Adaptive Optics Real-Time Compensation Method, Device, Storage Medium and System

Abstract

The present invention provides an adaptive optical real-time compensation method, device, storage medium and system, which relates to the field of optical communication technologies. The method includes: extracting historical wavefront phase diagrams of P historical time steps before the current time step from the wavefront phase diagram; constructing node features and an adjacency matrix based on each pixel node in the historical wavefront phase diagrams to obtain the graph data of the historical wavefront phase diagrams, and inputting the graph data into a pre-trained graph multi-head attention model to output predicted wavefront phases for Q future time steps; reconstructing the predicted wavefront phases into a phase distribution form suitable for correction by a phase modulator, and mapping the reconstructed predicted wavefront phases into control voltage signals of the phase modulator and then inputting the control voltage signals into the phase modulator, so that the phase modulator adjusts the phase structure based on the control voltage signals to perform real-time phase compensation on the actual wavefront phases for Q future time steps. The present invention can solve the problem that the existing wavefront phase prediction and turbulence compensation methods have time delay.

Description

Technical Field

[0001] The present invention relates to the field of optical communication technologies, and in particular, to an adaptive optical real-time compensation method, device, storage medium, and system. Background Art

[0002] Atmospheric laser communication is regarded as a core technology to break through the bottleneck of wireless access due to its high bandwidth, strong anti-interference ability, and high security, and shows great application potential in fields such as satellite communication and long-distance ground transmission. However, its signal transmission highly depends on the stability of the atmospheric channel and is vulnerable to atmospheric turbulence interference. Atmospheric turbulence can cause wavefront phase distortion and light intensity scintillation effects of the light beam, seriously reducing the reliability and transmission quality of the communication link. Therefore, how to compensate for the wavefront phase distortion caused by turbulence in real time and accurately has become a key challenge to improve the performance of atmospheric laser communication. Traditional wavefront adaptive optical technologies rely on high-precision wavefront sensors and complex optical paths, with high hardware costs and difficult to be deployed on a large scale. Therefore, wavefront-free adaptive optical technologies have received attention.

[0003] Currently, traditional wavefront-free adaptive optical technologies include: using iterative optimization algorithms (such as stochastic parallel gradient descent) or neural network-based phase prediction models to directly recover wavefront distortion from light intensity information.

[0004] However, the wavefront damage phase information caused by atmospheric turbulence has spatial and temporal correlations. The temporal and spatial characteristics of atmospheric turbulence are physically coupled. Therefore, they cannot be separately and independently predicted during the turbulence prediction process. Wavefront phase prediction established through a certain independent feature or a prediction generator based on a neural network structure will result in time delay in turbulence compensation. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide an adaptive optical real-time compensation method, device, storage medium, and system to eliminate or improve one or more defects existing in the prior art. It can solve the problem of time delay in existing wavefront phase prediction and turbulence compensation methods.

[0006] One aspect of the present invention provides an adaptive optical real-time compensation method, which includes the following steps:

[0007] Extract historical wavefront phase diagrams corresponding to P historical time steps before the current time step from the currently received wavefront phase diagram, and each historical wavefront phase diagram corresponds to a historical time step vector; where P is an integer greater than 1;

[0008] Based on the light intensity attributes corresponding to each pixel node and the connection relationship between pixel nodes in the historical wavefront phase diagram, construct node features and an adjacency matrix to obtain the graph data of the historical wavefront phase diagram;

[0009] Input the graph data and the historical time step vector into the pre-trained graph multi-head attention model, and output the predicted wavefront phases for Q future time steps; the model includes a first fully-connected layer, an encoding layer, a transformed attention layer, a decoding layer, and a second fully-connected layer connected in series; the first fully-connected layer is used to convert the graph data into spatial feature embeddings, and fuse the time feature embeddings of the historical time step vector and the future time step vector to obtain spatio-temporal feature embeddings; the encoding layer is used to encode the spatio-temporal feature embeddings and output historical features; the transformed attention layer is used to map the historical features to future features corresponding to the future time steps; the decoding layer is used to decode the future features and output a preliminary prediction result; the second fully-connected layer is used to optimize the preliminary prediction result and output the predicted wavefront phases; where Q is an integer greater than 1;

[0010] Reconstruct the predicted wavefront phases into a phase distribution form suitable for phase modulator correction, and map the reconstructed predicted wavefront phases to the control voltage signals of the phase modulator and then input them into the phase modulator, so that the phase modulator adjusts the phase structure based on the control voltage signals to perform real-time phase compensation on the actual wavefront phases of Q future time steps.

[0011] In some embodiments of the present invention, the historical time step vector is a joint time vector obtained by concatenating a week encoding and a moment encoding; converting the graph data into spatial feature embeddings, and fusing the time feature embeddings of the historical time step vector and the future time step vector to obtain spatio-temporal feature embeddings includes:

[0012] Calculate the graph data through a biased second-order random walk algorithm and convert it into a spatial feature embedding matrix;

[0013] Map the joint time vector to a time feature embedding vector;

[0014] Align the spatial feature embedding matrix and the time feature embedding vector node by node and fuse them to obtain spatio-temporal feature embeddings.

[0015] In some embodiments of the present invention, the encoding layer and the decoding layer each include L self-attention modules connected in series; each self-attention module includes a spatial attention sub-module, a time attention sub-module, and a gated fusion sub-module; the spatial attention sub-module includes K parallel spatial attention heads, and the time attention sub-module includes K parallel time attention heads; where K is an integer greater than 1;

[0016] Each spatial attention head is used to take the weighted sum of nodes output by the spatial attention sub-module in the previous self-attention module as the hidden state of pixel nodes, connect each pixel node hidden state with the corresponding spatio-temporal feature embedding, and calculate the similarity between each pixel node and other pixel nodes through the scaled dot product algorithm; normalize the similarity to obtain the pixel node attention score corresponding to each pixel node; based on the pixel node attention score and the pixel node hidden state, calculate the weighted sum of nodes corresponding to each pixel node; merge the outputs of K spatial attention heads to obtain the output result of the spatial attention sub-module;

[0017] Each temporal attention head is used to take the weighted sum of nodes output by the temporal attention sub-module in the previous self-attention module as the hidden state of pixel nodes, connect each pixel node hidden state with the corresponding spatio-temporal feature embedding, and calculate the temporal step similarity between the temporal step corresponding to each pixel node and other temporal steps through the scaled dot product algorithm; normalize the temporal step similarity to obtain the temporal step attention score; based on the temporal step attention score and the pixel node step hidden state, calculate the weighted sum of nodes corresponding to each pixel node; merge the outputs of K temporal attention heads to obtain the output result of the temporal attention sub-module;

[0018] The gated fusion sub-module is used to output the fusion of the output results of the spatial attention sub-module and the temporal attention sub-module through the gated fusion mechanism.

