A method for constructing a multi-modal deep learning model for super surface electromagnetic property estimation, medium, device and application

Abstract

This invention belongs to the field of electronic information technology and discloses a method, medium, device, and application for constructing a multimodal deep learning model for estimating the electromagnetic properties of metasurfaces. The method involves binarizing the metasurface pattern corresponding to each structural parameter vector in the metasurface parameter space; constructing a multimodal deep learning model (MMN) and initializing trainable parameters; calculating the accurate electromagnetic response corresponding to the structural parameter vector using the finite element method, treating the structural parameter vector and the binarized matrix as inputs and the electromagnetic response as the output, and pairing them to form a training sample set; and training the MMN model using gradient descent based on the loss function and the training sample set. The MMN model obtained by this invention can accurately estimate the electromagnetic properties of metasurfaces, and the required training sample set size is only 39% of that required by a single-modal model.

Description

Technical Field

[0001] This invention belongs to the field of electronic information technology, and in particular relates to a method, medium, device and application for constructing a multimodal deep learning model for estimating the electromagnetic properties of metasurfaces. Background Technology

[0002] Currently, estimating the electromagnetic properties of metasurfaces using deep learning is a widely studied problem, encompassing fields such as wireless communication and electromagnetic simulation. A major research trend in estimating the electromagnetic properties of metasurfaces using deep learning is to improve the accuracy and efficiency of the estimation process. However, the use of multimodal deep learning models has been less explored in this regard.

[0003] Generally, the accuracy and efficiency of estimating the electromagnetic properties of metasurfaces using deep learning depend on two aspects: first, the size of the training sample set required to build and train the deep learning model; and second, the ability of the deep learning model itself to extract features from the training sample set. Larger scale and stronger feature extraction capabilities result in higher accuracy and efficiency, and vice versa. After obtaining the training sample set, existing methods extract features from the data in the training sample set by constructing a single-modality deep learning model. For example, fully connected layers and corresponding activation and normalization layers are used to extract features from the metasurface parameter vector. In this case, the deep learning model only utilizes data from the metasurface parameter vector modality. Alternatively, convolutional layers and corresponding normalization layers are used to extract features from the metasurface pattern shape. In this case, the deep learning model only utilizes data from the metasurface pattern shape modality.

[0004] Currently, various methods exist for optimizing unimodal deep learning models to improve their ability to extract data features. These include optimizing the model's structure by mathematically optimizing the optimal number of network layers and neurons in each layer, or optimizing the training method by adjusting the loss function used during training. However, in practical applications, optimizing the structure of unimodal deep learning models is often counterproductive, as larger models generally have stronger feature extraction capabilities. Therefore, for a given training dataset, simply increasing the model size usually yields better electromagnetic response estimation results. However, such methods are typically limited by hardware computing power; excessively large model sizes lead to a dramatic increase in computational load during training.

[0005] For a given metasurface structure, either the structural parameter vector or the pattern shape can accurately characterize the structure. Therefore, the structural parameter vector and the pattern shape represent different modalities of the same metasurface structure. Clearly, the structural parameter vector and the pattern shape (usually represented in matrix form) have drastically different probability distributions. Therefore, a multimodal deep learning model can be constructed to simultaneously extract features of the metasurface structure from both the structural parameter vector and the pattern shape, thereby improving the accuracy and efficiency of the electromagnetic property estimation process.

[0006] Based on the above analysis, the existing technology has the following problems and shortcomings: Single-modal deep learning models can only utilize a single modality, such as the structural parameters of a metasurface structure in vector form or the pattern shape of a metasurface structure in matrix form. Using only a single modality does not fully exploit the feature extraction capabilities of the deep learning model, reducing the accuracy and efficiency of the deep learning model in the electromagnetic property estimation process.

[0007] The difficulty in solving the above problems and defects lies in the fact that, in order to utilize both vector-based structural parameter data and matrix-based pattern shape data simultaneously, a deep learning model that can simultaneously input both vector-based and matrix-based data needs to be constructed and trained. This model needs to be able to extract features from both the structural parameter vectors and the pattern shapes and fuse these features. The fused features will then be used to estimate the electromagnetic response of the metasurface.

