Multi-structure terahertz metamaterial sensor reverse design method based on improved conditional generative network

By improving the conditional generative network combined with generative adversarial network and forward neural network, the complexity and parameter optimization problems of terahertz metamaterial sensor design in traditional methods are solved, and efficient and accurate multi-structure terahertz metamaterial sensor design is achieved, significantly improving the design of terahertz metamaterial sensors. Design efficiency and computational stability are obtained.

CN119026454APending Publication Date: 2024-11-26HENAN UNIVERSITY OF TECHNOLOGY
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410949526.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

It is difficult to design multi-structure terahertz metamaterial sensors that meet the requirements in the prior art. Traditional methods rely on experience and trial and error, difficult to optimize parameters, and difficult to achieve the liberalized design of terahertz metamaterials.

Method used

Using a multi-structure terahertz metamaterial sensor reverse design method based on improved conditional generation network, combining SAGAN, CGAN and WGAN algorithms to construct a generative adversarial network, generate and verify the terahertz metamaterial structure through generator and discriminator, using positive Verify the design results to the neural network.

Benefits of technology

The efficient reverse design of terahertz metamaterial sensor is realized, which improves design efficiency and image generation quality, significantly saves calculation time and ensures the accuracy of design results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026454A_ABST
    Figure CN119026454A_ABST
Patent Text Reader

Abstract

The invention provides a multi-structure terahertz metamaterial sensor reverse design method based on an improved conditional generative network. The method comprises the following steps: (1) constructing a terahertz metamaterial sensor structure, and establishing sensor structure parameters and a sample data set corresponding to an absorption curve; (2) constructing and training a reverse design model of the multi-structure terahertz metamaterial sensor; in the training process of the generative adversarial network, inputting a conditional tensor formed by an absorption curve and a randomly generated noise vector for cascade, forming conditional noise, and transmitting the conditional noise to a generator to generate enough pseudo sample data; the discriminator calculates the Wasserstein distance between the real sample data and the pseudo sample data to realize the reverse design of the terahertz metamaterial sensor structure; and (3) constructing and training a forward neural network model to verify whether the terahertz metamaterial sensor structure designed by the generative adversarial network meets design requirements or not. According to the invention, the efficient reverse design of the terahertz metamaterial sensor is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of terahertz metamaterials, and in particular to a reverse design method for a multi-structure terahertz metamaterial sensor based on an improved conditional generative network. Background Art

[0002] Terahertz metamaterial sensor is a highly sensitive sensor device that uses the unique electromagnetic response of terahertz metamaterials to detect different chemical substances, biological molecules or environmental changes. In the terahertz band, the sensor can accurately capture the unique fingerprint information of materials or substances, opening up new ways for the detection and application of trace substances. The typical terahertz metamaterial absorber is a "sandwich" structure of metal resonance-insulator-metal, which can be used as a sensor. The metal resonance layer can be designed into different shapes, such as double opening shape, "H" shape, cross shape, symmetrical cutting line shape, etc., to achieve the effect of flexible regulation of electromagnetic waves.

[0003] In order to design a metamaterial sensor that meets the requirements, traditional design methods often rely on experience and trial and error. Usually, it is necessary to first determine the device function, consult relevant device literature, select appropriate materials, number of layers, structure and shape, etc., and then conduct test analysis after modeling through simulation software. Finally, the physical object is prepared according to the model for experiment, and the test results are verified to see if they meet the requirements. The requirements are adjusted according to the results, and this is repeated. Methods that rely on artificial experience design are no longer able to meet the needs. The trial and error process guided by traditional experiments makes it difficult to identify the most representative features from the many parameters to be optimized, which limits the search for undeveloped areas in the solution space.

[0004] In recent years, with the rapid development of artificial intelligence, deep learning has provided new ideas for sensor design. Deep learning can be used for data processing, feature extraction, and structural optimization. However, most of the current research work is the reverse design of a single structure, which makes it difficult to achieve the free design of multiple structures of terahertz metamaterials. Therefore, improving the model's degree of freedom in predicting the shape of terahertz metamaterials, and the ability to predict and generalize the structural parameters of terahertz metamaterials are still the top priorities of current research. Summary of the invention

[0005] The present invention aims to solve the problems of complex structure, limited control methods and relatively single function of some current terahertz metamaterial devices. To this end, the present invention proposes a reverse design method for a multi-structure terahertz metamaterial sensor based on an improved conditional generative network that can achieve broadband absorption and electromagnetic induced transparency.