[0019] In some embodiments of the present invention, the transform attention layer includes a first branch, a second branch and an output layer;

[0020] The first branch is used to perform a non-linear projection on the temporal feature embedding to obtain a non-linear projection result; perform an inner product operation on the non-linear projection result and then perform a normalization process to obtain the historical-future temporal step attention score, which is used as the output result of the first branch; the historical-future temporal step attention score is used to indicate the correlation between the future temporal step and the historical temporal step;

[0021] The second branch is used to perform a non-linear projection on the historical feature to obtain the output result of the second branch;

[0022] The output layer is used to perform an inner product operation on the output result of the first branch and the output result of the second branch, and perform a matrix concatenation on the result of the inner product operation to obtain the future feature.

[0023] In some embodiments of the present invention, the first branch sequentially includes a first non-linear projection layer, a first inner product operation layer and a normalization layer;

[0024] The second branch includes a second non-linear projection layer;

[0025] The output layer sequentially includes a second inner product operation layer and a matrix splicing layer.

[0026] In some embodiments of the present invention, the training process of the graph multi-head attention model includes:

[0027] Inputting the sample training graph data and the sample time series information into the initial graph multi-head attention model to output a prediction result; the initial graph multi-head attention model has the same model structure as the graph multi-head attention model; the sample training graph data and the sample time series information are extracted from the wavefront phase graph with time series correlation;

[0028] Comparing the prediction result with the real data corresponding to the sample training graph data and the sample time series information, and evaluating the prediction result through a preset loss function to obtain a loss result; the preset loss function includes the mean absolute error loss function, the root mean square error loss function, and the mean absolute percentage error loss function;

[0029] Based on the loss result, iteratively updating and optimizing the parameters of each layer in the initial graph multi-head attention model through the backpropagation algorithm and the gradient descent algorithm to obtain the graph multi-head attention model.

[0030] In some embodiments of the present invention, after reconstructing the predicted wavefront phase into a phase distribution form suitable for phase modulator correction and mapping the reconstructed predicted wavefront phase into the control voltage signal of the phase modulator and inputting it into the phase modulator, it further includes:

[0031] Comparing the compensated wavefront phase with the predicted wavefront phase to obtain an error result;

[0032] Generating a new control voltage signal based on the error result and sending it to the phase modulator so that the phase modulator adjusts the phase structure based on the new control voltage signal.

[0033] Another aspect of the present invention provides an electronic device, including a processor, a memory, and a computer program / instructions stored on the memory. The processor is used to execute the computer program / instructions. When the computer program / instructions are executed, the device implements the steps of the adaptive optical real-time compensation method as described above.

[0034] Another aspect of the present invention provides a computer-readable storage medium, on which a computer program / instructions are stored. When the computer program / instructions are executed by a processor, the steps of the adaptive optical real-time compensation method as described above are implemented.

[0035] Another aspect of the present invention provides an adaptive optical real-time compensation system, and the system includes:

[0036] A wavefront phase receiving module, configured to receive a wavefront phase map in real time and send the wavefront phase map to the historical phase extraction module;

[0037] A historical phase extraction module, which is used to extract historical wavefront phase diagrams corresponding to P historical time steps before the current time step from the currently received wavefront phase diagram, and each historical wavefront phase diagram corresponds to a historical time step vector; and send the historical wavefront phase diagrams to the graph data construction module;

[0038] A graph data construction module, which is used to construct node features and an adjacency matrix based on the light intensity attributes corresponding to each pixel node in the historical wavefront phase diagram and the connection relationship between pixel nodes, so as to obtain the graph data of the historical wavefront phase diagram; and send the graph data to the wavefront phase prediction module;

[0039] A wavefront phase prediction module, which is used to input the graph data and the historical time step vector into a pre-trained graph multi-head attention model, and output predicted wavefront phases for Q future time steps; and send the predicted wavefront phases to the voltage signal generation module; wherein, the model includes a first fully connected layer, an encoding layer, a transformed attention layer, a decoding layer, and a second fully connected layer connected in series; the first fully connected layer is used to convert the graph data into a spatial feature embedding, and fuse the time feature embedding of the historical time step vector and the future time step vector to obtain a spatio-temporal feature embedding; the encoding layer is used to encode the spatio-temporal feature embedding and output historical features; the transformed attention layer is used to map the historical features to future features corresponding to future time steps; the decoding layer is used to decode the future features and output a preliminary prediction result; the second fully connected layer is used to optimize the preliminary prediction result and output the predicted wavefront phases;

[0040] A voltage signal generation module, which is used to reconstruct the predicted wavefront phases into a phase distribution form suitable for correction by a phase modulator, and map the reconstructed predicted wavefront phases into a control voltage signal of the phase modulator and then input the control voltage signal into the phase modulator;

[0041] A phase modulator, which is used to adjust the phase structure based on the control voltage signal and perform real-time phase compensation on the actual wavefront phases for Q future time steps.

[0042] The adaptive optical real-time compensation method and device of the present invention can solve the problem that the existing wavefront phase prediction and turbulence compensation methods have time delay. By accurately predicting the spatio-temporal related turbulence model through the graph multi-head attention model, jointly estimating the wavefront phase data in the time dimension and the space dimension, and combining the hidden state of the historical wavefront phase with the spatio-temporal feature embedding, the real-time performance and accuracy of wavefront phase compensation are significantly improved.

[0043] Additional advantages, objects, and features of the present invention will be partly set forth in the description which follows, and will partly become obvious to those of ordinary skill in the art upon examination of the following, or may be learned by practice of the present invention. The objects and other advantages of the present invention may be realized and attained by the structure particularly pointed out in the specification and the drawings.