[0008] The significance of solving the above problems and defects is that the multimodal deep learning model constructed in this invention utilizes two different modes of the metasurface, which can improve the accuracy of the metasurface electromagnetic response estimation without changing the size of the training sample set; or, in order to achieve specific electromagnetic response estimation performance, it can reduce the requirement for the size of the training sample set, thereby improving the estimation efficiency. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention provides a method, medium, device, and application for constructing a multimodal deep learning model for estimating the electromagnetic properties of metasurfaces.

[0010] This invention is implemented as follows: a method for constructing a multimodal deep learning model for estimating the electromagnetic properties of metasurfaces, the method comprising:

[0011] Initialize the structure parameter vector of the metasurface in the parameter space;

[0012] Binarize the metasurface pattern corresponding to each structural parameter vector;

[0013] Construct a multimodal deep learning model MMN and initialize trainable parameters;

[0014] The finite element method is used to calculate the accurate electromagnetic response corresponding to the structural parameter vector. The structural parameter vector and the binarized matrix are regarded as inputs, and the electromagnetic response is regarded as output. They are paired to form a training sample set.

[0015] The MMN model is trained using gradient descent based on the loss function and the training sample set.

[0016] Furthermore, the method for constructing a multimodal deep learning model for estimating the electromagnetic properties of metasurfaces specifically includes the following steps:

[0017] The first step is to sample a metasurface structure of scale N in a given parameter space using a uniform sampling method. The parameter vector corresponding to the metasurface structure can be represented by a set X = {x1, x2, ..., x...}. N}express;

[0018] The second step is to use 3D modeling software to model the metasurface and extract its surface pattern for any parameter vector x in the set X.

[0019] The third step is to binarize the metasurface pattern, where the areas covered by metal or resistive film are binarized as 1, and other parts are binarized as 0. The binarization matrix corresponding to each metasurface structure in the parameter space can be represented by a set P = {p1, p2, ..., p...} N}express;

[0020] The fourth step involves constructing an MMN model using fully connected layers, one-dimensional batch normalization layers, convolutional layers, two-dimensional batch normalization layers, activation layers, and regularization layers. The MMN model has two input nodes and one output node, and the set of parameters of the neurons in the model is W.

[0021] The fifth step is to use the finite element method to calculate the electromagnetic property vector of the metasurface corresponding to each vector in set X, thus obtaining the set Y = {y1, y2, ..., y...}. N};

[0022] Step 6: Taking the set of structural parameter vectors X and the corresponding set of binary matrices P as input, and the set of electromagnetic properties Y as output, construct the training sample set Z = {z1, z2, ..., z...} N}={(X,P,Y)}={(x1,p1,y1),(x2,p2,y2),…,(x N ,p N ,y N )};

[0023] Step 7: In turn, select x with index i from the sample set Z. i and p i As input to the MMN, the MMN model is used to perform forward propagation on the samples. For any x...i and p i obtain the floating - point value vector y generated by the output - node neuron i ';

[0024] In the eighth step, according to the floating - point value vector y i ' and y in the original training sample set i , use the mean - square error loss function to calculate the loss function value L of the model MMN i ;

[0025] In the ninth step, according to the loss function value L i back - calculate the gradient G corresponding to the trainable parameters in the model MMN i , and according to the gradient G i use the gradient - descent method to update the parameters W of the MMN model;

[0026] In the tenth step, repeat steps seven to nine to complete the training of the MMN model.

[0027] Furthermore, the construction method of the MMN model in the multi - modal deep - learning model construction method for estimating the electromagnetic characteristics of the metasurface includes:

[0028] Step one, construct M fully - connected layers FC1, FC2, …, FC M , M - 1 one - dimensional batch - normalization layers BN1D1, BN1D2, …, BN1D M-1 , M - 1 activation layers RE1, RE2, …, RE M-1 , T convolutional layers CONV1, CONV2, …, CONV T , T two - dimensional batch - normalization layers BN2D1, BN2D2, …, BN2D T and an output layer OC;