[0006] A first aspect of the present invention provides a method for reverse designing a multi-structure terahertz metamaterial sensor based on an improved conditional generative network, the method comprising: (1) Construct a terahertz metamaterial sensor structure, and based on the constructed terahertz metamaterial sensor structure, establish a sample data set of sensor structure parameters and corresponding absorption curves; (2) Combining SAGAN, CGAN and WGAN algorithms, construct and train a reverse design model for multi-structure terahertz metamaterial sensors; The multi-structure terahertz metamaterial sensor reverse design model is a generative adversarial network based on the SACW-GAN algorithm; In the training process of the generative adversarial network, the input is a conditional tensor consisting of an absorption curve and a randomly generated noise vector cascaded to form conditional noise passed to the generator to generate sufficient pseudo sample data; the discriminator calculates the Wasserstein distance between the real sample data and the pseudo sample data to achieve the reverse design of the terahertz metamaterial sensor structure; (3) Construct and train a forward neural network model to verify whether the terahertz metamaterial sensor structure designed by the generative adversarial network meets the design requirements.

[0007] Based on the above, the real sample data y and the pseudo sample data The Wasserstein distance between is expressed as:

[0008] in, , , represents the absorption curve extracted from the training data, It represents the absorption curve dataset generated by the generator, E represents the expectation, and D represents the Wasserstein distance.

[0009] Based on the above, the input layer of the generator is the absorption spectrum line data and random noise; the absorption spectrum line is connected to the noise through the concat layer, and then passes through 6 deconvolution layers in sequence to convert the two-dimensional data into three-dimensional data; in the last two deconvolution layers, the self-attention mechanism is added; the activation function uses the Sigmoid function.

[0010] Based on the above, the real sample data and the pseudo sample data generated by the generator are used as the input of the discriminator. The input data passes through 5 deconvolution layers in sequence to convert the image data into digital features. In the last two deconvolution layers, the self-attention mechanism is added; the tanh function is selected as the activation function.

[0011] Based on the above, the constructed terahertz metamaterial sensor structures include four types: circular open ring type, square open ring type, H type and F type; The four terahertz metamaterial sensor structures are all composed of three layers. The first layer is a metal resonance layer made of gold, the second layer is a dielectric layer made of polytetrafluoroethylene, and the third layer is a substrate layer made of gold. The terahertz metamaterial sensor structure is represented by four parameters: Unit period u, resonator line width w of the metal resonator layer, thickness h of the dielectric layer, resonator length a of the metal resonator layer and thickness of the substrate layer 0.2 um; For the unit period u, the circular open ring type is marked as 78-82um, the square open ring type is marked as 75-84um, the H type is marked as 65-74um, and the F type is marked as 75-84um; For the dielectric layer thickness h, the circular open ring type is marked as 28-32um, the square open ring type is marked as 35-44um, the H type is marked as 20-29um, and the F type is marked as 55-64um; For the resonator length a, the circular open ring type is marked as 16-20um, the square open ring type is marked as 35-44um, the H type is marked as 20-29um, and the F type is marked as 25-34um; For the resonator line width w, the circular open ring type is marked as 2-5um, the square open ring type is marked as 3-5um, the H type is marked as 3-5um, and the F type is marked as 3-5um.

[0012] Based on the above, the sensor structure parameters and the corresponding absorption curve sample data set are established as follows: The resonators of the four terahertz metamaterial sensor structures are numbered, where the circular open ring type is marked as "1", the square open ring type is marked as "2", the H type is marked as "3", and the F type is marked as "4"; The sampling frequencies of the absorption curves of the four terahertz metamaterial sensor structures were set to 0.5-1.5 THz, 1.5-2.5 THz, 2.5-3.5 THz, and 0.5-1.5 THz, respectively. The absorption curves were sampled at 201 points within the above frequency ranges. In this sample data set, 3000 sets of structural parameters and absorption curves are calculated for each terahertz metamaterial sensor structure, with a total of 12000*206 dimensional data, of which 12000*201 is the absorption curves of the four terahertz metamaterial sensor structures, the 202nd column is the type code of the four terahertz metamaterial sensor structures, and the 203rd-206th columns are the structural parameters of the four terahertz metamaterial sensor structures.

