Method, system, device and readable storage medium for constructing freeform metasurface

Abstract

The application discloses a kind of construction methods and systems of free shape metasurface, method includes the following steps: obtaining the electromagnetic wave data that is refracted and reflected by preset metasurface;Pretreatment is carried out, and the actual electric field distribution data of metasurface center plane, the electromagnetic wave phase distribution data in the same plane in far field are obtained, and the matrix tensor of frequency and angle is obtained by discrete processing to incident frequency data and incident angle data;The training and verification of the series neural network model constructed are carried out, and the series neural network model is obtained;Initial phase change cell arrangement parameter is obtained;Actual electric field distribution data and electromagnetic wave phase distribution are combined to verify initial phase change cell arrangement parameter, and optimization is carried out, and the phase change cell arrangement after optimization is obtained, and the optimal metasurface is constructed based on the phase change cell arrangement after optimization.The series neural network model realized by global reverse design can break the limitation of unit coupling and multiple scattering, and improve the stealth effect.

Description

Technical Field

[0001] This invention relates to the field of metasurface technology, and in particular to a method, system, apparatus, and readable storage medium for constructing free-form metasurfaces. Background Technology

[0002] Invisibility, or stealth, has been a long-standing topic throughout human civilization. To realize this dream, humanity has continuously explored and strived for improvement. Especially with the advancement of technology, unprecedented enthusiasm has been ignited in the fields of metamaterials and transformation optics. Researchers have been constantly exploring and refining these technologies. Transformation optics can make objects invisible by guiding the flow of light around them without disturbing their internal regions. The fundamental physics behind this is attributed to the formal invariance of Maxwell's equations: a coordinate transformation can compress the normal free space of a volume into a shell, requiring only the constitutive parameters of the volume and the electromagnetic field. Beyond electromagnetism, transformation optics has now become a fashionable tool, entering fields such as sound, heat flow, and water waves. Theoretically, this method is perfect; however, in experiments, it is limited by the anisotropic and non-uniform composition of large-volume materials. Significant efforts have been devoted to mitigating these limitations, such as bilinear transformation optics with singularity-free features and quasi-conformal transformations for ground-plane cloaks. However, these trade-offs also weaken stealth performance to some extent, making the application scenarios more specific.

[0003] With the development of technology, cloaking or invisibility achieved through metasurfaces has become a mainstream approach. This is achieved by adding a deliberately designed metasurface layer to the hidden region or object surface. The resulting scattered field can be reconstructed into a field similar to a pure background, making it appear to an observer as if the region or object does not exist. Among current technologies, metasurface cloaks are the most researched. Due to their significant advantages such as negligible thickness, simple fabrication, and low loss, metasurface cloaks have been widely welcomed by academia and industry. Building upon this, significant progress has been made in researching metasurface cloaks across different spectra, extending their application from reflection geometry to transmission geometry. Furthermore, by combining active components and intelligent algorithms, it holds the promise of transforming traditional static cloaking methods into dynamic cloaking capable of self-guided responsiveness to constantly changing external stimuli and environments. These advancements are essential for numerous practical applications involving moving objects and dynamic environments.

[0004] Free-form metasurface cloak cloaking, as an emerging technology, has broad application prospects, but it is subject to some inherent limitations. For example, because using complex shapes to design metasurface cloaks encounters many unpredictable scattering events that are difficult to generalize into specific criteria, almost all metasurface cloaks have been proven to be convex. Alternatively, metasurface cloaks typically operate within a narrow angle of incidence, usually less than 35°, and the specific angle depends on the geometry of the hidden object. If the angle of incidence increases, all incident energy will be reflected back to the incident surface and cannot be reflected to the specular direction where it should flow. Or, due to unavoidable metasurface dispersion, the operating bandwidth is very narrow, even single-frequency. Or, when designing metasurface cloaks, each metasurface unit is assumed to be an independent entity, which ignores the potential interactions with adjacent elements that can lead to many scattering uncertainties.