[0029] Step two, set FC1 as the first input node of MMN, and FC1 is responsible for receiving x in the training sample; set CONV1 as the second input node of MMN, and CONV1 is responsible for receiving p in the training sample;

[0030] Step three, connect the first Q (Q < M) fully - connected layers, the first Q one - dimensional batch - normalization layers, and the first Q activation layers in the order of {FC1, BN1D1, RE1, FC2, …, FC Q , BN1D Q , RE Q}, and the output of the previous - layer network is used as the input of the next - layer network;

[0031] Step four, connect the T convolutional layers and the T two - dimensional batch - normalization layers in the order of {CONV1, BN2D1, CONV2, …, CONV TBN2D T The layers are connected in sequence, with the output of the previous layer serving as the input of the next layer.

[0032] Step 5, RE Q The output vector and BN2D T The output vectors are added together to obtain vector I;

[0033] Step six, sort the remaining fully connected layers, one-dimensional batch normalized layers, activation layers, and output layers according to {FC} Q+1 BN1D Q+1 RE Q+1 ,…,FC M The layers OC are connected sequentially, with the output of the previous layer serving as the input of the next layer. FC Q+1 The input to the layer is vector I; the output of the neurons in the OC layer is the output of the MMN model.

[0034] Another object of the present invention is to provide a computer device comprising a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the following steps:

[0035] Initialize the structure parameter vector of the metasurface in the parameter space;

[0036] Binarize the metasurface pattern corresponding to each structural parameter vector;

[0037] Construct a multimodal deep learning model MMN and initialize trainable parameters;

[0038] The finite element method is used to calculate the accurate electromagnetic response corresponding to the structural parameter vector. The structural parameter vector and the binarized matrix are regarded as inputs, and the electromagnetic response is regarded as output. They are paired to form a training sample set.

[0039] The MMN model is trained using gradient descent based on the loss function and the training sample set.

[0040] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0041] Initialize the structure parameter vector of the metasurface in the parameter space;

[0042] Binarize the metasurface pattern corresponding to each structural parameter vector;

[0043] Construct a multimodal deep learning model MMN and initialize trainable parameters;

[0044] The finite element method is used to calculate the accurate electromagnetic response corresponding to the structural parameter vector. The structural parameter vector and the binarized matrix are regarded as inputs, and the electromagnetic response is regarded as output. They are paired to form a training sample set.

[0045] The MMN model is trained using gradient descent based on the loss function and the training sample set.

[0046] Another objective of this invention is to provide an information data processing terminal, which is used to implement the multimodal deep learning model construction method for estimating the electromagnetic properties of metasurfaces. The information data processing terminal includes: a wireless communication terminal and an electromagnetic simulation terminal.

[0047] The multimodal deep learning model construction method was tested using a dual-layer metasurface absorber-sense integrated structure. Compared to a single-modal deep learning model trained with the complete training sample set, the multimodal deep learning model can achieve the same estimation accuracy using only 39% of the training samples. The size of its training sample set and the mean squared error performance of the corresponding single-modal or multimodal deep learning models are shown in the attached figure. Figure 3 As shown in the figure, the dashed line represents the mean squared error performance of the single-modal deep learning model, and the solid line represents the mean squared error performance of the multimodal deep learning model. The lower the mean squared error value, the better the model's performance. Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart of a method for constructing a multimodal deep learning model for estimating the electromagnetic properties of metasurfaces, provided in an embodiment of the present invention.

[0050] Figure 2 This is a flowchart of the MMN model construction method provided in the embodiments of the present invention;

[0051] Figure 3 This is a mean squared error performance curve of the single-modal and multi-modal deep learning models provided in this embodiment of the invention when trained using the same percentage of training sample set data;

[0052] Figure 4 This is a bar chart showing the size of the training sample set required for a single-modal or multi-modal model to achieve the desired mean squared error performance, as provided in the embodiments of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0054] To address the problems existing in the prior art, this invention provides a method, medium, device, and application for constructing a multimodal deep learning model for estimating the electromagnetic properties of metasurfaces. The invention will be described in detail below with reference to the accompanying drawings.