[0013] The second aspect of the present invention provides a multi-structure terahertz metamaterial sensor reverse design system based on an improved conditional generative network, comprising: A parameter and data set construction module is used to: construct a terahertz metamaterial sensor structure, and based on the constructed terahertz metamaterial sensor structure, establish a sample data set of sensor structure parameters and corresponding absorption curves; The reverse design model module is used to: combine SAGAN, CGAN and WGAN algorithms to construct and train a reverse design model for multi-structure terahertz metamaterial sensors; The multi-structure terahertz metamaterial sensor reverse design model is a generative adversarial network based on the SACW-GAN algorithm; In the training process of the generative adversarial network, the input is a conditional tensor consisting of an absorption curve and a randomly generated noise vector cascaded to form conditional noise passed to the generator to generate sufficient pseudo sample data; the discriminator calculates the Wasserstein distance between the real sample data and the pseudo sample data to achieve the reverse design of the terahertz metamaterial sensor structure; The forward neural network model module is used to construct and train a forward neural network model to verify whether the terahertz metamaterial sensor structure designed by the generative adversarial network meets the design requirements.

[0014] A third aspect of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the reverse design method of a multi-structure terahertz metamaterial sensor based on an improved conditional generative network are implemented.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the reverse design method for a multi-structure terahertz metamaterial sensor based on an improved conditional generative network.

[0016] Compared with the prior art, the present invention has outstanding substantive features and significant progress, specifically: (1) By combining the SAGAN, CGAN, and WGAN algorithms, efficient reverse design of terahertz metamaterial sensors was achieved; (2) Using deep learning methods, the quality of generated images and training stability are improved; (3) A forward neural network was constructed to ensure the accuracy of the reverse design results; (4) Improved design efficiency and significantly saved calculation time. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the structure of the terahertz metamaterial sensor structure in the present invention; wherein, (a) Round opening type (b) Square opening type (c) H type (d) F type.

[0018] Figure 2 The figure is a flow chart of reverse design of metamaterial structure by SACW-GAN model in the method of the present invention.

[0019] Figure 3 This is a network structure diagram of the SACW-GAN model generator in the method of the present invention.

[0020] Figure 4 Schematic diagram of the network structure of the SACW-GAN model discriminator in the method of the present invention.

[0021] Figure 5 It is the iterative convergence process of the forward neural network in the method of the present invention.

[0022] Figure 6 It is a comparison diagram of the absorption curves predicted by the forward neural network and numerically simulated in the method of the present invention; wherein, (a) Circular open ring type (b) Square open ring type (c) F type (d) H type.

[0023] Figure 7 is a comparison diagram of the target absorption spectrum in the method of the present invention and the absorption curve predicted by the forward neural network; wherein, (a) Circular open ring type (b) Square open ring type (c) F type (d) H type. DETAILED DESCRIPTION

[0024] This embodiment proposes a reverse design method for a multi-structure terahertz metamaterial sensor based on an improved conditional generative network, comprising the following steps: Sensor structure construction The constructed terahertz metamaterial sensor structures include four types: circular open ring type, square open ring type, H type and F type, such as Figure 1 As shown; The four terahertz metamaterial sensor structures are composed of three layers. The first layer is made of gold (conductivity is 3.56×10 7 S / m) of metal resonance layer, the second layer is a dielectric layer made of polytetrafluoroethylene (dielectric constant is 2.1+0.00027i), and the third layer is a layer made of gold (conductivity is 3.56×10 7 S / m) substrate layer; The structure of the terahertz metamaterial sensor is represented by four parameters: unit period u, resonator line width w of the metal resonance layer, dielectric layer thickness h, and resonator length a of the metal resonance layer. By changing the structural parameters and analyzing the electromagnetic response of the metamaterial sensor, it is found that the thickness of the substrate layer has little effect on the absorption curve, so the thickness of the substrate layer used is 0.2 um.

