Joint discriminant generative adversarial network-based metasurface structure design method

Abstract

The invention relates to the field of electromagnetic super technology, in particular to a metasurface structure design method based on a joint discriminant generative adversarial network, which comprises the following steps: designing periodic arrangement of an electromagnetic metasurface unit structure and related parameters, inputting the periodic arrangement into electromagnetic simulation software to obtain electromagnetic response, and constructing a forward prediction data set and a reverse design data set; respectively segmenting the forward prediction data set and the reverse design data set into a training set and a test set; a forward prediction network model is constructed, a forward prediction data set is sent to a simulator network simulator for forward prediction for training, and parameters of the forward prediction network model are fixed after training is completed. According to the method, the pre-trained forward prediction network and the discriminator network for discriminating the spectrum similarity are added into the reverse design network, so that the network can well understand the relationship between the structural design and the overall optical response, more attention is paid to the generation of the absorption spectrum of the structure, a new design can be generated, and the design efficiency is improved. And the method can also be expanded to analysis of other metasurface structures.

Description

Technical Field

[0001] The present invention relates to the technical field of electromagnetic metamaterials, and in particular to a super surface structure design method based on a joint discriminant generative adversarial network. Background Art

[0002] The concept of electromagnetic metamaterials was first proposed in 1999, referring to artificially manufactured, periodic three-dimensional composite materials. Electromagnetic metamaterials can produce physical properties that existing materials in nature do not have, such as negative dielectric constant and negative magnetic permeability, through the periodic arrangement of artificial structures. Electromagnetic metasurfaces refer to artificial layered materials with a thickness less than the wavelength. Electromagnetic metasurfaces are considered to be the two-dimensional counterparts of electromagnetic metamaterials. Different from electromagnetic metamaterials, electromagnetic metasurfaces use the sudden phase and amplitude generated by the sudden change of electromagnetic waves on both sides of the metasurface to control the phase and amplitude distribution of the reflected field and the transmitted field in space. Electromagnetic metasurfaces can realize flexible control of electromagnetic wave amplitude, phase and other characteristics. Compared with traditional electromagnetic metamaterials, they have excellent characteristics such as low loss and good ductility.

[0003] Supercell surface design refers to the design of a surface structure for electromagnetic wave absorption within a specific frequency range, which is usually used in the preparation of absorbing materials or absorbing coatings. The absorption spectrum describes the absorption capacity of a material or surface to electromagnetic waves of different frequencies. In supercell surface design, there may be a variety of different surface structure designs, but the absorption spectra corresponding to these designs may tend to be similar or overlapping. Different surface designs may lead to similar absorbing performance, which is a "many-to-one" relationship. The specific manifestation of this "many-to-one" relationship depends on factors such as the material, structure and preparation process used in the supercell surface design. When designing a supercell surface, these factors are usually adjusted to achieve the desired absorbing performance, and it may be found that there is a certain degree of overlap or similarity between different designs.

[0004] Generative adversarial network (GAN) models are used for inverse design of supercell surfaces to provide cell structures with arbitrary patterns. GAN consists of a generative network that generates images and a discriminative network that distinguishes between generated images and real images. The generative network is trained to generate real images to deceive the discriminative network, and the discriminative network is trained not to be deceived by the generative network. The two networks compete with each other in each training step; ultimately, the competition leads to mutual improvement of each network, so that the generative network can generate realistic images with higher quality than when learning alone. In order to reduce time-consuming iterative simulations, data-driven design methods based on deep learning have been introduced into nanophotonics. Using generative adversarial networks to design nanophotonic structures can be free from the constraints of predefined shapes. For a given input absorption spectrum, the network produces an ideal design in the form of an image. The conditional deep convolutional generative adversarial network (cDCGAN) algorithm is an algorithm developed in recent years to solve the instability problem of GAN. The training of the generator and the discriminator depends not only on random noise vectors, but also on additional conditional information. This conditional information can be of any type, such as category labels, text descriptions, etc. By introducing conditional information, cDCGAN can control the samples generated by the generator to meet the provided conditions.