[0055] like Figure 1 As shown, the multimodal deep learning model construction method for estimating the electromagnetic properties of metasurfaces provided by this invention includes the following steps:

[0056] S101: Initialize the structure parameter vector of the metasurface in the parameter space;

[0057] S102: Binarize the metasurface pattern corresponding to each structural parameter vector;

[0058] S103: Construct the multimodal deep learning model MMN and initialize trainable parameters;

[0059] S104: The finite element method is used to calculate the accurate electromagnetic response corresponding to the structural parameter vector, and the structural parameter vector and the binarized matrix are regarded as inputs, and the electromagnetic response is regarded as outputs, and the pairing is used to form a training sample set;

[0060] S105: Train the MMN model using gradient descent based on the loss function and the training sample set.

[0061] The multimodal deep learning model construction method for estimating the electromagnetic properties of metasurfaces provided in this invention can also be implemented by those skilled in the art using other steps. Figure 1 The multimodal deep learning model construction method for estimating the electromagnetic properties of metasurfaces provided by this invention is merely a specific embodiment.

[0062] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0063] The multimodal deep learning model construction method for estimating the electromagnetic properties of metasurfaces provided by this invention specifically includes the following steps:

[0064] The first step is to sample a metasurface structure of scale N in a given parameter space using a uniform sampling method. The parameter vector corresponding to the metasurface structure can be represented by a set X = {x1, x2, ..., x...}. N}express;

[0065] The second step is to use 3D modeling software to model the metasurface and extract its surface pattern for any parameter vector x in the set X.

[0066] The third step is to binarize the metasurface pattern, where the areas covered by metal or resistive film are binarized as 1, and other parts are binarized as 0. The binarization matrix corresponding to each metasurface structure in the parameter space can be represented by a set P = {p1, p2, ..., p...} N}express;

[0067] The fourth step involves constructing an MMN model using fully connected layers, one-dimensional batch normalization layers, convolutional layers, two-dimensional batch normalization layers, activation layers, and regularization layers. The MMN model has two input nodes and one output node, and the set of parameters of the neurons in the model is W.

[0068] The fifth step is to use the finite element method to calculate the electromagnetic property vector of the metasurface corresponding to each vector in set X, thus obtaining the set Y = {y1, y2, ..., y...}. N};

[0069] Step 6: Taking the set of structural parameter vectors X and the corresponding set of binary matrices P as input, and the set of electromagnetic properties Y as output, construct the training sample set Z = {z1, z2, ..., z...} N}={(X,P,Y)}={(x1,p1,y1),(x2,p2,y2),…,(x N ,p N ,y N )};

[0070] Step 7: In turn, select x with index i from the sample set Z. i and p i As input to the MMN, the MMN model is used to perform forward propagation on the samples. For any x... i and p i Obtain the floating-point numerical vector y generated by the output node neuron. i ′;

[0071] Step 8: Based on the floating-point numerical vector y i ′ and y in the original training sample set i The loss function value L of the MMN model is calculated using the mean squared error loss function. i ;

[0072] Step 9: Based on the loss function value L i The gradient G corresponding to the trainable parameters in the reverse computation model MMN is calculated. i And according to the gradient G i Update the parameters W of the MMN model using gradient descent.

[0073] Step 10: Repeat steps 7 through 9 to complete the training of the MMN model.

[0074] like Figure 2 As shown, the MMN model construction method provided in this embodiment of the invention includes:

[0075] Step 1: Construct M fully connected layers FC1, FC2, ..., FC M M-1 one-dimensional batch normalized layers BN1D1, BN1D2, ..., BN1D M-1 M-1 activation layers RE1, RE2, ..., RE M-1 T convolutional layers CONV1, CONV2, ..., CONV1, and T two-dimensional batch normalized layers BN2D1, BN2D2, ..., BN2D T And an output layer OC;

[0076] Step 2: Set FC1 as the first input node of MMN, and FC1 is responsible for receiving x from the training samples; set CONV1 as the second input node of MMN, and CONV1 is responsible for receiving p from the training samples.