[0025] Sample data set creation The material properties, structural shape and other factors of terahertz metamaterial devices jointly determine the frequency and amplitude of their electromagnetic response. In this embodiment, the resonators of the four terahertz metamaterial sensor structures are numbered, where the circular open ring type is marked as "1", the square open ring type is marked as "2", the H type is marked as "3", and the F type is marked as "4". Different terahertz metamaterial sensors are represented by numbers, which is conducive to the effective identification of the model. For different terahertz metamaterial sensor shapes, considering the characteristics of the absorption curve and the effective frequency band of operation, the sampling frequencies of the absorption curves of the four terahertz metamaterial sensor structures are set to 0.5-1.5 THz, 1.5-2.5 THz, 2.5-3.5 THz, and 0.5-1.5THz, respectively, and the absorption curves are sampled at 201 points within the above frequency range; Specifically, through joint simulation of CST and MATLAB, the absorption curves corresponding to the structural parameters of four terahertz metamaterial sensor structures are calculated; the VBA script in CST is converted into a language that MATLAB can recognize to realize the calculation of the absorption curve. The absorption curve is discretized into 201 points, and the structural parameters and absorption curves of the terahertz metamaterial sensor are merged in the same csv file. In this sample data set, 3000 sets of structural parameters and absorption curves are calculated for each terahertz metamaterial sensor structure, with a total of 12000*206 dimensional data, of which 12000*201 is the absorption curve of the four terahertz metamaterial sensor structures, the 202nd column is the type code of the four terahertz metamaterial sensor structures, and the 203rd-206th columns are the structural parameters of the four terahertz metamaterial sensor structures. In this sample data set, 80% of the data is used for model network training, and 20% of the data is used for model testing; The reverse design model degree of freedom of the multi-structure terahertz metamaterial sensor of this embodiment is higher than that of a single structure. The parameter distribution ranges of the four terahertz metamaterial sensor structures are shown in Table 1: Table 1. Design parameter distribution of four terahertz metamaterial sensors

[0026] Generator and discriminator model construction and training This embodiment combines the advantages of SAGAN, CGAN and WGAN algorithms to construct a reverse design model of multi-structure terahertz metamaterial sensors based on the SACW-GAN algorithm. The working principle of this model is: first train the generator, add conditional information to the generated image data part, generate a structure that conforms to the label content, and finally introduce a self-attention mechanism module between the two last layers of the network to improve the perception and expression capabilities of the network. The discriminator evaluates the similarity between the real sample and the generated sample by calculating the Wasserstein distance between them. Then use the back propagation algorithm to reversely adjust the parameters of the generator and the discriminator to minimize or maximize the Wasserstein distance to achieve more accurate generation results.

[0027] Real sample data y and pseudo sample data The Wasserstein distance between is expressed as:

[0028] in, , , represents the absorption curve extracted from the training data, It represents the absorption curve dataset generated by the generator, E represents the expectation, and D represents the Wasserstein distance.

[0029] Figure 2 The generative adversarial network framework and model training process based on the SACW-GAN algorithm are shown. The input is a cascade of a conditional tensor (1*1*201) composed of an absorption curve and a randomly generated noise vector (1*1*100) to form a conditional noise (1*1*301) passed to the generator to generate enough pseudo sample data (64*64*2). The input real sample data is flattened into a 3D tensor (64*64*201*2), and the discriminator calculates the Wasserstein distance between the real sample data and the pseudo sample data to realize the reverse design of the terahertz metamaterial sensor structure.

[0030] In the training process of the generative adversarial network model based on the SACW-GAN algorithm, the goal of the discriminator is to learn the data feature adjustment weights by maximizing the Wassertein distance between the real sample and the generated sample. The Wassertein distance measures the degree of difference between the real sample and the generated sample. Through the calculation and analysis of this distance, the generator can generate pseudo data that is closer to the real sample. In order to increase the stability of the training and make the results closer to the real data, the generator and the discriminator are trained alternately. The discriminator is updated 4 times per iteration, and the generator is updated once.

[0031] The generative network of the model is mainly used to generate terahertz metamaterial structures that meet the expected absorption spectrum. The network structure is as follows Figure 3 As shown in the figure. The input layer of the generator is 201-dimensional absorption spectrum data and 100-dimensional random noise. The absorption spectrum is connected to the noise through the concat layer, and then it passes through 6 deconvolution layers with nodes of 1024, 512, 512, 256, 256, and 128 respectively, converting the two-dimensional data into three-dimensional data. In the last two deconvolution layers, the self-attention mechanism is added, the nodes are 256 and 128 respectively, the activation function uses the Sigmoid function, and the output is 64*64*1 image data.

