Multi-mode multifunctional composite metasurface design method, device and medium

Abstract

The invention relates to a multi-modal multifunctional composite metasurface design method and device and a medium, and the method comprises the steps: carrying out the parameterized sampling of the key size of a unit structure through Latin hypercube sampling, and generating a training sample set; electromagnetic simulation is carried out on the sampled structure to obtain two-state electromagnetic response data, and a mixed variable database is obtained; according to the target response, generating a plurality of groups of candidate structures through a generator in the improved conditional variation auto-encoder generative adversarial network, screening the candidate structures through a predictor, calculating a prediction comprehensive error, and selecting a preset number of candidate structures with the minimum prediction comprehensive error to perform full-wave simulation verification; and obtaining a simulation double-state response, calculating a simulation comprehensive error, and judging the current candidate structure according to a preset comprehensive error threshold value. Compared with the prior art, the method has the advantages of multi-state collaboration, low calculation cost, high degree of freedom and the like.

Description

Technical Field

[0001] The present invention relates to the field of metasurface inverse design, and more particularly to a multi-modal multi-functional composite metasurface design method, device, and medium. Background Art

[0002] In recent years, significant progress has been made in metasurface inverse design technology, but there are still many technical bottlenecks. The existing technologies can be mainly divided into the following three development stages: In the traditional optimization algorithm stage, the research mainly relied on heuristic methods such as genetic algorithms and particle swarm optimization, and achieved electromagnetic response matching by iteratively searching geometric parameters. However, such methods require manual setting of empirical parameters, have a long design cycle, and are difficult to meet the requirements of multi-state function switching. Due to the lack of theoretical guidance, the optimization process is prone to falling into local optima, resulting in a significant deviation between the generated structure and the target response. Especially when dealing with scenarios that require continuous phase modulation such as polarization conversion, it is difficult to ensure the modulation accuracy and structural stability. With the introduction of machine learning technology, an inverse design method based on artificial neural networks has been proposed to improve the design efficiency by establishing the mapping relationship between the target response and the structural parameters. However, this method still has obvious limitations. First, it can only optimize for a single working state independently. If multi-state function switching is required, such as switching between polarization conversion and wave absorption functions, separate models need to be trained for each state, resulting in waste of computing resources and extension of the design cycle. Second, although the discrete variable method can improve the structural diversity, the computational complexity increases exponentially when the design scale expands, while the continuous variable method is easily restricted by local optima and is difficult to balance the requirements of high-degree-of-freedom design and high-precision modulation, especially in the design of complex asymmetric structures. In addition, pure continuous variable methods, such as geometric parameter optimization, are prone to falling into local optima and are difficult to handle complex structures, such as asymmetric opening designs. The electromagnetic response accuracy of the generated structures is limited. When realizing dual-band wave absorption by continuous variable optimization, it is difficult to meet the response requirements of both bands at the same time. Although recent explorations of multi-objective optimization algorithms have tried to balance multiple performance indicators, they still face severe challenges in high-degree-of-freedom scenarios. On the one hand, the same target response may correspond to multiple effective structures, and traditional methods are difficult to balance diversity and convergence, resulting in poor consistency of the electromagnetic responses of the generated structures. On the other hand, the existing dataset construction relies on random sampling, and a large number of invalid samples lead to high simulation costs and lack of in-depth constraints on the physical mechanism of electromagnetic responses, making the physical rationality and interpretability of the generated structures insufficient and difficult to be directly applied to the design of actual devices.

