristics. Load the continuous dynamic turbulence phase screen onto SLM-2 to simulate the phase perturbation of the vortex light by turbulence. The light beam passes through SLM-2 to obtain continuously distorted vortex light. After passing through the beam collimation and filtering system, use a CCD camera to detect the beam intensity distribution and collect it on the propagation path of the light beam. Make the collected batch of dataset pictures into a network of appropriate size as the network input; (3) The pre-correction model first passes through a coefficient mapping module improved based on the Convmixer model, and uses a convolutional network to learn the turbulence characteristics in a large number of distorted vortex light intensity maps and predict the turbulence phase screens corresponding to the first 16 frames of distorted vortex light images; (4) The lead prediction module part takes the turbulence information predicted by the coefficient mapping module as historical data input, cyclically predicts the wavefront aberration caused by the future 1-6 frame turbulence perturbations, and loads the lead-predicted turbulence phase compensation screen onto SLM-2 before the arrival of the next frame at the turbulence evolution frequency to achieve high-precision real-time correction of the distorted vortex light.

2. A method for predicting and correcting the distortion of vortex beams in dynamic turbulence according to claim 1, characterized in that: In the pre-correction model, the coefficient mapping module learns the turbulence characteristics in a large number of distorted vortex light intensity maps based on a convolutional network and predicts the turbulence phase screens corresponding to the first 16 frames of distorted vortex light images, and takes it as historical data input to the lead prediction module that extracts time series information based on a recurrent neural network.

3. A method for predicting and correcting the distortion of a vortex beam in advance under dynamic turbulence according to claim 1 or claim 2, characterized in that: In the pre-correction model, the 16-frame turbulence screens predicted by the coefficient mapping module are used as historical data input to the lead prediction module to predict the 17th frame of turbulence information, and then the 2nd to 17th frames are used as the historical data of the lead prediction module to predict the 18th frame. That is, a sliding window with a size of 1*16 and a step of 1 is used to continuously update the training set input to the lead prediction module to train the lead prediction module, so as to establish the non-linear mapping relationship between the past state and the future state of the turbulence phase screen and realize the function of predicting the future 1 to 6 frame turbulence phase screens from the known turbulence phase time series.

Images