[0005] When cDCGAN is applied in the reverse design of metamaterials, the conditional information can represent the design specifications or goals that the generated metamaterial structure meets. Given a spectrum A and a latent vector z, the conditional information can be additional information describing the desired optical performance. The generator will use this conditional information to adjust the generated metamaterial structure, using cDCGAN to generate metamaterial structures with specific optical properties, rather than just generating structures from random noise to meet the provided optical performance requirements. The generated designs are presented in the form of images, so they basically provide any possible design to meet the required optical properties without being limited to specific structures. In general, the cDCGAN model is used to implement a framework for reverse design of super-unit surface nanostructures. The network is trained in the form of a game between the generator and the discriminator to meet the on-demand design goals. The generated results focus on the similarity between the generated images used to represent the super-unit surface structure and the existing samples in the training set, and there is no effective constraint on the degree of fit of the spectral response corresponding to the structure to the target response. Summary of the invention

[0006] The purpose of the present invention is to overcome the above-mentioned problems or at least partially solve the above-mentioned problems, and a super surface structure design method based on a joint discriminant generative adversarial network is proposed. By effectively introducing spectral response loss in the model architecture design and the objective function design, the network back propagation process mainly relies on the degree of fit between the spectral response corresponding to the generated surface structure design and the target spectral response to optimize the network parameters, which effectively avoids the problem that traditional reverse design can only generate super unit surface structures in a similar image domain based on the spectrum.

[0007] To achieve the above object, the present invention provides the following technical solution: a method for designing a hypersurface structure based on a joint discriminant generative adversarial network, comprising the following steps:

[0008] S1. Design the periodic arrangement of the electromagnetic metasurface unit structure and related parameters, and input them into the electromagnetic simulation software to obtain the electromagnetic response, and construct a forward prediction data set and a reverse design data set; both the forward prediction data set and the reverse design data set include the generated electromagnetic response absorption spectrum data and the metasurface pattern;

[0009] S2, split the forward prediction data set and the reverse design data set into a training set and a test set respectively, where the proportion of the training set is greater than the proportion of the test set;

[0010] S3, constructing a forward prediction network model, sending the segmented forward prediction data set to the simulator network Simulator for forward prediction for training, and fixing the forward prediction network model parameters after the training is completed;

[0011] S4, constructing a reverse design network model, including a generator network Generator for generating a metasurface pattern, an image discriminator network Discriminator_A for judging the authenticity of the pattern, and a spectrum discriminator network Discriminator_B for judging the authenticity of the spectrum, and initializing network parameters;

[0012] S5. Send the training set in the reverse-engineered dataset into the generator network Generator, send the generated images and the real images into the image discriminator network Discriminator_A, send the spectrum predicted by the generated images and the real spectrum into the spectral discriminator network Discriminator_B, and perform iterative cycle training.

[0013] In a preferred embodiment, the electromagnetic supersurface unit includes an upper layer, a middle layer and a lower layer; the upper layer is divided into a coding pattern area; the middle layer dielectric substrate is made of F4B; and the lower layer is covered with full metal copper.

[0014] In a preferred embodiment, in S1, the steps of constructing a forward prediction data set and a reverse design data set include:

[0015] Draw different metasurface unit patterns, construct corresponding electromagnetic metasurface units, and send them to electromagnetic simulation software to obtain corresponding electromagnetic responses;

[0016] The encoding matrix of the metasurface unit pattern is used as the network input. In this process, the absorbance Amp is converted into the reflectivity S 11 and transmittance S 12 The relationship is calculated as follows:

[0017]

[0018] Through the reflectivity S 11 and transmittance S 12 Calculate the wave absorption rate Amp, perform uniform frequency sampling on the wave absorption rate Amp, express it as a multi-dimensional vector as the label of the data set, and construct a forward prediction data set;

[0019] The absorbance Amp is represented as a multi-dimensional vector and used as input at the same time, and its corresponding metasurface unit matrix is ​​used as a label to construct a reverse design dataset.

[0020] In a preferred embodiment, in the S3, the forward prediction network model constructed, the simulator network Simulator adopts the ResNet18 structure, the optimizer of the network Simulator is set to Adam, and the learning rate is 1e-3;

[0021] use Represents the network loss function, where y is the true value, is the predicted value, m is the dimension, and the convergence of the model is improved through a dynamic adjustment mechanism. During the training process, the network's prediction of the absorption rate is gradually optimized. After the training is completed, the forward prediction network model parameters are fixed.