[0003] Chinese Patent Application CN119249869A discloses a method for optimizing the design of metasurfaces based on reinforcement learning. It constructs a neural network to map states to action values, optimizes decisions through the interaction between the agent and the environment, and uses an experience replay pool and a dual network to enhance stability. However, this technical solution only optimizes for a single target and cannot jointly optimize multiple working states. It requires repeated training of the model, resulting in low efficiency. Moreover, it relies on real-time electromagnetic simulation to provide reward feedback, leading to a long training cycle and high resource consumption. Therefore, how to achieve the balance of multi-state collaborative design ability, design freedom, and precision, and construct a dataset guided by theory to achieve efficient, accurate, and physically reasonable multifunctional metasurface design is a technical problem that needs to be solved. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned existing technologies and provide a multi-modal multifunctional composite metasurface design method, device, and medium. Through the definition of mixed variable parameters, the joint optimization of dual-state responses, the construction of a dataset guided by theory, and the multi-candidate generation and screening mechanism, an efficient, high-precision, and physically reasonable metasurface design is achieved.

[0005] The purpose of the present invention can be realized by the following technical solutions:

[0006] According to one aspect of the present invention, a multi-modal multifunctional composite metasurface design method is provided, and the specific steps include:

[0007] S1. Parametric sampling of the key dimensions of the unit structure is performed through Latin hypercube sampling to generate a training sample set; electromagnetic simulation is performed on the sampled structure to obtain dual-state electromagnetic response data, and the data is subjected to format conversion and enhancement processing to obtain a mixed variable database;

[0008] S2. According to the mixed variable data and the target response, the generator in the improved conditional variational autoencoder generative adversarial network generates several groups of candidate structures, and the predictor screens the candidate structures to obtain the predicted dual-state response, calculates the predicted comprehensive error, and selects a preset number of candidate structures with the smallest predicted comprehensive error for full-wave simulation verification;

[0009] S3. The simulated dual-state response is obtained through full-wave simulation, the simulated comprehensive error is calculated, and the current candidate structure is judged according to the preset comprehensive error threshold. If the simulated comprehensive error is less than the comprehensive error threshold, the corresponding structure is output.

[0010] Furthermore, the unit structure in S1 includes a polarization conversion unit designed based on the Jones matrix theory, an absorbing unit optimized by combining the Fabry-Pérot cavity theory, and a basic structure unit.

[0011] Further, the mixed variable data in S1 includes a discrete design matrix and a continuous angle parameter, where the discrete design matrix is the material distribution state of the metasurface unit, and the continuous angle parameter is the continuous variation range of the opening angle of the metasurface unit.

[0012] Further, the improved conditional variational autoencoder generative adversarial network in S2 includes an encoder, a generator, a predictor, and a discriminator, which are trained and optimized through a comprehensive loss function; wherein, the encoder inputs the mixed variable structure parameters and the bistatic target response, and outputs a latent variable that conforms to a normal distribution; the generator generates candidate structure parameters based on the latent variable and the target response; the predictor predicts the bistatic response of the candidate structure, and the discriminator discriminates the authenticity of the generated structure and the real structure.

[0013] Further, the comprehensive loss function includes a reconstruction loss, a relative entropy loss, a predicted response matching loss, and a feature matching loss; wherein, the reconstruction loss is used to constrain the parameter consistency between the generated structure and the real structure, and the parameters include the discrete design matrix and the continuous angle parameter, the relative entropy loss is used to constrain the latent variable distribution to be close to the Gaussian prior distribution, the predicted response matching loss is used to constrain the predicted bistatic response of the generated structure to be consistent with the target response, and the feature matching loss is used to constrain the deep features of the generated structure and the real structure to be similar.

[0014] Further, when generating the candidate structure in S2, the latent variable is obtained through random sampling, and the discrete design matrix is 30×30 pixels, and the continuous angle parameter coverage range is from 30° to 320°.

[0015] Further, the predicted comprehensive error in S2 includes the root mean square error between the predicted bistatic responses corresponding to the current candidate structure in the two states and the corresponding target responses respectively.

[0016] Further, the simulation comprehensive error in S3 includes the root mean square error between the simulated bistatic responses corresponding to the current candidate structure in the two states and the corresponding target responses respectively.

[0017] According to the second aspect of the present invention, there is provided an electronic device, including a memory and a processor, wherein a computer program is stored on the memory, and when the processor executes the program, the method described above is implemented.