[0044] Those skilled in the art will understand that the objects and advantages that can be achieved by the present invention are not limited to those specifically described above, and the above and other objects that the present invention can achieve will be more clearly understood according to the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings described herein are for further understanding of the present invention, form a part of this application, and do not limit the present invention. In the drawings:

[0046] Figure 1 is a flowchart of an adaptive optical real-time compensation method provided for an embodiment of the present invention.

[0047] Figure 2 is a schematic diagram of model training provided for an embodiment of the present invention.

[0048] Figure 3 is a schematic diagram of a model structure provided for an embodiment of the present invention.

[0049] Figure 4 is a schematic diagram of the structure of a conversion attention layer provided for an embodiment of the present invention.

[0050] Figure 5 is a schematic diagram of the principle of an adaptive optical real-time compensation method provided for an embodiment of the present invention.

[0051] Figure 6 is a block diagram of an adaptive optical real-time compensation system provided for an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] To make the objects, technical solutions, and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with the embodiments and the drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but do not limit the present invention.

[0053] Here, it should also be noted that in order to avoid obscuring the present invention due to unnecessary details, only the structures and / or processing steps closely related to the solution of the present invention are shown in the drawings, and other details less related to the present invention are omitted.

[0054] It should be emphasized that when the term "comprising / including" is used herein, it refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0055] Here, it should also be noted that if not otherwise specified, the term "connection" in this text can not only refer to direct connection, but also represent indirect connection with intermediates.

[0056] In the following, embodiments of the present invention will be described with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0057] The adaptive optical real-time compensation method provided by this application will be introduced in detail below.

[0058] In some embodiments of the present invention, the execution subject of the adaptive optical real-time compensation method provided by this application is an electronic device, which can be a terminal such as a computer, a mobile phone, a tablet computer, a camera, etc., or can also be a server, a digital signal processor (DSP) or a field programmable gate array (FPGA), etc. This embodiment does not limit the implementation manner of the electronic device.

[0059] This embodiment provides an adaptive optical real-time compensation method, as Figure 1 shown, the method at least includes steps S101 to S104:

[0060] Step S101, extract the historical wavefront phase diagrams corresponding to the previous P historical time steps before the current time step from the currently received wavefront phase diagram, and each historical wavefront phase diagram corresponds to a historical time step vector. Wherein, P is an integer greater than 1.

[0061] In some embodiments of the present invention, the real-time wavefront phase diagram is received in real time from the atmospheric turbulence channel by a Charge-Coupled Device (CCD) in combination with a wavefront-free detection phase retrieval algorithm and then sent to the electronic device.

[0062] Among them, the wavefront-free detection phase retrieval algorithm includes but is not limited to the Fourier Transform Method (FTM), the Iterative Phase Retrieval Algorithm, or the Maximum Likelihood Method (MLM). This embodiment does not limit the implementation manner of the wavefront-free detection phase retrieval algorithm.

[0063] In actual implementation, the wavefront phase can also be received in real time through devices such as a fiber optic sensor, a high-resolution camera, or a micro-electro-mechanical systems scanning mirror. The implementation manner of receiving the wavefront phase is not limited in this embodiment.

[0064] In some embodiments of the present invention, the time intervals between adjacent wavefront phase diagrams are the same, and the historical wavefront phase diagrams corresponding to P historical time steps are a set of wavefront phase diagrams with temporal correlation. Each historical wavefront phase diagram corresponds to a historical time step vector. Among them, the historical time step vector is a joint time vector obtained by splicing a week code and a moment code.

[0065] For example: taking one day with T time steps as an example, the week and moment of each time step are respectively encoded as a week code and a moment code , then the joint time vector is expressed as .

[0066] Step S102: Based on the light intensity attributes corresponding to each pixel node in the historical wavefront phase diagram and the connection relationship between the pixel nodes, construct node features and an adjacency matrix to obtain the graph data of the historical wavefront phase diagram.

[0067] In some embodiments of the present invention, based on the connection relationship between the pixel points in the historical wavefront phase diagram, the historical wavefront phase diagram is represented as a graph G = (V, D, A). Among them, V is the set of pixel nodes, and N = |V| represents the total number of pixel nodes; D is the set of edges, representing the adjacent connections between pixel nodes; represents the adjacency matrix, including the connectivity between each pixel node.

[0068] At the same time, the light intensity attribute of each pixel node is extracted as the node feature of the pixel node and mapped into the graph signal in graph G. At this time, the graph data of the wavefront phase at time step is represented as the graph signal on graph G. Correspondingly, the graph data of the historical wavefront phase diagram can be represented as , where P represents the number of historical time steps, C represents the dimension of the pixel point features, and N represents the number of pixel points.

[0069] Step S103: Input the graph data and the historical time step vector into a pre-trained graph multi-head attention model, and output the predicted wavefront phases of Q future time steps. Among them, Q is an integer greater than 1.

[0070] In some embodiments of the present invention, based on the prediction ability of the pre-trained graph multi-head attention model for spatio-temporal sequences, the wavefront phase information for Q future time steps of the atmospheric turbulence channel is predicted. Then, the wavefront phase is reconstructed by adjusting the phase modulator, thereby achieving the purpose of real-time wavefront phase compensation.

[0071] Among them, the graph multi-head attention model includes a first fully connected layer, an encoding layer, a transformed attention layer, a decoding layer, and a second fully connected layer connected in series. The training process of the graph multi-head attention model at least includes steps S1031 to S1033:

[0072] Step S1031: Input the sample training graph data and the sample time series information into the initial graph multi-head attention model, and output a prediction result. Among them, the initial graph multi-head attention model has the same model structure as the graph multi-head attention model.

[0073] As Figure 2 shown, the sample training graph data and the sample time series information are extracted from the wavefront phase map with temporal correlation. Specifically, in some embodiments of the present invention, a charge-coupled device is combined with the wavefront sensing-free phase retrieval technique to obtain multiple sets of wavefront phase maps with temporal correlation within a continuous time period. 70% of the data in the multiple sets of wavefront phase maps is used for training, 10% of the data is used for verification, and 20% of the data is used for testing.