[0005] Due to these limitations, no structure or method with better stealth capabilities has yet emerged. Summary of the Invention

[0006] This invention addresses the shortcomings of existing technologies by providing a method, system, apparatus, and readable storage medium for constructing free-form metasurfaces.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] A method for constructing a free-form metasurface includes the following steps:

[0009] Electromagnetic wave data refracted and reflected by a preset metasurface is acquired. The electromagnetic wave data includes near-field distribution data, far-field distribution data, incident frequency data, and incident angle data. The near-field distribution data is electric field distribution data, and the far-field data is radar cross-sectional area distribution.

[0010] Preprocessing the near-field and far-field distribution data respectively yields the actual electric field distribution data of the metasurface center plane and the electromagnetic wave phase distribution data in the same plane in the far field. Discretizing the incident frequency and incident angle data yields the frequency and angle matrix tensors.

[0011] The tandem neural network model was trained and validated using actual electric field distribution data, electromagnetic wave phase distribution data, and matrix tensors to obtain the tandem neural network model.

[0012] The actual electric field distribution data to be measured, the electromagnetic wave phase distribution, the incident frequency and the incident angle are used as inputs to the series neural network model to obtain the initial phase transition unit arrangement parameters.

[0013] The initial phase transition unit arrangement parameters were verified by combining actual electric field distribution data and electromagnetic wave phase distribution, and the initial phase transition unit arrangement parameters were used as the phase transition unit arrangement parameters of the target metasurface.

[0014] The initial phase change unit arrangement parameters are optimized to obtain an optimized phase change unit arrangement. An optimal metasurface is constructed based on the optimized phase change unit arrangement, wherein the optimized phase change unit arrangement includes metasurface design parameters.

[0015] As one possible implementation, obtaining the optimized phase change unit arrangement and constructing an optimal metasurface based on the optimized phase change unit arrangement includes the following steps:

[0016] The design parameters of the phase transition unit are changed based on relevant characteristics, and the electromagnetic response generated by the metasurface is changed by changing the design parameters to obtain the corresponding electromagnetic response.

[0017] Based on the electromagnetic response comparison with the actual electromagnetic response of the reference plane, the arrangement of the phase change units is adjusted according to the design parameters of the phase change units so that the measured near-field distribution data and far-field distribution data are similar to the field distribution data of the reference plane. The corresponding design parameters of the phase change units are the design parameters of the phase change units in the optimal metasurface.

[0018] As one possible implementation, the cascaded neural network model includes a reverse neural network unit and a forward neural network unit, wherein the output layer of the reverse neural network unit is the input layer of the forward neural network unit;

[0019] The input to the inverse neural network unit is the actual electric field distribution data, electromagnetic wave phase distribution data, and matrix tensor; the output of the inverse neural network unit is the phase transition unit arrangement parameters.

[0020] The input to the forward neural network unit is the phase transition unit arrangement parameters, and the output of the forward neural network unit is the predicted electric field distribution data and electromagnetic wave phase distribution data.

[0021] As one possible implementation, the inverse neural network unit includes a first input layer and a second input layer. The first input layer includes three channels, and the input for each channel is: actual electric field distribution data, incident frequency data, and incident angle data. The second input layer inputs electromagnetic wave phase distribution data.

[0022] The output layer of the forward neural network unit includes three channels, and the output of each channel is the predicted electric field distribution data, the predicted incident frequency data, and the predicted incident angle data.

[0023] As one possible implementation, the cascaded neural network model is configured as follows:

[0024] The gradient of the loss value propagating in the inverse neural network unit is defined by a composite loss function, which is as follows: Loss = α·(1-PCC) 2 )+β·MSE

[0025] in, Indicates mean square error.

[0026] This represents the Pearson correlation coefficient.

[0027] Y i Indicates the truth value. α represents the predicted value, μ represents the average value, n represents the total number of elements in each data matrix, and α and β represent the weights of each loss function, which are used to optimize the inverse neural network unit to obtain the optimization result.