[0022] In a preferred embodiment, in the S4, the reverse design network model is constructed, the generator network Generator is composed of a deconvolution layer, receives the absorbance vector as input, and generates a matrix representing the metasurface pattern through a deconvolution operation; the image discriminator network Discriminator_A is composed of a convolution layer, the real image in the data set and the image generated by the generator and the absorption spectrum are input into the discriminator, the image features are converted into digital features, and the discrimination probability of the authenticity of the image is output; the spectrum discriminator Discriminator_B is composed of a linear layer, the real spectrum and the predicted spectrum of the generated image are input, and the discrimination probability of the authenticity of the spectrum is output.

[0023] In a preferred embodiment, in S2, the forward prediction data set input is a 64×64 encoding matrix, and the corresponding label is a 51-dimensional absorbance Amp vector; the reverse design data set input is a 51-dimensional absorbance Amp vector, and the corresponding label is a 64×64 encoding matrix.

[0024] In a preferred embodiment, in S5, each round of iterative calculation process includes the following steps:

[0025] S5.1, feeding the absorbance vector of the reverse-designed data set into the generator network Generator, concatenating the input vector with random noise, and processing it through multiple deconvolution layers and activation functions to generate a metamaterial structure image;

[0026] S5.2. The generated metamaterial structure image and the original metamaterial structure image corresponding to the absorbance vector are sent to the image discriminator network Discriminator_A. The input vector is concatenated with random noise, and after being processed by multiple convolutional layers and activation functions, the probability for judging the authenticity of the image is output.

[0027] S5.3, the generated metamaterial structure image is sent to the forward prediction network, the predicted absorbance vector and the real absorbance vector are input into the spectrum discriminator Discriminator_B, and after being processed by multiple linear layers and activation functions, the probability for judging the authenticity of the spectrum is output;

[0028] S5.4. Calculate the error through the image prediction value and the spectral prediction value, and perform the back propagation algorithm to update the network parameters of the real part generator network Generator, the image discriminator network Discriminator_A and the spectral discriminator network Discriminator_B.

[0029] In a preferred embodiment, in S5.2, the loss of the image discriminator network Discriminator_A is defined as:

[0030]

[0031] Where m is the batch size, the superscript i represents the i-th data in the batch, the loss is the cross entropy between the image dataset x and the generated data G(A,z), and the discriminator output is between 0 and 1. In order to distinguish between real and fake images, D(x i ) approaches 1, D(G(A i ,z i )) approaches 0, and this loss defines the similarity of the pattern between the image data and the generated image.

[0032] In a preferred embodiment, in said S5.3, the loss of the spectral discriminator Discriminator_B is defined as:

[0033]

[0034] The loss is measured by the spectral data A and the predicted spectrum The cross entropy of the discriminator is between 0 and 1. In order to distinguish true and false spectra, D(A i ) approaches 1, Approaching 0, this loss defines the similarity between the spectral data and the predicted spectrum.

[0035] In a preferred embodiment, in said S5.4, the parameters of the generator are updated by back-propagation through the two discriminator losses:

[0036]

[0037] In the game between the generator and the discriminator, D(G(A i ,z i ))and will approach 1, which introduces the reconstruction error of the spectral response. λ is a parameter that balances the spectrum and image loss, and λ=0.1 is selected.

[0038] Compared with the prior art, the present invention has the following beneficial effects: by adding the pre-trained forward prediction network and the discriminator network for discriminating spectral similarity into the inverse design network, the network can well understand the relationship between the structural design and its overall optical response, and pay more attention to the absorption spectrum of the generated structure. This network structure is not limited to predefined structures, but can also generate new designs and can be extended to analyze other metasurface structures, so it can be widely used in the system design of nanophotonic structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic diagram of the structure of the electromagnetic supersurface unit of the present invention;

[0040] Figure 2 The structure and training flow chart of the forward network model of the present invention;

[0041] Figure 3 The structure and training flow chart of the reverse-design network model of the present invention;

[0042] Figure 4 Schematic diagram of the starting template of the electromagnetic supersurface unit of the present invention;