[0018] According to the third aspect of the present invention, there is provided a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the method described above is implemented.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] (1) Improve the multi-state collaborative design ability: Through the bistate response joint optimization framework, the multi-state functional requirements can be met through one-time training, avoiding the complex process caused by independently training models for each state. By integrating multi-state information in the latent variable space, repeated training is avoided, significantly shortening the design cycle and improving the efficiency and consistency of multi-modal metasurface design.

[0021] (2) Co-optimize the design freedom and regulation accuracy: Adopt a mixed variable parameter definition, namely a discrete design matrix and continuous angle parameters. The discrete matrix supports flexible material distribution, and the continuous angle parameters enable fine regulation of the opening direction. Combining the Jones matrix theory to constrain the phase symmetry and the Fabry-Pérot cavity theory to optimize the absorbing unit not only ensures the design ability of high-degree-of-freedom complex structures but also improves the electromagnetic response accuracy and physical rationality of the generated structures through physical theory constraints, solving the contradiction in traditional single-variable methods where discrete variables have high complexity and continuous variables are prone to falling into local optima.

[0022] (3) Reduce the computational cost: Generate training samples through theory-guided basic unit selection and Latin hypercube sampling, and improve the effectiveness of the dataset through data augmentation. In addition, the multi-candidate generation mechanism and predictor pre-screening effectively reduce the number of simulation validations of invalid samples, reducing the consumption of computing resources while ensuring design reliability and significantly improving the design efficiency and applicability. Brief Description of the Drawings

[0023] Figure 1 It is a flow framework diagram of a multi-modal multi-functional composite metasurface design method;

[0024] Figure 2 It is a structural diagram of an improved conditional variational autoencoder generative adversarial network. Detailed Embodiments

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] As Figure 1 shown, it is a multi-modal multi-functional composite metasurface design method, and the specific steps include:

[0027] S1. Parametrically sample the key dimensions of the unit structure through Latin hypercube sampling to generate a training sample set; perform electromagnetic simulation on the sampled structure to obtain bistate electromagnetic response data, and perform format conversion and enhancement processing on the data to obtain a mixed variable database;

[0028] S2. According to the mixed variable data and the target response, several groups of candidate structures are generated by the generator in the improved conditional variational autoencoder generative adversarial network, and the candidate structures are screened by the predictor to obtain the predicted bistatic response, calculate the predicted comprehensive error, and select a preset number of candidate structures with the smallest predicted comprehensive error for full-wave simulation verification;

[0029] S3. Obtain the simulated bistatic response through full-wave simulation, calculate the simulated comprehensive error, and judge the current candidate structure according to the preset comprehensive error threshold. If the simulated comprehensive error is less than the comprehensive error threshold, output the corresponding structure.

[0030] In S1, the unit structure includes a polarization conversion unit designed based on the Jones matrix theory, an absorbing unit optimized by combining the Fabry-Pérot cavity theory, and a basic structure unit. Among them, the polarization conversion unit is used to ensure phase symmetry, and the absorbing unit is used to enhance resonance coupling. Latin hypercube sampling is used to parametrically sample the key dimensions of the unit structure, generating seven thousand training samples to uniformly cover the parameter space and improve sample diversity. The mixed variable data includes a discrete design matrix and continuous angular parameters, where the discrete design matrix is the material distribution state of the metasurface unit, and the continuous angular parameters are the continuous variation range of the opening angle of the metasurface unit. The bistatic electromagnetic response is the non-conductive state response and the conductive state response. The conversion and enhancement of the data include converting the simulation curve into a square wave format, setting it to 1 when the response is greater than or equal to 0.9, and setting it to 0.1 when it is less than 0.1, as well as symmetric transformations such as mirroring and central symmetry to generate a 3-fold augmented dataset, improving the generalization ability of the model and reducing the risk of overfitting.