[0028] Throughout the invention, the gradient of the loss value serves as a guide for adjusting the weights of the tandem neural network model after learning. This adjustment is based on currently known information (such as field distribution, angular frequency information, and cell arrangement). When the output result from a comparison with known data is unsatisfactory, it is necessary to know where the tandem neural network model should be adjusted; the gradient tells the model where to adjust more quickly. Therefore, by incorporating a composite loss function, it is possible to accurately determine where the tandem neural network model needs adjustment, thereby obtaining more accurate predictions of electric field distribution data, incident frequency data, and incident angle data.

[0029] As one possible implementation, the preset metasurface includes at least two convex shapes, each convex shape having multiple phase transition units, wherein the size of the phase transition units is selected according to the phase transition requirements of their location, and the phase transition units face the microwave band.

[0030] As one possible implementation, the electromagnetic wave data refracted and reflected by the preset metasurface is generated by a microwave detection module;

[0031] The microwave detection module includes a microwave generating unit and a microwave detection unit. The microwave generating unit is used to generate electromagnetic wave data that is refracted and reflected by a preset metasurface.

[0032] The microwave detection unit is used to detect the electromagnetic wave data to obtain near-field distribution data and far-field distribution data.

[0033] A system for constructing free-form metasurfaces includes a data acquisition module, a data preprocessing module, a data processing module, and a parameter optimization module;

[0034] The data acquisition module is used to acquire electromagnetic wave data refracted and reflected by a preset metasurface. The electromagnetic wave data includes near-field distribution data, far-field distribution data, incident frequency data, and incident angle data. The near-field distribution data is electric field distribution data, and the far-field distribution data is radar cross-sectional area distribution.

[0035] The data preprocessing module preprocesses the near-field distribution data and the far-field distribution data respectively to obtain the actual electric field distribution data of the metasurface center plane and the electromagnetic wave phase distribution data in the same plane in the far field. The incident frequency data and incident angle data are discretized to obtain the frequency and angle matrix tensor.

[0036] The data processing module trains and verifies the constructed tandem neural network model using actual electric field distribution data, electromagnetic wave phase distribution data, and matrix tensors to obtain the tandem neural network model. It then uses the actual electric field distribution data to be measured, electromagnetic wave phase distribution, incident frequency, and incident angle as inputs to the tandem neural network model to obtain initial phase transition unit arrangement parameters. Finally, it verifies the initial phase transition unit arrangement parameters using the actual electric field distribution data and electromagnetic wave phase distribution, and uses these initial phase transition unit arrangement parameters as the phase transition unit arrangement parameters for the target metasurface.

[0037] The parameter optimization module is used to optimize the initial phase change unit arrangement parameters to obtain the optimized phase change unit arrangement, and to construct an optimal metasurface based on the optimized phase change unit arrangement, wherein the optimized phase change unit arrangement includes metasurface design parameters.

[0038] As one possible implementation, the cascaded neural network model is configured as follows:

[0039] The gradient of the loss value propagating in the inverse neural network unit is defined by a composite loss function, which is as follows: Loss = α·(1-PCC) 2 )+β·MSE

[0040] in, Indicates mean square error.

[0041] This represents the Pearson correlation coefficient.

[0042] Y i Indicates the truth value. α represents the predicted value, μ represents the average value, n represents the total number of data in each data matrix, and α and β represent the weights of each loss function used to optimize the inverse neural network unit.

[0043] As one possible implementation, the preset metasurface includes at least two convex shapes, each convex shape having multiple phase transition units, wherein the size of the phase transition units is selected according to the phase transition requirements of their location, and the phase transition units face the microwave band.

[0044] As one possible implementation, it also includes a microwave detection module;

[0045] The microwave detection module includes a microwave generating unit and a microwave detection unit. The microwave generating unit is used to generate electromagnetic wave data that is refracted and reflected by a preset metasurface.

[0046] The microwave detection unit is used to detect the electromagnetic wave data to obtain near-field distribution data and far-field distribution data.