[0043] Figure 5 This is a downward trend diagram of the loss function during the forward network training process of the present invention;

[0044] Figure 6 To reverse design the network for the difference between the design value and the simulation value of the existing pattern;

[0045] Figure 7 The difference between the design value and simulation value of the reverse design network for a new pattern. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] See also Figure 1-4 The present invention provides a technical solution: a method for designing a hypersurface structure based on a joint discriminant generative adversarial network, comprising the following steps:

[0048] S1. Design the periodic arrangement of the electromagnetic metasurface unit structure and related parameters, and input them into the electromagnetic simulation software to obtain the electromagnetic response, and construct a forward prediction data set and a reverse design data set; both the forward prediction data set and the reverse design data set include the generated electromagnetic response absorption spectrum data and the metasurface pattern;

[0049] In specific implementation, the electromagnetic metasurface unit is as follows: Figure 1 As shown, Figure 1 (a) is a front view, (b) is a side view, the electromagnetic metasurface unit includes an upper layer, a middle layer and a lower layer; the upper layer is divided into a coding pattern area, which is divided into a 64×64 matrix in the coding pattern area, the conductivity of the coding pattern metal copper patch is 5.8e+007S / m, and the thickness is t1=0.1mm; the material of the middle layer dielectric substrate is F4B, the dielectric constant is 2.65(1+0.003i), the width is L=2mm, and the thickness is t2=0.2mm; the lower layer is covered with full metal copper, the conductivity is 5.8e+007S / m, and the thickness is t3=0.1mm. The electromagnetic response of the electromagnetic metasurface unit of this structure depends on the upper coding pattern style.

[0050] In specific implementation, the steps for constructing the forward prediction dataset and the reverse design dataset are as follows:

[0051] Draw different 64×64 pixel metasurface unit patterns, construct corresponding electromagnetic metasurface units, and send them to electromagnetic simulation software to obtain corresponding electromagnetic responses. The frequency range is set to 120-160 Thz.

[0052] The 64×64 encoding matrix of the metasurface unit pattern is used as the network input. In this process, the absorbance Amp is compared with the reflectivity S 11 and transmittance S 12 The relationship is calculated as follows:

[0053]

[0054] Through the reflectivity S 11 and transmittance S 12 Calculate the wave absorption rate Amp, perform uniform frequency sampling on the wave absorption rate Amp, express it as a multi-dimensional vector as the label of the data set, and construct a forward prediction data set;

[0055] The absorbance Amp is represented as a 51-dimensional vector and used as input, and the corresponding metasurface unit matrix is ​​used as a label to construct a reverse design dataset.

[0056] S2, split the forward prediction dataset and the reverse design dataset into 80% training set and 20% test set respectively;

[0057] In the specific implementation, the forward prediction data set input is a 64×64 encoding matrix, and the corresponding label is a 51-dimensional absorbance Amp vector; the reverse design data set input is a 51-dimensional absorbance Amp vector, and the corresponding label is a 64×64 encoding matrix.

[0058] S3, construct a forward prediction network model, send the segmented forward prediction data set to the simulator network Simulator for forward prediction for training, and fix the forward prediction network model parameters after the training is completed;

[0059] In the specific implementation, the forward prediction network model constructed, the simulator network Simulator adopts the ResNet18 structure, the optimizer of the network Simulator is set to Adam, and the learning rate is 1e-3;

[0060] use Represents the network loss function, where y is the true value, is the predicted value, m is the dimension, and the convergence of the model is improved through a dynamic adjustment mechanism. During the training process, the network's prediction of the absorption rate is gradually optimized. After the training is completed, the forward prediction network model parameters are fixed.

[0061] S4, constructing a reverse design network model, including a generator network Generator for generating a metasurface pattern, an image discriminator network Discriminator_A for judging the authenticity of the pattern, and a spectrum discriminator network Discriminator_B for judging the authenticity of the spectrum, and initializing network parameters;

[0062] In the specific implementation, the reverse design network model is constructed, such as Figure 3 As shown in the figure, the generator network Generator is composed of deconvolution layers, which inputs a 51-dimensional absorbance vector Amp and generates a 64×64×1 matrix representing the metasurface pattern through deconvolution operations; the image discriminator network Discriminator_A is composed of convolution layers, and the real image in the data set and the image generated by the generator and the absorption spectrum are input into the discriminator, the image features are converted into digital features, and the discrimination probability 0 or 1 is output; the spectral discriminator Discriminator_B is composed of linear layers, which inputs the real spectrum and the predicted spectrum of the generated image, and outputs the discrimination probability 0 or 1.