[0074] In addition, in actual implementation, if the number of pixel points in the original image is too large, the pixel data of the image can also be preprocessed, that is, reduced to a relatively small order of magnitude, such as 64×64. After reducing the spatial information of the multiple sets of wavefront phase maps, each phase image pixel is flattened into a one-dimensional vector, and all the image data is aggregated into a matrix. Set the time interval information between the phase maps, and extract the sample time series information.

[0075] To construct the adjacency matrix, each pixel point is regarded as a node, and the adjacent relationship is used as an edge to form a graph structure. Assume that each pixel point is only related to its adjacent points, the edge value is set to 1, and the edge value between other points is set to 0, thereby obtaining the adjacency matrix. At the same time, extract the light intensity information of each pixel point as the node feature and generate a spatial embedding matrix through a biased second-order random walk algorithm (p = 2, q = 1), and aggregate it into the sample training graph data.

[0076] Input the sample training graph data and the sample time series information into the initial graph multi-head attention model to construct spatio-temporal feature embeddings, and output a prediction result by the initial graph multi-head attention model.

[0077] Step S1032: Compare the prediction result with the real data corresponding to the sample training graph data and the sample time series information, evaluate the prediction result through a preset loss function, and obtain a loss result.

[0078] Among them, the preset loss function includes the mean absolute error loss function, the root mean square error loss function, or the mean absolute percentage error loss function.

[0079] Step S1033: Based on the loss result, use the backpropagation algorithm and the gradient descent algorithm to iteratively update and optimize the parameters of each layer in the initial graph multi-head attention model to obtain the graph multi-head attention model.

[0080] In some embodiments of the present invention, the model is trained by an Adaptive Moment Estimation (Adam) optimizer, and the initial learning rate is set to 0.001.

[0081] In the obtained graph multi-head attention model, the first fully connected layer is used to convert the graph data into spatial feature embeddings. At the same time, since the spatial feature embeddings only provide static representations and cannot represent the dynamic correlations of pixel nodes, based on this, as Figure 3 shown, the present invention further proposes temporal feature embeddings. The P historical time step vectors and the Q future time step vectors are converted into a vector through the first fully connected layer , and the P historical time steps and the Q future time steps are both embedded with temporal features to obtain temporal feature embeddings, denoted as , where , to provide the dynamic correlations of spatial pixel nodes. Then, the spatial feature embeddings and the temporal feature embeddings are fused as spatio-temporal feature embeddings to obtain the temporally correlated pixel node representations . Among them, P and Q are integers greater than 1, including 11, 12, and 13, etc. The values of P and Q can be adjusted according to the actual situation, and the present embodiment does not limit the values of P and Q; N = |V| represents the total number of pixel nodes; D is the edge set, representing the adjacency connections between pixel nodes.

[0082] Specifically, the first fully connected layer is used to convert the graph data into spatial feature embeddings and fuse the temporal feature embeddings of the historical time step vectors and the future time step vectors to obtain spatio-temporal feature embeddings. Among them, converting the graph data into spatial feature embeddings and fusing the temporal feature embeddings of the historical time step vectors and the future time step vectors to obtain spatio-temporal feature embeddings includes: calculating the graph data through the biased second-order random walk algorithm and converting it into a spatial feature embedding matrix; mapping the joint time vector into a temporal feature embedding vector; and aligning the spatial feature embedding matrix and the temporal feature embedding vector node by node and fusing them to obtain spatio-temporal feature embeddings.

[0083] When converting the graph data of the historical wavefront phase diagram and the historical time step vectors into spatio-temporal feature embeddings through the first fully connected layer After that, the spatio-temporal features are embedded into the input encoding layer for encoding. Specifically, the encoding layer is used to encode each pixel node data of the historical wavefront phase diagram as grid node information into the model, encode the spatio-temporal feature embedding, and output historical features.

[0084] In some embodiments of the present invention, as Figure 3 shown, the encoding layer includes L cascaded self-attention blocks. Among them, L is an integer greater than 1, including 3, 4 or 5. In actual implementation, the value of L can be adjusted according to actual needs. The present embodiment does not limit the value of L.

[0085] Each self-attention block includes a spatial attention sub-module, a temporal attention sub-module, and a gated fusion sub-module. The spatial attention sub-module is used to model the dynamic spatial correlation of the historical wavefront phase diagram; the temporal attention sub-module is used to model the non-linear temporal data correlation; the gated fusion sub-module is used to fuse the spatial and temporal representations.

[0086] Among the L cascaded self-attention blocks, the input of the th self-attention block is the spatio-temporal feature embedding , and the input of the 2nd to the Lth self-attention blocks is the output of the previous self-attention block.

[0087] Among the 2nd to the Lth self-attention blocks, taking the th self-attention block as an example, the input of the th self-attention block is denoted as , that is, the output of the -1th self-attention block. In the th self-attention block, the output of the spatial attention sub-module is denoted as , and the output of the temporal attention sub-module is denoted as . Among them, the hidden states of the vertices of the time step are respectively denoted as and , is the set of , is the set of .

[0088] After the output of the spatial attention sub-module and the output of the temporal attention sub-module, the output of the spatial attention sub-module and the output of the temporal attention sub-module are fused through the gated fusion sub-module to obtain the output of the th self-attention block.

[0089] Since the influence between different pixel nodes is highly dynamic, the present invention adaptively captures the correlation between different pixel nodes through the spatial attention sub-module. The key idea is to dynamically assign different weights to different pixel nodes at different time steps.

[0090] Take the th self-attention block as an example. In the th self-attention block, for any pixel node, a weighted sum is calculated from all pixel nodes as the hidden state of the pixel node. The set of hidden states of all pixel nodes is the output of the spatial attention sub-module .

[0091] Specifically, taking the pixel node at time step as an example, its weighted sum can be calculated by the following formula:

[0092]

[0093] In the formula, represents the hidden state of the pixel node at time step , that is, the weighted sum of the pixel node -1 in the output of the th self-attention block at time step ; represents the set of all pixel nodes; is the attention score of the pixel node to the pixel node . The sum of the attention scores is equal to 1, expressed as .

[0094] At any time step, since the current wavefront phase may affect the correlation between pixel nodes. Based on this, in this embodiment, the phase feature and the graph structure are considered to learn the attention score.