[0031] In S2, the improved conditional variational autoencoder generative adversarial network includes an encoder, a generator, a predictor, and a discriminator, and is trained and optimized through a comprehensive loss function. The specific structure is as Figure 2 shown.

[0032] Among them, the encoder inputs the mixed variable structure parameters x = [x m , x a and the bistatic target response y = [y s1 , y s2 . x m is the surface pixel distribution matrix, x a is the opening angle of the surface pixel distribution matrix. The metal sheet within the opening angle range will be modified to VO2 material to realize the distribution definition of the phase change material. y s1 is the electromagnetic response corresponding to the input structure in the first state; the parameter y s2Represents the electromagnetic response corresponding to the input structure in the second state. The output is a latent variable whose latent variable conforms to a normal distribution, indicating that the design parameters are compressed into a low-dimensional sample space and conform to a normal distribution.

[0033] The generator generates candidate structural parameters based on the latent variable and the target response. The latent variable is randomly sampled between 0 and 1. The diversified design generation lies in integrating the low-dimensional latent vector latent variable into the predefined target parameters. Through the differentiation strategy, the "one-to-many" mapping problem is fundamentally solved.

[0034] The predictor predicts the bistate response of the candidate structure generated by the generator to obtain the predicted bistate response corresponding to the bistate. The discriminator discriminates the authenticity of the generated structure and the real structure. Its input is the real sample or the predicted structure. When the input is the real sample, it is the training sample set, and the corresponding label is the all-one matrix. When the input is the predicted structure, the corresponding label is the all-zero matrix.

[0035] The comprehensive loss function of the model includes reconstruction loss, relative entropy loss, predicted response matching loss, and feature matching loss. Among them, the reconstruction loss is used to constrain the parameter consistency between the generated structure and the real structure. The parameters include the discrete design matrix and the continuous angle parameter. The relative entropy loss is used to constrain the latent variable distribution to be close to the Gaussian prior distribution. The predicted response matching loss is used to constrain the predicted bistate response of the generated structure to be consistent with the target response. The feature matching loss is used to constrain the deep features of the generated structure and the real structure to be similar.

[0036] When generating the candidate structure in S2, the latent variable is obtained by random sampling, the discrete design matrix is 30×30 pixels, and the coverage range of the continuous angle parameter is 30° to 320°.

[0037] The predicted comprehensive error in S2 includes the root mean square error between the predicted bistate response corresponding to the current candidate structure in the bistate and the corresponding target response. That is, the generator is based on the target response y = [y s1 , y s2 and the random latent variable z ∼ N(0, I), and generates 15 groups of candidate structures x′ = [x′ m , x′ a through the deconvolution network, where the predicted discrete matrix x′ m is 30×30 pixels, and the predicted continuous angle parameter x a ′ covers 30° - 320°. Utilize the continuity and diversity of the latent space to explore potential effective designs, break through the limitations of traditional single-solution designs, and ensure full exploration of the design space. The predictor screens the candidate structures. Input the candidate structure x′, and the predictor outputs the bistate response y′ = [y′ s1 , y′ s2, calculate the comprehensive error MSE pre = MSE((y' s1 , y s1 ) + MSE(y' s2 , y s2 ); where y' s1 is the predicted response corresponding to the prediction structure in the first state, and y s1 is the target response of the first state. Among them, y' s2 is the predicted response corresponding to the prediction structure in the second state, and y s2 is the target response of the second state. Select the top 3 groups of structures with the smallest error to enter the full-wave simulation verification. Eliminate 75% of the invalid structures through the predictor, reduce the number of full-wave simulations from 15 times to 3 times, and reduce the calculation cost.