[0063] S5. Send the training set in the reverse-engineered dataset into the generator network Generator, send the generated images and the real images into the image discriminator network Discriminator_A, send the spectrum predicted by the generated images and the real spectrum into the spectral discriminator network Discriminator_B, and perform iterative cycle training.

[0064] In specific implementation, each round of iterative calculation process includes the following steps:

[0065] S5.1. Send the absorbance vector of the reverse-engineered data set into the generator network Generator, concatenate the 51-dimensional spectral data A and the 20-dimensional Gaussian random noise z through the concat operation, and pass them into the five deconvolution network layers of the generator in sequence, where the number of nodes in the deconvolution layer is 64, 32, 16, 8, and 1 respectively. Finally, use the TanH activation function to limit the output to [-1,1] to generate a metamaterial structure image.

[0066] S5.2. The generated metamaterial structure image and the original metamaterial structure image corresponding to the absorbance vector are sent to the image discriminator network Discriminator_A and concatenated with the absorbance vector. After passing through 5 convolution layers, the image features are converted into digital features, which are composed of 5 convolution structures. Finally, the Sigmoid activation function is used for binary classification to discriminate the output 0 or 1, which is applied to the classification of images to predict whether the input is a real pattern or a generated pattern. The loss of the image discriminator network Discriminator_A is defined as:

[0067]

[0068] Where m is the batch size, the superscript i represents the i-th data in the batch, the loss is the cross entropy between the image dataset x and the generated data G(A,z), and the discriminator output is between 0 and 1. In order to distinguish between real and fake images, D(x i) approaches 1, D(G(A i ,z i )) approaches 0, and this loss defines the similarity of the pattern between the image data and the generated image;

[0069] S5.3. The generated metamaterial structure image is sent to the forward prediction network, and the predicted absorbance vector and the real absorbance vector are input into the spectrum discriminator Discriminator_B. After passing through 4 linear layers, the Sigmoid activation function is used for binary classification. The discriminant output is 0 or 1, and the predicted input is a real spectrum or a false spectrum. The loss of the spectrum discriminator Discriminator_B is defined as:

[0070]

[0071] The loss is measured by the spectral data A and the predicted spectrum The cross entropy of the discriminator is between 0 and 1. In order to distinguish true and false spectra, D(A i ) approaches 1, Approaching 0, this loss defines the similarity between the spectral data and the predicted spectrum;

[0072] S5.4. Calculate the error through the image prediction value and the spectral prediction value, and perform the back propagation algorithm to update the network parameters of the real generator network Generator, the image discriminator network Discriminator_A, and the spectral discriminator network Discriminator_B. The parameters of the generator are updated by the back propagation of the loss of the two discriminators:

[0073]

[0074] In order to make the generated image be able to deceive the discriminator well, in the game between the generator and the discriminator, D(G(A i ,z i ))and will approach 1. The reconstruction error of the spectral response is introduced, and λ is a parameter that balances the spectrum and image loss. In order to make the absorption spectrum of the reverse-designed graphic structure as similar as possible to that of the original graphic structure, rather than considering the structural image similarity, λ=0.1 is selected.

[0075] Examples are listed below to illustrate the present invention:

[0076] Embodiment 1:

[0077] The training dataset for deep learning consists of 5800 MIM surface unit cell designs. Figure 4As shown, these designs are derived from six starting shape templates: open ring, cross, H, bow tie, square, and donut. Full-wave simulations are performed on each structure (under TE polarization at vertical incidence) to generate corresponding 1001-point absorption spectra in the 120-160THz frequency band. The response range in the training dataset can be expanded in future studies to enhance the prediction ability of the network. 80% of the data in the dataset is used as a training set, 10% of the data is used as a validation set, and 10% is used as a test set. The image of each dimension of the data in the dataset is a single-channel png image with a resolution of 64×64, and the corresponding label is a 51-dimensional vector Absorption.