[0095] Specifically, the hidden state of the pixel node is connected to the corresponding spatio-temporal feature embedding, and the scaled dot product method is used to calculate the similarity between pixel nodes, expressed by the following formula:

[0096] In the formula, represents the hidden state of the pixel node at time step , that is, the weighted sum of the pixel node -1 in the output of the th self-attention block at time step ; represents the hidden state of the pixel node at time step The corresponding spatio-temporal feature embedding; Denote the time step The pixel nodes in The hidden state of, i.e., the time step output by the -1 self-attention blocks The pixel nodes in The weighted sum; Denote the time step The pixel nodes in The corresponding spatio-temporal feature embedding; Denote the concatenation operation; <•,•> denotes the inner product operator, and 2D means The corresponding dimension.

[0097] After that, through softmax normalization, the attention scores of the pixel nodes For the pixel nodes Can be expressed by the following formula:

[0098]

[0099] In the formula, , And Denote the pixel nodes; Denote the set of all pixel nodes; Denote the pixel node And the pixel node The similarity between; Denote the pixel node And the pixel node The similarity between.

[0100] After obtaining the attention scores of the pixel nodes After that, through the attention scores And , calculate to obtain And update the hidden state of the pixel nodes After calculating the hidden states of all pixel nodes, the output of the spatial attention sub-module of the th self-attention block is obtained .

[0101] In some embodiments of the present invention, in order to stabilize the learning process, the spatial attention sub-module is extended to a multi-head mechanism. Each spatial attention sub-module includes K parallel spatial attention heads, and different spatial attention heads have different learnable projections. Wherein, K is an integer greater than 1, including 8, 9 or 10, and the value of K can be adjusted according to actual needs during actual implementation. In this embodiment, the value of K is not limited.

[0102] ​Specifically, each spatial attention head is used to take the weighted sum of nodes output by the spatial attention sub-module in the previous self-attention module as the hidden state of the pixel node, connect each pixel node hidden state with the corresponding spatio-temporal feature embedding, and calculate the similarity between each pixel node and other pixel nodes through the scaled dot product algorithm; normalize the similarity to obtain the pixel node attention score corresponding to each pixel node; calculate the weighted sum of nodes corresponding to each pixel node based on the pixel node attention score and the pixel node hidden state; merge the outputs of K spatial attention heads to obtain the output result of the spatial attention sub-module.

[0103] Taking the -th spatial attention head as an example, connecting each pixel node hidden state with the corresponding spatio-temporal feature embedding, and calculating the similarity between each pixel node and other pixel nodes through the scaled dot product algorithm can be expressed by the following formula:

[0104]

[0105] In the formula, represents the hidden state of the pixel node at time step , that is, the weighted sum of the pixel node at time step output by the -1-th self-attention block; represents the corresponding spatio-temporal feature embedding of the pixel node at time step ; represents the hidden state of the pixel node at time step , that is, the weighted sum of the pixel node at time step output by the -1-th self-attention block; represents the corresponding spatio-temporal feature embedding of the pixel node at time step ; represents the connection operation; <•,•> represents the inner product operator, and d represents the corresponding dimension; represents the non-linear mapping of the -th spatial attention head, and the function expression is .

[0106] At the -th spatial attention head, normalize the similarity to obtain the node attention score corresponding to each node, which can be expressed by the following formula:

[0107]

[0108] In the formula, , and represent pixel nodes; represents the set of all pixel nodes; represents the th pixel node in the th spatial attention head and the similarity between the pixel node represents the th pixel node in the th spatial attention head and the similarity between the pixel node

[0109] The outputs of K spatial attention heads can be combined, which can be expressed by the following formula:

[0110]

[0111] In the formula, represents the non-linear mapping of the th spatial attention head, and the function expression is ; represents the hidden state of the pixel node at time step , that is, the weighted sum of the pixel node -1 output by the previous self-attention block at time step of the pixel node .

[0112] Correspondingly, each time attention head is used to take the weighted sum of the nodes output by the time attention sub-module in the previous self-attention module as the hidden state of the pixel node, connect each pixel node hidden state with the corresponding spatio-temporal feature embedding, and calculate the time step similarity between the time step corresponding to each pixel node and other time steps through the scaled dot product algorithm; normalize the time step similarity to obtain the time step attention score; based on the time step attention score and the pixel node step hidden state, calculate the weighted sum corresponding to each pixel node; combine the outputs of K time attention heads to obtain the output result of the time attention sub-module.

[0113] Taking the th spatial attention head as an example, connect each pixel node hidden state with the corresponding spatio-temporal feature embedding, and the similarity between the time step corresponding to each pixel node and other pixel nodes can be calculated through the scaled dot product algorithm, which can be expressed by the following formula:

[0114]

[0115] In the formula, Indicates the time step and the similarity between; Indicates the time step of the pixel nodes hidden state, i.e., the time step output by the -1 self-attention block; of the pixel nodes weighted sum; Indicates the time step of the pixel nodes corresponding spatio-temporal feature embedding; Indicates the time step of the pixel nodes hidden state, i.e., the time step output by the -1 self-attention block; of the pixel nodes weighted sum; Indicates the time step of the pixel nodes corresponding spatio-temporal feature embedding; Indicates the connection operation; <•,•> indicates the inner product operator, and d represents the corresponding dimension; and Indicates the non-linear mapping of the th spatial attention head.

[0116] At the th spatial attention head, the similarity is normalized to obtain the node attention score corresponding to each node, which can be expressed by the following formula:

[0117]

[0118] In the formula, , and indicate the time step; Indicates the set of all time steps; Indicates the th time attention head, the attention score between the time step and ; Indicates the th time attention head, the attention score between the time step and ;

[0119] The outputs of K time attention heads are combined, which can be expressed by the following formula:

[0120]

[0121] In the formula, represents the non-linear mapping of the -th spatial attention head; represents the hidden state of the pixel node at time step , that is, the hidden state of the pixel node output by the time attention sub-module of the -th self-attention block at time step ; represents the attention score between time step and time step .

[0122] After the output of the spatial attention sub-module in the -th self-attention block and the output of the time attention sub-module, the gated fusion sub-module is used to fuse the output results of the spatial attention sub-module and the time attention sub-module through a gated fusion mechanism and then output, to obtain the output of the -th self-attention block. The output of the final encoding layer is represented as .