[0077] Step 3: Assign the first Q fully connected layers, the first Q one-dimensional batch normalized layers, and the first Q activation layers according to {FC1, BN1D1, RE1, FC2, ..., FC...} Q BN1D Q RE Q The layers are connected in sequence, with the output of the previous layer serving as the input of the next layer.

[0078] Step 4: Divide the T convolutional layers and T two-dimensional batch normalization layers into {Conv1, BN2D1, CONV2, ..., CONV}. T BN2D T The layers are connected in sequence, with the output of the previous layer serving as the input of the next layer.

[0079] Step 5, RE Q The output vector and BN2D T The output vectors are added together to obtain vector I;

[0080] Step six, sort the remaining fully connected layers, one-dimensional batch normalized layers, activation layers, and output layers according to {FC} Q+1 BN1D Q+1 RE Q+1 ,…,FC M The layers OC are connected sequentially, with the output of the previous layer serving as the input of the next layer. FC Q+1 The input to the layer is vector I; the output of the neurons in the OC layer is the output of the MMN model.

[0081] The technical effects of the present invention will be described in detail below with reference to specific embodiments.

[0082] Example 1:

[0083] Taking a two-layer metasurface integrated absorber-reflector structure as an example, the bottom and top surfaces of this structure are both made of copper, with different metal covering patterns on the bottom and top surfaces. A dielectric layer with a relative permittivity of 1.1 and a relative permeability of 1 is filled between the metal layers on the bottom and top surfaces. A dataset of 2500 samples is constructed using a random sampling method, with 2000 samples used as the training set and 500 samples as the test set. Two modal characterizations are performed on all metasurfaces in the dataset, including vectorized structural parameters and binary patterns in matrix form. A multimodal model is constructed using the aforementioned multimodal deep learning model construction method for estimating the electromagnetic properties of metasurfaces, containing a total of 7 fully connected layers and 10 convolutional layers. For the fully connected layers, the first 5 are used to extract features from the structural parameter vectors, and the last 2 are used to further extract features from the fused features. For the convolutional layers, the metal patterns on the bottom and top surfaces are extracted using 10 convolutional layers. The unimodal model contains seven fully connected layers, and its connection structure is consistent with that of the fully connected layers in the multimodal model. The only difference between the two deep learning models is that the multimodal model uses an additional pattern shape as a second modality data and fuses features from the structural parameter vector and the pattern shape.

[0084] Treating 2000 training samples as 100% data, we trained unimodal and multimodal models using different percentages of data, and then tested the trained deep learning models using the same 500 samples. Figure 3 The figure illustrates the mean squared error performance of the unimodal and multimodal deep learning models when using the same percentage of data. The horizontal axis represents the percentage of data samples used during training relative to the total training dataset, and the vertical axis represents the mean squared error of the electromagnetic response estimation of the trained model on the test dataset. The figure shows that the multimodal model can achieve the same performance as the unimodal model using only 39% of the training data, indicating that the multimodal model has higher estimation efficiency. Figure 4 The figure illustrates the size of the training sample set required for a single-modal or multimodal model to achieve the desired mean squared error performance. The horizontal axis represents the desired mean squared error performance for a single-modal or multimodal model, and the vertical axis represents the size of the training sample set required to achieve the desired performance. The figure shows that the multimodal model always requires fewer training samples to achieve the desired performance, indicating that the multimodal deep learning model has higher estimation accuracy.