[0078] According to step 3, the forward prediction network model is built, and the forward prediction data set is sent to the forward prediction network model for iterative cycle training. In the case of the Adam optimizer, the learning rate is dynamically adjusted, the initial learning rate is 0.01, and the learning rate is adjusted to the original 0.1 every 250 rounds of training. The training cycle is set to 1000.

[0079] Figure 5 The graph shows the decrease in the total loss value during 1000 training cycles, and the training set loss is similar to the test set loss, and finally they decrease synchronously and converge to a lower value. The forward prediction network has high accuracy and generalization. After 1000 cycles of training, the loss on the validation set is 0.0052, and a relatively accurate forward prediction result can be obtained.

[0080] The parameters of the trained forward prediction network are fixed. According to step 4, a reverse design network model is constructed, including a generator network Generator for generating a hypersurface pattern, a discriminator network Discriminator_A for judging the authenticity of the pattern, and a discriminator network Discriminator_B for judging the authenticity of the spectrum, and iterative cycle training is performed.

[0081] The MAE and model accuracy of the absorption spectrum are defined as:

[0082]

[0083] Randomly select cases from the test set and perform reverse design based on the absorption curve of the new pattern. The difference between the design value and the simulation value of the reverse design network is as follows: Figure 6 and Figure 7 As shown in Figure 1, the similarity between the design value and the simulation value of the reverse design network in the absorbance vector is extremely high. Specifically, the specific values ​​of the simulation values ​​of different types of super unit patterns are shown in Table 1. The accuracy is improved by about 1.2% compared with the traditional cDCGAN. The reverse design network has achieved a high accuracy and can meet the needs of fast and accurate design.

[0084]

[0085] Table 1 Reverse design network simulation value table

[0086] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A hypersurface structure design method based on joint discriminant generative adversarial network, characterized in that: The following steps are involved: S1. Design the periodic arrangement of the electromagnetic metasurface unit structure and related parameters, and input them into the electromagnetic simulation software to obtain the electromagnetic response, and construct a forward prediction data set and a reverse design data set; both the forward prediction data set and the reverse design data set include the generated electromagnetic response absorption spectrum data and the metasurface pattern; S2, split the forward prediction data set and the reverse design data set into a training set and a test set respectively, where the proportion of the training set is greater than the proportion of the test set; S3, constructing a forward prediction network model, sending the segmented forward prediction data set to the simulator network Simulator for forward prediction for training, and fixing the forward prediction network model parameters after the training is completed; S4, constructing a reverse design network model, including a generator network Generator for generating a metasurface pattern, an image discriminator network Discriminator_A for judging the authenticity of the pattern, and a spectrum discriminator network Discriminator_B for judging the authenticity of the spectrum, and initializing network parameters; S5. Send the training set in the reverse-engineered dataset into the generator network Generator, send the generated images and the real images into the image discriminator network Discriminator_A, send the spectrum predicted by the generated images and the real spectrum into the spectral discriminator network Discriminator_B, and perform iterative cycle training.

2. The method for designing a hypersurface structure based on a joint discriminant generative adversarial network according to claim 1, characterized in that: The electromagnetic super surface unit includes an upper layer, a middle layer and a lower layer; the upper layer is divided into a coding pattern area; the middle layer dielectric substrate is made of F4B; and the lower layer is covered with full metal copper.

3. The method for designing a hypersurface structure based on a joint discriminant generative adversarial network according to claim 2, characterized in that: In S1, the steps of constructing a forward prediction data set and a reverse design data set include: Draw different metasurface unit patterns, construct corresponding electromagnetic metasurface units, and send them to electromagnetic simulation software to obtain corresponding electromagnetic responses; The encoding matrix of the metasurface unit pattern is used as the network input. In this process, the absorbance Amp is converted into the reflectivity S 11 and transmittance S 12 The relationship is calculated as follows: Through the reflectivity S 11 and transmittance S 12 Calculate the wave absorption rate Amp, perform uniform frequency sampling on the wave absorption rate Amp, express it as a multi-dimensional vector as the label of the data set, and construct a forward prediction data set; The absorbance Amp is represented as a multi-dimensional vector and used as input at the same time, and its corresponding metasurface unit matrix is ​​used as a label to construct a reverse design dataset.