[0123] To alleviate the error propagation effect between different prediction time steps in a long time range, in some embodiments of the present invention, a transform attention layer is added between the encoding layer and the decoding layer to simulate the direct relationship between each future time step and each historical time step, and transform the wavefront phase feature output by the encoding into the feature of the future time step as the input of the decoding layer. Specifically, the transform attention layer is used to model the direct association between the historical time step and the future time step, and map the historical feature to the future feature corresponding to the future time step.

[0124] In some embodiments of the present invention, the transform attention layer includes a first branch, a second branch, and an output layer.

[0125] Among them, the first branch is used to perform non-linear projection on the time feature embedding to obtain a non-linear projection result; after performing an inner product operation on the non-linear projection result and then performing a normalization process, a historical-future time step attention score is obtained as the output result of the first branch; the historical-future time step attention score is used to indicate the correlation between the future time step and the historical time step.

[0126] As Figure 4 shown, the first branch sequentially includes a first non-linear projection layer, a first inner product operation layer, and a normalization layer; the implementation of the first non-linear projection layer and the first inner product operation layer can be represented by the following formula:

[0127]

[0128] In the formula, represents that in the transform attention layer, for the th attention head corresponding to the future time step pixel node the spatio-temporal feature embedding is non-linearly projected through the non-linear projection function for non-linear projection; represents that in the transform attention layer, for the th attention head corresponding to the historical time step pixel node the spatio-temporal feature embedding is non-linearly projected through the non-linear projection function for non-linear projection; <•,•> represents the inner product operator, and d represents the corresponding dimension.

[0129] The implementation of the normalization layer can be represented by the following formula:

[0130]

[0131] In the formula, represents the future time step; , , and represent the historical time steps.

[0132] The second branch is used to non-linearly project the historical features to obtain the output result of the second branch; the output layer is used to perform an inner product operation on the output results of the first branch and the second branch, and concatenate the results of the inner product operation to obtain the future features. As Figure 4 shown, the second branch includes a second non-linear projection layer; the output layer sequentially includes a second inner product operation layer and a matrix concatenation layer. The implementation of the second branch and the output layer can be represented by the following formula:

[0133]

[0134] In the formula, represents the projection of the output historical features through the non-linear projection function ; represents the future time step; , , and represent the historical time steps; represents the th attention head; K represents the number of attention heads.

[0135] When the transform attention layer outputs the future features After that, the future features are used as the input of the decoding layer, which is used to decode the future features and output a preliminary prediction result. By stacking L spatio-temporal self-attention blocks on it, the output is used as the preliminary prediction result . Finally, the preliminary prediction result is input into the second fully-connected layer, which is used to optimize the preliminary prediction result and output the predicted wavefront phase, generating the predicted wavefront phase for Q future time steps .

[0136] Step S104: Reconstruct the predicted wavefront phase into a phase distribution form suitable for phase modulator correction, map the reconstructed predicted wavefront phase into the control voltage signal of the phase modulator and then input it into the phase modulator, so that the phase modulator adjusts the phase structure based on the control voltage signal to perform real-time phase compensation on the actual wavefront phase of Q future time steps.

[0137] Reference Figure 5 , taking the deformable mirror as an example of the phase modulator for illustration. After obtaining the predicted wavefront phase, input the predicted wavefront phase into the controller. Through the controller, under the condition of fusing the spatial correlation and temporal correlation of the wavefront phase, the reconstructed predicted wavefront phase is obtained, and the control voltage is generated based on the preset control algorithm and input into the deformable mirror. Among them, the controller includes devices such as a computer, a digital signal processor (DSP), or a field programmable gate array (FPGA), etc. The preset control algorithm includes linear control (such as matrix multiplication) or nonlinear control (such as optimization algorithm), etc. This embodiment does not limit the implementation manner of the controller and the implementation manner of the control algorithm. The deformable mirror works under the control of the control voltage output by the controller. By changing parameters such as the optical path and refractive index of wavefront transmission, the phase structure is adjusted to achieve phase compensation. The total delay that can be reduced by the compensation is , improving the real-time performance of phase compensation.

[0138] In addition, in order to improve the accuracy of phase compensation, in some embodiments of the present invention, after the phase modulator completes the compensation, the compensated wavefront phase is compared with the predicted wavefront phase, and feedback regulation is used to achieve precise phase compensation.

[0139] Specifically, after reconstructing the predicted wavefront phase into a phase distribution form suitable for phase modulator correction, mapping the reconstructed predicted wavefront phase into the control voltage signal of the phase modulator and then inputting it into the phase modulator, it further includes: comparing the compensated wavefront phase with the predicted wavefront phase to obtain an error result; generating a new control voltage signal based on the error result and sending it to the phase modulator, so that the phase modulator adjusts the phase structure based on the new control voltage signal.

[0140] For example: Reference Figure 5, re - input the CCD sensor, combine with the wavefront - sensing - free phase retrieval technology, send the phase information back to the controller, generate a new control voltage signal and send it to the deformable mirror, so that the deformable mirror adjusts the phase structure based on the new control voltage signal.

[0141] In summary, the adaptive optics real - time compensation method provided by the present invention extracts the historical wavefront phase diagrams corresponding to P historical time steps before the current time step from the currently received wavefront phase diagram, and each historical wavefront phase diagram corresponds to a historical time - step vector; constructs node features and an adjacency matrix based on the light - intensity attributes corresponding to each pixel node in the historical wavefront phase diagram and the connection relationship between pixel nodes to obtain the graph data of the historical wavefront phase diagram; inputs the graph data and the historical time - step vector into a pre - trained graph multi - head attention model to output the predicted wavefront phases for Q future time steps; reconstructs the predicted wavefront phases into a phase distribution form suitable for phase modulator correction, maps the reconstructed predicted wavefront phases into the control voltage signals of the phase modulator and inputs them into the phase modulator, so that the phase modulator adjusts the phase structure based on the control voltage signals to perform real - time phase compensation on the actual wavefront phases for Q future time steps; can solve the problem of latency in the existing wavefront phase prediction and turbulence compensation methods, accurately predict the spatio - temporal - related turbulence model through the graph multi - head attention model, jointly estimate the wavefront phase data in the time dimension and the space dimension, and combine the hidden state of the historical wavefront phase with spatio - temporal features to significantly improve the real - time performance and accuracy of wavefront phase compensation.