[0085] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0086] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for constructing a multimodal deep learning model for estimating the electromagnetic properties of metasurfaces, characterized in that, The method for constructing a multimodal deep learning model for estimating the electromagnetic properties of metasurfaces includes: Initialize the metasurface's structure parameter vector in the parameter space, corresponding to the first step: The first step is to sample a metasurface structure of scale N in a given parameter space using a uniform sampling method. The parameter vector corresponding to the metasurface structure can be represented by a set X = {x1, x2, ..., x...}. N }express; Binarize the metasurface pattern corresponding to each structural parameter vector, corresponding to steps two and three: The second step is to use 3D modeling software to model the metasurface and extract its surface pattern for any parameter vector x in the set X. The third step is to binarize the metasurface pattern, where the areas covered by metal or resistive film are binarized as 1, and other parts are binarized as 0. The binarization matrix corresponding to each metasurface structure in the parameter space can be represented by a set P = {p1, p2, ..., p...} N }express; Construct the multimodal deep learning model MMN and initialize the trainable parameters, corresponding to step four: The fourth step involves constructing an MMN model using fully connected layers, one-dimensional batch normalization layers, convolutional layers, two-dimensional batch normalization layers, and activation layers. The MMN model has two input nodes and one output node. The set of parameters for the neurons in the model is W. The specific structure of the model can be described using the following construction steps: Step 1: Construct M fully connected layers FC1, FC2, ..., FC M M-1 one-dimensional batch normalized layers BN1D1, BN1D2, ..., BN1D M-1 M-1 activation layers RE1, RE2, ..., RE M-1 T convolutional layers CONV1, CONV2, ..., CONV T T two-dimensional batch normalized layers BN2D1, BN2D2, ..., BN2D T And an output layer OC; In step two, FC1 is set as the first input node of MMN, and FC1 is responsible for receiving the parameter vector x; CONV1 is set as the second input node of MMN, and CONV11 is responsible for receiving the binarized matrix p. Construction step three: Connect the first Q (Q < M) fully connected layers, the first Q one-dimensional batch normalization layers, and the first Q activation layers in sequence according to the order of {FC1, BN1D1, RE1, FC2, …, FC Q , BN1D Q , RE Q}, where the output of the previous layer network is used as the input of the next layer network; Step four involves constructing the T convolutional layers and T two-dimensional batch normalization layers according to {CONV1, BN2D1, CONV2, ..., CONV...}. T BN2D T The layers are connected in sequence, with the output of the previous layer serving as the input of the next layer. Step 5 of the construction process: RE Q The output vector and BN2D T The output vectors are added together to obtain vector I; Step six involves constructing the remaining fully connected layers, one-dimensional batch normalized layers, activation layers, and output layers according to {FC}. Q+1 BN1D Q+1 RE Q+1 ,…,FC M The layers OC are connected sequentially, with the output of the previous layer serving as the input of the next layer. FC Q+1 The input to the layer is vector I; the output of the neurons in the OC layer is the output of the MMN model. The finite element method is used to calculate the accurate electromagnetic response corresponding to the structural parameter vector. The structural parameter vector and the binarized matrix are regarded as inputs, and the electromagnetic response is regarded as the output. These are paired to form a training sample set, corresponding to steps five and six: The fifth step is to use the finite element method to calculate the electromagnetic property vector of the metasurface corresponding to each vector in set X, thus obtaining the set Y = {y1, y2, ..., y...}. N }; Step 6: Taking the set of structural parameter vectors X and the corresponding set of binary matrices P as input, and the set of electromagnetic properties Y as output, construct the training sample set Z = {z1, z2, ..., z...} N }={(X,P,Y)}={(x1,p1,y1),(x2,p2,y2),…,(x N ,p N ,y N )}; Based on the loss function and the training sample set, train the MMN model using gradient descent, corresponding to steps seven through ten: Step 7: In turn, select x with index i from the sample set Z. i and p i As input to the MMN, the MMN model is used to perform forward propagation on the samples. For any x... i and p i Obtain the floating-point numerical vector y generated by the output node neuron. i ′; Step 8: Based on the floating-point numerical vector y i ′ and y in the original training sample set i The loss function value L of the MMN model is calculated using the mean squared error loss function. i ; Step 9: Based on the loss function value L i Calculate the gradient G corresponding to the trainable parameters in the MMN model. i And according to the gradient G i Update the parameters W of the MMN model using gradient descent. Step 10: Repeat steps 7 through 9 to complete the training of the MMN model.

2. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the multimodal deep learning model construction method for estimating the electromagnetic properties of metasurfaces as described in claim 1.

3. A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the multimodal deep learning model construction method for estimating the electromagnetic properties of metasurfaces as described in claim 1.

4. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the multimodal deep learning model construction method for estimating the electromagnetic properties of metasurfaces as described in claim 1. The information data processing terminal includes: a wireless communication terminal and an electromagnetic simulation terminal.

Images