4. The method for designing a hypersurface structure based on a joint discriminant generative adversarial network according to claim 3, characterized in that: In the S3, the forward prediction network model is constructed, the simulator network Simulator adopts the ResNet18 structure, the optimizer of the network Simulator is set to Adam, and the learning rate is 1e-3; use Represents the network loss function, where y is the true value, is the predicted value, m is the dimension, and the convergence of the model is improved through a dynamic adjustment mechanism. During the training process, the network's prediction of the absorption rate is gradually optimized. After the training is completed, the forward prediction network model parameters are fixed.

5. The method for designing a hypersurface structure based on a joint discriminant generative adversarial network according to claim 4, characterized in that: In the S4, the reverse design network model is constructed, the generator network Generator is composed of deconvolution layers, receives the absorbance vector as input, and generates a matrix representing the metasurface pattern through deconvolution operation; the image discriminator network Discriminator_A is composed of convolution layers, the real image in the data set and the image generated by the generator and the absorption spectrum are input into the discriminator, the image features are converted into digital features to output the discrimination probability of the authenticity of the image; the spectrum discriminator Discriminator_B is composed of linear layers, the real spectrum and the predicted spectrum of the generated image are input, and the discrimination probability of the authenticity of the spectrum is output.

6. The method for designing a hypersurface structure based on a joint discriminant generative adversarial network according to claim 5, characterized in that: In S2, the forward prediction data set input is a 64×64 encoding matrix, and the corresponding label is a 51-dimensional absorbance Amp vector; the reverse design data set input is a 51-dimensional absorbance Amp vector, and the corresponding label is a 64×64 encoding matrix.

7. The method for designing a hypersurface structure based on a joint discriminant generative adversarial network according to claim 6, characterized in that: In S5, each round of iterative calculation process includes the following steps: S5.1, feeding the absorbance vector of the reverse-designed data set into the generator network Generator, concatenating the input vector with random noise, and processing it through multiple deconvolution layers and activation functions to generate a metamaterial structure image; S5.

2. The generated metamaterial structure image and the original metamaterial structure image corresponding to the absorbance vector are sent to the image discriminator network Discriminator_A. The input vector is concatenated with random noise, and after being processed by multiple convolutional layers and activation functions, the probability for judging the authenticity of the image is output. S5.3, the generated metamaterial structure image is sent to the forward prediction network, the predicted absorbance vector and the real absorbance vector are input into the spectrum discriminator Discriminator_B, and after being processed by multiple linear layers and activation functions, the probability for judging the authenticity of the spectrum is output; S5.

4. Calculate the error through the image prediction value and the spectral prediction value, and perform the back propagation algorithm to update the network parameters of the real part generator network Generator, the image discriminator network Discriminator_A and the spectral discriminator network Discriminator_B.

8. The method for designing a hypersurface structure based on a joint discriminant generative adversarial network according to claim 7, characterized in that: In S5.2, the loss of the image discriminator network Discriminator_A is defined as: Where m is the batch size, the superscript i represents the i-th data in the batch, the loss is the cross entropy between the image dataset x and the generated data G(A,z), and the discriminator output is between 0 and 1. In order to distinguish between real and fake images, D(x i ) approaches 1, D(G(A i ,z i )) approaches 0, and this loss defines the similarity of the pattern between the image data and the generated image.

9. The method for designing a hypersurface structure based on a joint discriminant generative adversarial network according to claim 7, characterized in that: In S5.3, the loss of the spectral discriminator Discriminator_B is defined as: The loss is measured by the spectral data A and the predicted spectrum The cross entropy of the discriminator is between 0 and 1. In order to distinguish true and false spectra, D(A i ) approaches 1, Approaching 0, this loss defines the similarity between the spectral data and the predicted spectrum.

10. The method for designing a hypersurface structure based on a joint discriminant generative adversarial network according to claim 7, characterized in that: In S5.4, the parameters of the generator are updated by back-propagation through the two discriminator losses: In the game between the generator and the discriminator, D(G(A i ,z i ))and will approach 1, which introduces the reconstruction error of the spectral response. λ is a parameter that balances the spectrum and image loss, and λ=0.1 is selected.

Images