[0038] In S3, the simulation comprehensive error includes the root mean square error between the simulated dual-state responses corresponding to the current candidate structure in the dual state and the corresponding target responses respectively. Through the simulated dual-state response y sim = [y s1_sim , y s2_sim , calculate the simulation comprehensive error MSE sim = MSE(y s1_sim , y s1 ) + MSE(y s2_sim , y s2 ). Among them, y s1_sim is the simulated response corresponding to the prediction structure in the first state, and y s2_sim is the simulated response corresponding to the prediction structure in the second state; through the root mean square error MSE(y s1_sim , y s1 ) calculation, obtain the gap with the target response y s1 of the first state; through the root mean square error MSE(y s2_sim , y s2 ) calculation, obtain the gap with the target response y s2 of the second state. If the comprehensive error MSE sim < 0.02, the design passes the verification; otherwise, regenerate the candidate structure. This verification process ensures that the electromagnetic response of the generated structure deviates from the target by less than the engineering acceptable range, guaranteeing the practicability of the design.

[0039] Taking the electromagnetic responses of the metasurface in two different states as the targets, explicitly integrating the multi-state functional requirements into the loss function, and realizing the joint optimization of the bistate responses by sharing the structural parameters, thus avoiding time-consuming processes such as the repeated iteration of the responses for each state and the alignment of structural parameters, and significantly improving the optimization efficiency. In addition, the idea of sampling the latent space of the improved conditional variational autoencoder generative adversarial network is also utilized. A latent variable is added to the input target to distinguish designs with the same type of response but different corresponding structures in the high-dimensional space, thereby avoiding the one-to-many problem that may cause the generative network not to converge in the inverse design. The discrete matrix and continuous angle parameters are jointly encoded to balance the structural diversity and the generation accuracy. The discrete matrix enables flexible regulation of the material layout, where the 0-2 states respectively represent the distributions of air, VO2, and PEC, and the continuous parameter supports the continuous change of the opening angle of the phase control material from 30° to 320°. The phase symmetry is constrained by the Jones matrix theory, and the absorbing unit is optimized in combination with the Fabry-Pérot cavity theory to ensure the physical rationality of the generated structure. Therefore, the algorithm supports complex electromagnetic function designs while making the generated structure comply with the constraints of theoretical knowledge. Through the multi-candidate generation and performance pre-screening mechanism, the generator generates multiple groups of candidate structures through random sampling of the latent variable, and the predictor evaluates their performance and screens the optimal solution. The multi-candidate generation covers the latent space to avoid falling into local optimal solutions, and the true simulation error and the target error of the optimal candidate are lower than 0.02. The pre-screening mechanism not only ensures that the generated structure meets the target performance, but also improves the reliability and data utilization efficiency of the algorithm. The predictor learns the structure-response mapping relationship during the training process and can be used for candidate screening without additional training, significantly reducing data redundancy and computational costs.

[0040] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the described modules can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0041] The electronic device of the present invention includes a central processing unit (CPU), which can execute various appropriate actions and processes according to the computer program instructions stored in the read-only memory (ROM) or the computer program instructions loaded from the storage unit into the random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other through a bus. The input / output (I / O) interface is also connected to the bus.

[0042] Multiple components in the device are connected to the I / O interface, including: an input unit, such as a keyboard, a mouse, etc.; an output unit, such as various types of displays, speakers, etc.; a storage unit, such as a disk, an optical disc, etc.; and a communication unit, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit allows the device to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks. The processing unit executes the various methods and processes described above, such as the method of the present invention. For example, in some embodiments, the method of the present invention can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as the storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device via the ROM and / or the communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of the method of the present invention described above can be executed. Alternatively, in other embodiments, the CPU can be configured to execute the method of the present invention by any other suitable means (e.g., by means of firmware).

[0043] The functions described above herein can be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that can be used include: Field Programmable Gate Arrays (FPGA), Application Specific Integrated Circuits (ASIC), Application Specific Standard Products (ASSP), System on a Chip (SOC), Complex Programmable Logic Devices (CPLD), and so on.