[0142] Figure 6 FIG. 7 is a block diagram of an adaptive optics real - time compensation system provided by an embodiment of the present application. The system at least includes the following modules: a wavefront phase receiving module 610, a historical phase extraction module 620, a graph data construction module 630, a wavefront phase prediction module 640, a voltage signal generation module 650, and a phase modulator 660.

[0143] The wavefront phase receiving module 610 is configured to receive the wavefront phase diagram in real - time and send the wavefront phase diagram to the historical phase extraction module 620;

[0144] The historical phase extraction module 620 is configured to extract the historical wavefront phase diagrams corresponding to P historical time steps before the current time step from the currently received wavefront phase diagram, and each historical wavefront phase diagram corresponds to a historical time - step vector; send the historical wavefront phase diagrams to the graph data construction module 630;

[0145] The graph data construction module 630 is configured to construct node features and an adjacency matrix based on the light - intensity attributes corresponding to each pixel node in the historical wavefront phase diagram and the connection relationship between pixel nodes to obtain the graph data of the historical wavefront phase diagram; send the graph data to the wavefront phase prediction module 640;

[0146] The wavefront phase prediction module 640 is configured to input the graph data and the historical time step vector into a pre-trained graph multi-head attention model, and output the predicted wavefront phase for Q future time steps; and send the predicted wavefront phase to the voltage signal generation module 650; wherein, the model includes a first fully-connected layer, an encoding layer, a transformed attention layer, a decoding layer, and a second fully-connected layer connected in series; the first fully-connected layer is configured to convert the graph data into a spatial feature embedding, and fuse the temporal feature embedding of the historical time step vector and the future time step vector to obtain a spatio-temporal feature embedding; the encoding layer is configured to encode the spatio-temporal feature embedding and output historical features; the transformed attention layer is configured to map the historical features to future features corresponding to the future time steps; the decoding layer is configured to decode the future features and output a preliminary prediction result; the second fully-connected layer is configured to optimize the preliminary prediction result and output the predicted wavefront phase.

[0147] The voltage signal generation module 650 is configured to reconstruct the predicted wavefront phase into a phase distribution form suitable for correction by the phase modulator 660, and map the reconstructed predicted wavefront phase into a control voltage signal of the phase modulator 660 and then input it into the phase modulator 660.

[0148] The phase modulator 660 is configured to adjust the phase structure based on the control voltage signal and perform real-time phase compensation on the actual wavefront phase for Q future time steps.

[0149] For related details, refer to the above embodiments.

[0150] It should be noted that: when the adaptive optical real-time compensation system provided in the above embodiments performs adaptive optical real-time compensation, only the division of the above functional modules is used as an example for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs to complete all or part of the functions described above. In addition, the adaptive optical real-time compensation system provided in the above embodiments and the embodiments of the adaptive optical real-time compensation method belong to the same concept. For the specific implementation process, refer to the method embodiments and will not be elaborated here.

[0151] Correspondingly to the above method, another aspect of the present invention provides an electronic device, including a processor, a memory, and a computer program / instruction stored in the memory. The processor is configured to execute the computer program / instruction, and when the computer program / instruction is executed, the device implements the steps of the adaptive optical real-time compensation method as described above.

[0152] Another aspect of the present invention provides a computer-readable storage medium, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the adaptive optical real-time compensation method as described above are implemented.

[0153] Those of ordinary skill in the art should understand that the various exemplary components, systems, and methods described in connection with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Specifically, whether to implement in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, and so on. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium or transmitted via a data signal carried in a carrier wave on a transmission medium or a communication link.

[0154] It should be clear that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order between steps after understanding the spirit of the present invention.

[0155] In the present invention, the features described and / or illustrated for one embodiment can be used in the same or a similar manner in one or more other embodiments, and / or combined with the features of other embodiments or replace the features of other embodiments.

[0156] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, various changes and variations can be made to the embodiments of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive optical real-time compensation method, characterized in that: The method comprises the following steps: Extracting the historical wavefront phase diagram corresponding to P historical time steps before the current time step from the currently received wavefront phase diagram, each historical wavefront phase diagram corresponds to a historical time step vector; wherein P is an integer greater than 1; Based on the light intensity attribute corresponding to each pixel node in the historical wavefront phase map and the connection relationship between the pixel nodes, a node feature and an adjacency matrix are constructed to obtain the graph data of the historical wavefront phase map; The graph data and the historical time step vector are input into the pre-trained graph multi-head attention model, and the predicted wavefront phases of Q future time steps are output; the model includes a first fully connected layer, an encoding layer, a conversion attention layer, a decoding layer, and a second fully connected layer connected in series; the first fully connected layer is used to convert the graph data into a spatial feature embedding, and fuse the time feature embeddings of the historical time step vector and the future time step vector to obtain a spatiotemporal feature embedding; the encoding layer is used to encode the spatiotemporal feature embedding and output historical features; the conversion attention layer is used to map the historical features to future features corresponding to the future time step; the decoding layer is used to decode the future features and output a preliminary prediction result; the second fully connected layer is used to optimize the preliminary prediction result and output the predicted wavefront phase; wherein Q is an integer greater than 1; The predicted wavefront phase is reconstructed into a phase distribution form suitable for phase modulator correction, and the reconstructed predicted wavefront phase is mapped to a control voltage signal of the phase modulator and then input into the phase modulator, so that the phase modulator adjusts the phase structure based on the control voltage signal and performs real-time phase compensation on the actual wavefront phases of the Q future time steps.

2. The adaptive optical real-time compensation method according to claim 1, characterized in that: The historical time step vector is a joint time vector obtained by concatenating the week code and the time code; the graph data is converted into a spatial feature embedding, and the time feature embedding of the historical time step vector and the future time step vector is fused to obtain a spatiotemporal feature embedding, including: The graph data is calculated by a biased second-order random walk algorithm and converted into a spatial feature embedding matrix; Mapping the joint time vector into a time feature embedding vector; The spatial feature embedding matrix and the temporal feature embedding vector are aligned node by node and fused to obtain the spatiotemporal feature embedding.