[0047] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0048] An apparatus for constructing a free-form metasurface includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method described above.

[0049] This invention, by adopting the above technical solutions, has significant technical effects:

[0050] This invention combines the advantages of deep learning with the universality of freeform metasurface applications. From a non-physical perspective, it utilizes a cascaded neural network model implemented through global inverse design to overcome the limitations of metasurface shape. The freeform metasurface constructed using this method can generate multiple scattering, thus providing the ability to reshape energy distributions with large degrees of freedom. Considering the nonlocal effects and complex internal scattering in existing technologies, the cascaded neural network model implemented through a global inverse design strategy establishes a connection between the metasurface and its electromagnetic response. This cascaded neural network model includes inverse and forward neural network units. These units allow multiple input or output constraints to simultaneously possess electric and far-field constraints within the same framework, contributing to more accurate predictions. The accuracy rates of the inverse and forward neural network units are 93.9% and 95.2%, respectively, or even higher. Compared to traditional local metasurface designs, the design of freeform metasurfaces can generate electromagnetic responses to different frequencies and incident angles more quickly, thus broadening both the incident angle and the frequency bandwidth.

[0051] The shape of the metasurface of the present invention is not limited to a single convex shape, which makes the metasurface universal for stealth objects and can broaden the range of incident angles that can be used for stealth.

[0052] The cascaded neural network model, implemented using global inverse design, can break the limitations of unit coupling and multiple scattering, thereby improving the stealth effect. Furthermore, during the training process of the cascaded neural network model, it can learn different frequencies and different incident angles corresponding to the same unit arrangement together, achieving stealth at higher frequencies (greater than 20%) and a wider range of incident angles (90°). Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a schematic diagram of the overall process of the method of the present invention;

[0055] Figure 2 This is a schematic diagram of the overall structure of the system of the present invention;

[0056] Figure 3 These are design drawings of existing metasurfaces;

[0057] Figure 4 This is a schematic diagram of a metasurface constructed based on the method or system of this invention;

[0058] Figure 5 This is a schematic diagram of the existing metasurface local field principle;

[0059] Figure 6 This is a schematic diagram of a metasurface structure according to a specific embodiment of the present invention;

[0060] Figure 7 This is a schematic diagram of the global inverter design used in the metasurface of this invention;

[0061] Figure 8 This is a schematic diagram illustrating the principle of a cascaded neural network model;

[0062] Figures 9-10 Schematic diagram of the unit structure performance and traditional stealth effect in the example of this invention;

[0063] Figure 11 This is a schematic diagram of a serial neural network model in a specific embodiment;

[0064] Figures 12-17 This is a diagram illustrating the final stealth effect in a specific embodiment of the present invention. Detailed Implementation

[0065] The present invention will be further described in detail below with reference to the embodiments. The following embodiments are explanations of the present invention, but the present invention is not limited to the following embodiments.

[0066] Exemplary method:

[0067] A method for constructing a free-form metasurface includes the following steps:

[0068] Electromagnetic wave data refracted and reflected by a preset metasurface is acquired. The electromagnetic wave data includes near-field distribution data, far-field distribution data, incident frequency data, and incident angle data. The near-field distribution data is electric field distribution data, and the far-field data is radar cross-sectional area distribution.

[0069] Preprocessing the near-field and far-field distribution data respectively yields the actual electric field distribution data of the metasurface center plane and the electromagnetic wave phase distribution data in the same plane in the far field. Discretizing the incident frequency and incident angle data yields the frequency and angle matrix tensors.

[0070] The tandem neural network model was trained and validated using actual electric field distribution data, electromagnetic wave phase distribution data, and matrix tensors to obtain the tandem neural network model.

[0071] The actual electric field distribution data to be measured, the electromagnetic wave phase distribution, the incident frequency and the incident angle are used as inputs to the series neural network model to obtain the initial phase transition unit arrangement parameters.