[0044] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or a controller of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, such that when the program codes are executed by the processor or the controller, the functions / operations specified in the flowchart and / or the block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, executed partially on the machine and partially on a remote machine as an independent software package, or executed entirely on a remote machine or a server.

[0045] In the context of the present invention, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include electrical connections based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0046] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A multi-modal and multi-functional composite metasurface design method, characterized in that, The specific steps include: S1. Parametric sampling is performed on the critical dimensions of the unit structure through Latin hypercube sampling to generate a training sample set; electromagnetic simulation is performed on the sampled structure to obtain bistatic electromagnetic response data, and the data is subjected to format conversion and enhancement processing to obtain a mixed variable database; S2. According to the mixed variable data and the target response, the generator in the improved conditional variational autoencoder generative adversarial network generates several groups of candidate structures, and the predictor screens the candidate structures to obtain the predicted bistatic response, calculates the predicted comprehensive error, and selects a preset number of candidate structures with the smallest predicted comprehensive error for full-wave simulation verification; S3. The simulated bistatic response is obtained through full-wave simulation, the simulated comprehensive error is calculated, and the current candidate structure is judged according to the preset comprehensive error threshold. If the simulated comprehensive error is less than the comprehensive error threshold, the corresponding structure is output.

2. A multimodal and multifunctional composite metasurface design method according to claim 1, characterized in that The unit structure in S1 includes a polarization conversion unit designed based on the Jones matrix theory, an absorbing unit optimized by combining the Fabry-Pérot cavity theory, and a basic structure unit.

3. A multimodal and multifunctional composite metasurface design method according to claim 1, characterized in that, The mixed variable data in S1 includes a discrete design matrix and continuous angle parameters. The discrete design matrix is the material distribution state of the metasurface unit, and the continuous angle parameter is the continuous change range of the opening angle of the metasurface unit.

4. A multimodal and multifunctional composite metasurface design method according to claim 1, characterized in that The improved conditional variational autoencoder generative adversarial network in S2 includes an encoder, a generator, a predictor, and a discriminator, and is trained and optimized through a comprehensive loss function; among them, the encoder inputs the mixed variable structure parameters and the bistatic target response, and outputs a latent variable that conforms to a normal distribution; the generator generates candidate structure parameters based on the latent variable and the target response; the predictor predicts the bistatic response of the candidate structure, and the discriminator discriminates the authenticity of the generated structure and the real structure.

5. A multimodal and multifunctional composite metasurface design method according to claim 4, characterized in that The comprehensive loss function includes a reconstruction loss, a relative entropy loss, a predicted response matching loss, and a feature matching loss; among them, the reconstruction loss is used to constrain the parameter consistency between the generated structure and the real structure, and the parameters include the discrete design matrix and the continuous angle parameters. The relative entropy loss is used to constrain the latent variable distribution to be close to the Gaussian prior distribution. The predicted response matching loss is used to constrain the predicted bistatic response of the generated structure to be consistent with the target response, and the feature matching loss is used to constrain the deep features of the generated structure and the real structure to be similar.

6. A multimodal multifunctional composite metasurface design method according to claim 1, characterized in that When generating the candidate structure in S2, the latent variable is obtained through random sampling, the discrete design matrix is 30×30 pixels, and the continuous angle parameter coverage range is from 30° to 320°.

7. A multimodal and multifunctional composite metasurface design method according to claim 1, characterized in that The predicted comprehensive error in S2 includes the root mean square error between the predicted bistatic responses corresponding to the current candidate structure in the two states and the corresponding target responses respectively.

8. A multimodal and multifunctional composite metasurface design method according to claim 1, characterized in that, The simulated comprehensive error in S3 includes the root mean square error between the simulated bistatic responses corresponding to the current candidate structure in the two states and the corresponding target responses respectively.

9. An electronic device, comprising a memory and a processor, wherein a computer program is stored on the memory, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, on which a computer program is stored, characterized in that, When the program is executed by the processor, it implements the method according to any one of claims 1 to 8.

Images