[0032] The discriminator of the model is used to predict whether the input data is real data or generated data. The network structure is as follows Figure 4 As shown in the figure, the real sample data and the pseudo sample data generated by the generator are used as the input of the discriminator. The input data passes through 5 deconvolution layers to convert the image data into digital features. Similarly, the self-attention mechanism is added to the last two layers of the network structure, with 256 and 128 nodes respectively. The activation function selects the tanh function, and the discriminant output is 0 or 1 for binary classification.

[0033] The 12,000 pairs of metamaterial structural parameters and absorption line data were input into the constructed model for training. They were divided in a 4:1 ratio, with 9,600 pairs of data used for training and the rest used as test data. The generator and discriminator were trained at the same time. The learning rate was initially set to 0.001, the number of epochs was set to 8,000, the batch-size was set to 64, and Aadm was set as the optimizer of the model. The above parameters were then adjusted according to the mean square error to optimize the model and determine the optimal model structure.

[0034] The Loss value of the generator gradually stabilizes as the number of iterations increases. The images generated by 8000 iterations gradually become clearer, proving that the generator after training can generate images of terahertz metamaterial structures that meet the conditions. The discriminator parameter optimization approaches 0 after 10,000 iterations, but the convergence speed is faster than that of the generator, because the number of training times for the generator is 1 / 4 of the number of training times for the discriminator.

[0035] Design and training of forward neural networks In order to speed up the verification process of the reverse design model results, a deep neural network model is used to realize the forward prediction of the absorption curve. The model input is the structural parameters of four terahertz metamaterial sensors, and the output is the absorption curve. Before model training, it is still necessary to determine the hyperparameters of the model, optimize the number of network layers, number of neuron nodes and batch size of the model, and finally select 4 network layers, 50, 300, 300, 200 neural network nodes, and 128 batch size. The hidden layer still uses the RELU function, the output layer uses the Sigmoid function, the loss function uses the MSE, and the optimizer uses the Adam.

[0036] The 12,000 pairs of metamaterial structure parameters and generated absorption line data are divided into three parts in a ratio of 8:1:1, of which 80% are training data sets for training the model, 10% are validation sets for adjusting the parameters of the forward neural network, and 10% are test sets for evaluating the prediction performance of the model. The model iterations are set to 3,000 times, and the learning rate is selected as 0.001 to learn the tiny features in the data. The loss function of the model is as follows: Figure 5 As shown, the mean square error is 0.015.

[0037] In each of the four terahertz metamaterial sensors, a structure is randomly selected to verify the effectiveness of the network, and the absorption curve calculated by the CST full-wave simulation is compared with the absorption curve predicted by the forward neural network, such as Figure 6 The results show that the absorption curves of the four terahertz metamaterial sensors calculated by CST are roughly consistent with the absorption curves predicted by the model, with a high degree of fit, proving that the model can replace CST to verify the results of the reverse design model.

[0038] 1,000 sets of data were used for testing, and the absorption curve predicted by the forward neural network was compared with the absorption curve simulated by CST. The results showed that the forward neural network can replace the full-wave numerical calculation of CST to a certain extent. After the adversarial network generated the structural parameters, the forward neural network calculated the absorption curve in no more than 2 seconds. The model shortened the result verification time to a certain extent.

[0039] Results and Analysis The 1200 absorption curves of the test set were input into the trained improved generative adversarial network model to analyze, verify and evaluate the prediction performance of SACW-GAN. The 16 groups of absorption curves were input into the trained SACW-GAN model to generate the structural parameters of the metamaterial sensor that met the conditions, as shown in Table 2.

[0040] Table 2 Structural parameters of terahertz metamaterial sensor with target absorption curve

[0041] The absorption spectrum of the metamaterial is generated by calculating the reverse model through the forward neural network and compared with the target absorption spectrum, such as Figure 7 The black dotted line graph is the input absorption curve in the reverse design model, and the red line graph is the absorption curve predicted by the forward neural network for the generated metamaterial structure. When the input random absorption curve is single-peaked, the constructed network model can output the metamaterial sensor structure parameters that meet the conditions; however, when the absorption curve is double-peaked or multi-peaked, the absorption curve generated by the structure predicted by the model has the same trend as the target absorption curve, but there is a small difference in the absorption rate, which may be due to the problem of too large spacing during data sampling. The results show that the model can well predict the metamaterial structure parameters that meet the target conditions. Therefore, the model can be used for the structural design of terahertz metamaterial sensors, and the results can be verified by the forward neural network. The verified devices can be prepared and subsequent research work can be carried out. The optimal terahertz metamaterial sensor structure parameter determination process only takes a few seconds or even milliseconds, which speeds up the generation cycle of terahertz metamaterial sensors.