3. The adaptive optical real-time compensation method according to claim 1, characterized in that: The encoding layer and the decoding layer respectively include L serially connected self-attention modules; each self-attention module includes a spatial attention submodule, a temporal attention submodule and a gated fusion submodule; the spatial attention submodule includes K parallel spatial attention heads, and the temporal attention submodule includes K parallel temporal attention heads; wherein K is an integer greater than 1; Each spatial attention head is used to take the node weighted sum output by the spatial attention submodule in the previous self-attention module as the hidden state of the pixel node, embed the connection of each pixel node hidden state with the corresponding spatiotemporal feature, and calculate the similarity between each pixel node and other pixel nodes through the scaled dot product algorithm; normalize the similarity to obtain the pixel node attention score corresponding to each pixel node; calculate the node weighted sum corresponding to each pixel node based on the pixel node attention score and the pixel node hidden state; merge the outputs of K spatial attention heads to obtain the output result of the spatial attention submodule; Each temporal attention head is used to take the node weighted sum output by the temporal attention submodule in the previous self-attention module as the hidden state of the pixel node, embed the connection of each pixel node hidden state with the corresponding spatiotemporal feature, and calculate the time step similarity between the time step corresponding to each pixel node and other time steps through the scaled dot product algorithm; normalize the time step similarity to obtain the time step attention score; calculate the node weighted sum corresponding to each pixel node based on the time step attention score and the pixel node hidden state; merge the outputs of K temporal attention heads to obtain the output result of the temporal attention submodule; The gated fusion submodule is used to fuse the output result of the spatial attention submodule and the output result of the temporal attention submodule through a gated fusion mechanism and output them.

4. The adaptive optical real-time compensation method according to claim 1, characterized in that: The conversion attention layer includes a first branch, a second branch and an output layer; The first branch is used to perform nonlinear projection on the temporal feature embedding to obtain a nonlinear projection result; the nonlinear projection result is subjected to an inner product operation and then normalized to obtain a history-future time step attention score as an output result of the first branch; the history-future time step attention score is used to indicate the correlation between the future time step and the history time step; The second branch is used to perform nonlinear projection on the historical features to obtain an output result of the second branch; The output layer is used to perform an inner product operation on the output result of the first branch and the output result of the second branch, and to perform matrix concatenation on the results of the inner product operation to obtain the future features.

5. The adaptive optical real-time compensation method according to claim 4, characterized in that: The first branch includes a first nonlinear projection layer, a first inner product operation layer and a normalization layer in sequence; The second branch includes a second nonlinear projection layer; The output layer includes a second inner product operation layer and a matrix concatenation layer in sequence.

6. The adaptive optical real-time compensation method according to claim 1, characterized in that: The training process of the graph multi-head attention model includes: Input the sample training graph data and the sample time series information into the initial graph multi-head attention model, and output the prediction result; the initial graph multi-head attention model has the same model structure as the graph multi-head attention model; the sample training graph data and the sample time series information are extracted from a wavefront phase map with time series correlation; Compare the prediction result with the sample training graph data and the real data corresponding to the sample time series information, and evaluate the prediction result by a preset loss function to obtain a loss result; the preset loss function includes a mean absolute error loss function, a root mean square error loss function, or a mean absolute percentage error loss function; Based on the loss result, the parameters of each layer in the initial graph multi-head attention model are iteratively updated and optimized through the back propagation algorithm and the gradient descent algorithm to obtain the graph multi-head attention model.

7. The adaptive optical real-time compensation method according to claim 1, characterized in that: After reconstructing the predicted wavefront phase into a phase distribution form suitable for phase modulator correction, and mapping the reconstructed predicted wavefront phase into a control voltage signal of the phase modulator and inputting it into the phase modulator, the method further includes: Comparing the compensated wavefront phase with the predicted wavefront phase to obtain an error result; A new control voltage signal is generated based on the error result and sent to the phase modulator, so that the phase modulator adjusts the phase structure based on the new control voltage signal.

8. An electronic device comprising a processor, a memory and a computer program / instruction stored in the memory, characterized in that: The processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the device implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method as claimed in any one of claims 1 to 7 are implemented.

10. An adaptive optical real-time compensation system, characterized in that: The system comprises: A wavefront phase receiving module, used for receiving a wavefront phase image in real time and sending the wavefront phase image to a historical phase extraction module; The historical phase extraction module is used to extract the historical wavefront phase diagram corresponding to P historical time steps before the current time step from the currently received wavefront phase diagram, each historical wavefront phase diagram corresponds to a historical time step vector; and send the historical wavefront phase diagram to the graph data construction module; The graph data construction module is used to construct node features and an adjacency matrix based on the light intensity attribute corresponding to each pixel node in the historical wavefront phase graph and the connection relationship between the pixel nodes to obtain the graph data of the historical wavefront phase graph; and send the graph data to the wavefront phase prediction module; The wavefront phase prediction module is used to input the graph data and the historical time step vector into the pre-trained graph multi-head attention model, and output the predicted wavefront phase of Q future time steps; the predicted wavefront phase is sent to the voltage signal generation module; wherein the model includes a first fully connected layer, an encoding layer, a conversion attention layer, a decoding layer, and a second fully connected layer connected in series; the first fully connected layer is used to convert the graph data into a spatial feature embedding, and fuse the time feature embedding of the historical time step vector and the future time step vector to obtain a spatiotemporal feature embedding; the encoding layer is used to encode the spatiotemporal feature embedding and output the historical feature; the conversion attention layer is used to map the historical feature to the future feature corresponding to the future time step; the decoding layer is used to decode the future feature and output a preliminary prediction result; the second fully connected layer is used to optimize the preliminary prediction result and output the predicted wavefront phase; The voltage signal generating module is used to reconstruct the predicted wavefront phase into a phase distribution form suitable for phase modulator correction, and map the reconstructed predicted wavefront phase into a control voltage signal of the phase modulator and then input it into the phase modulator; The phase modulator is used to adjust the phase structure based on the control voltage signal to perform real-time phase compensation on the actual wavefront phases of the Q future time steps.

Images