[0042] Based on the same inventive concept, the embodiment of the present application also provides a multi-structure terahertz metamaterial sensor reverse design system based on an improved conditional generative network. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above method, so the specific limitations of one or more embodiments of the multi-structure terahertz metamaterial sensor reverse design system based on an improved conditional generative network provided below can be referred to the limitations of the method above, and will not be repeated here.

[0043] In an exemplary embodiment, a multi-structure terahertz metamaterial sensor reverse design system based on an improved conditional generative network is provided, comprising: A parameter and data set construction module is used to: construct a terahertz metamaterial sensor structure, and based on the constructed terahertz metamaterial sensor structure, establish a sample data set of sensor structure parameters and corresponding absorption curves; The reverse design model module is used to: combine SAGAN, CGAN and WGAN algorithms to construct and train a reverse design model for multi-structure terahertz metamaterial sensors; The multi-structure terahertz metamaterial sensor reverse design model is a generative adversarial network based on the SACW-GAN algorithm; In the training process of the generative adversarial network, the input is a conditional tensor consisting of an absorption curve and a randomly generated noise vector cascaded to form conditional noise passed to the generator to generate sufficient pseudo sample data; the discriminator calculates the Wasserstein distance between the real sample data and the pseudo sample data to achieve the reverse design of the terahertz metamaterial sensor structure; The forward neural network model module is used to construct and train a forward neural network model to verify whether the terahertz metamaterial sensor structure designed by the generative adversarial network meets the design requirements.

[0044] Each module in the above system can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module above.

[0045] In an exemplary embodiment, a computer device is provided, which may be a terminal. The computer device further includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected via a system bus, and the communication interface, the display unit, and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, the steps of the reverse design method of a multi-structure terahertz metamaterial sensor based on an improved conditional generative network are implemented. The display unit of the computer device is used to form a visually visible picture, which may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0046] Those skilled in the art will appreciate that the above structure is only a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different arrangement of components.

[0047] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a reverse design method of a multi-structure terahertz metamaterial sensor based on an improved conditional generative network are implemented.

[0048] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0049] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0050] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A reverse design method for multi-structure terahertz metamaterial sensors based on improved conditional generative networks, characterized in that: The steps include: (1) Construct a terahertz metamaterial sensor structure, and based on the constructed terahertz metamaterial sensor structure, establish a sample data set of sensor structure parameters and corresponding absorption curves; (2) Combining SAGAN, CGAN and WGAN algorithms, construct and train a reverse design model for multi-structure terahertz metamaterial sensors; The multi-structure terahertz metamaterial sensor reverse design model is a generative adversarial network based on the SACW-GAN algorithm; In the training process of the generative adversarial network, the input is a conditional tensor consisting of an absorption curve and a randomly generated noise vector cascaded to form conditional noise passed to the generator to generate sufficient pseudo sample data; the discriminator calculates the Wasserstein distance between the real sample data and the pseudo sample data to achieve the reverse design of the terahertz metamaterial sensor structure; (3) Construct and train a forward neural network model to verify whether the terahertz metamaterial sensor structure designed by the generative adversarial network meets the design requirements.

2. The method for reverse designing a multi-structure terahertz metamaterial sensor based on an improved conditional generative network according to claim 1, characterized in that: Real sample data y and pseudo sample data The Wasserstein distance between is expressed as:

3. Among them, , , represents the absorption curve extracted from the training data, It represents the absorption curve dataset generated by the generator, E represents the expectation, and D represents the Wasserstein distance.

4. The method for reverse designing a multi-structure terahertz metamaterial sensor based on an improved conditional generative network according to claim 1, characterized in that: The input layer of the generator is absorption spectrum line data and random noise; the absorption spectrum line is connected to the noise through the concat layer, and then passes through 6 deconvolution layers in sequence to convert the two-dimensional data into three-dimensional data; in the last two deconvolution layers, a self-attention mechanism is added; the activation function uses the Sigmoid function.

5. The method for reverse designing a multi-structure terahertz metamaterial sensor based on an improved conditional generative network according to claim 1, characterized in that: The real sample data and the pseudo sample data generated by the generator are used as the input of the discriminator. The input data passes through 5 deconvolution layers in sequence to convert the image data into digital features. In the last two deconvolution layers, the self-attention mechanism is added; the tanh function is selected as the activation function.

6. The method for reverse designing a multi-structure terahertz metamaterial sensor based on an improved conditional generative network according to claim 1, characterized in that: The constructed terahertz metamaterial sensor structures include four types: circular open ring type, square open ring type, H type and F type; The four terahertz metamaterial sensor structures are all composed of three layers. The first layer is a metal resonance layer made of gold, the second layer is a dielectric layer made of polytetrafluoroethylene, and the third layer is a substrate layer made of gold. The terahertz metamaterial sensor structure is represented by four parameters: Unit period u, resonator line width w of the metal resonator layer, thickness h of the dielectric layer, resonator length a of the metal resonator layer and thickness of the substrate layer 0.2 um; For the unit period u, the circular open ring type is marked as 78-82um, the square open ring type is marked as 75-84um, the H type is marked as 65-74um, and the F type is marked as 75-84um; For the dielectric layer thickness h, the circular open ring type is marked as 28-32um, the square open ring type is marked as 35-44um, the H type is marked as 20-29um, and the F type is marked as 55-64um; For the resonator length a, the circular open ring type is marked as 16-20um, the square open ring type is marked as 35-44um, the H type is marked as 20-29um, and the F type is marked as 25-34um; For the resonator line width w, the circular open ring type is marked as 2-5um, the square open ring type is marked as 3-5um, the H type is marked as 3-5um, and the F type is marked as 3-5um.

7. The method for reverse designing a multi-structure terahertz metamaterial sensor based on an improved conditional generative network according to claim 1, characterized in that: The sensor structure parameters and the corresponding absorption curve sample data set established are: The resonators of the four terahertz metamaterial sensor structures are numbered, where the circular open ring type is marked as "1", the square open ring type is marked as "2", the H type is marked as "3", and the F type is marked as "4"; The sampling frequencies of the absorption curves of the four terahertz metamaterial sensor structures were set to 0.5-1.5 THz, 1.5-2.5 THz, 2.5-3.5 THz, and 0.5-1.5 THz, respectively. The absorption curves were sampled at 201 points within the above frequency ranges. In this sample data set, 3000 sets of structural parameters and absorption curves are calculated for each terahertz metamaterial sensor structure, with a total of 12000*206 dimensional data, of which 12000*201 is the absorption curves of the four terahertz metamaterial sensor structures, the 202nd column is the type code of the four terahertz metamaterial sensor structures, and the 203rd-206th columns are the structural parameters of the four terahertz metamaterial sensor structures.

8. A multi-structure terahertz metamaterial sensor reverse design system based on an improved conditional generative network, characterized in that: include: A parameter and data set construction module is used to: construct a terahertz metamaterial sensor structure, and based on the constructed terahertz metamaterial sensor structure, establish a sample data set of sensor structure parameters and corresponding absorption curves; The reverse design model module is used to: combine SAGAN, CGAN and WGAN algorithms to construct and train a reverse design model for multi-structure terahertz metamaterial sensors; The multi-structure terahertz metamaterial sensor reverse design model is a generative adversarial network based on the SACW-GAN algorithm; In the training process of the generative adversarial network, the input is a conditional tensor consisting of an absorption curve and a randomly generated noise vector cascaded to form conditional noise passed to the generator to generate sufficient pseudo sample data; the discriminator calculates the Wasserstein distance between the real sample data and the pseudo sample data to achieve the reverse design of the terahertz metamaterial sensor structure; The forward neural network model module is used to construct and train a forward neural network model to verify whether the terahertz metamaterial sensor structure designed by the generative adversarial network meets the design requirements.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for reverse design of a multi-structure terahertz metamaterial sensor based on an improved conditional generative network according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for reverse design of a multi-structure terahertz metamaterial sensor based on an improved conditional generative network according to any one of claims 1 to 6 are implemented.

Citation Information

Cited By

  • Terahertz metamaterial enhanced spectrum detection system

    CN120404648A