[0072] The initial phase transition unit arrangement parameters were verified by combining actual electric field distribution data and electromagnetic wave phase distribution, and the initial phase transition unit arrangement parameters were used as the phase transition unit arrangement parameters of the target metasurface.

[0073] The initial phase change unit arrangement parameters are optimized to obtain an optimized phase change unit arrangement. An optimal metasurface is constructed based on the optimized phase change unit arrangement, wherein the optimized phase change unit arrangement includes metasurface design parameters.

[0074] This invention combines the advantages of deep learning with the universality of freeform metasurface applications. From a non-physical perspective, it utilizes a cascaded neural network model implemented through global inverse design to overcome the limitations of metasurface shape. The freeform metasurface constructed using this method can generate multiple scattering, thus providing the ability to reshape energy distributions with large degrees of freedom. Considering the nonlocal effects and complex internal scattering in existing technologies, the cascaded neural network model implemented through a global inverse design strategy establishes a connection between the metasurface and its electromagnetic response. This cascaded neural network model includes inverse and forward neural network units. These units allow multiple input or output constraints to simultaneously possess electric and far-field constraints within the same framework, contributing to more accurate predictions. The accuracy rates of the inverse and forward neural network units are 93.9% and 95.2%, respectively, or even higher. Compared to traditional local metasurface designs, the design of freeform metasurfaces can generate electromagnetic responses to different frequencies and incident angles more quickly, thus broadening both the incident angle and the frequency bandwidth.

[0075] This invention surpasses traditional stealth methods that rely on physical formula calculations. Because existing technologies neglect unit coupling and cannot avoid scattering effects, the designed metasurfaces are often limited to simple, gently curving structures, lacking adaptability to incident angles and frequencies, and exhibiting limited energy distribution. This invention uses a free-form metasurface with multiple curvatures, which can solve the energy distribution problem over large angles or wide bandwidths. However, designing the unit arrangement using traditional analytical methods is difficult. Therefore, a cascaded neural network model can be used for calculation, resulting in better metasurface design parameters and ultimately an optimal metasurface. The shape of this metasurface is free-form; no restrictions are placed on its shape in this invention.

[0076] The shape of the metasurface of the present invention is not limited to a single convex shape, which makes the metasurface universal for stealth objects and can broaden the range of incident angles that can be used for stealth.

[0077] The cascaded neural network model, implemented using global inverse design, can break the limitations of unit coupling and multiple scattering, thereby improving the stealth effect. Furthermore, during the training process of the cascaded neural network model, it can learn different frequencies and different incident angles corresponding to the same unit arrangement together, achieving stealth at higher frequencies (greater than 20%) and a wider range of incident angles (90°).

[0078] Example 1:

[0079] A method for constructing free-form metasurfaces, such as Figure 1 As shown, it includes the following steps:

[0080] S100. Acquire electromagnetic wave data refracted and reflected by a preset metasurface, wherein the electromagnetic wave data includes near-field distribution data, far-field distribution data, incident frequency data and incident angle data, the near-field distribution data is electric field distribution data, and the far-field distribution data is radar cross-sectional area distribution.

[0081] S200. Preprocess the near-field distribution data and the far-field distribution data respectively to obtain the actual electric field distribution data of the metasurface center plane and the electromagnetic wave phase distribution data in the same plane in the far field. Discretize the incident frequency data and incident angle data to obtain the frequency and angle matrix tensor.

[0082] S300. The cascaded neural network model is trained and verified using actual electric field distribution data, electromagnetic wave phase distribution data, and matrix tensors to obtain the cascaded neural network model.

[0083] S400. The actual electric field distribution data to be measured, the electromagnetic wave phase distribution, the incident frequency and the incident angle are used as inputs to the series neural network model to obtain the initial phase transition unit arrangement parameters.

[0084] S500. The initial phase change unit arrangement parameters are verified by combining actual electric field distribution data and electromagnetic wave phase distribution, and the initial phase change unit arrangement parameters are used as the phase change unit arrangement parameters of the target metasurface.

[0085] S600. Optimize the initial phase change unit arrangement parameters to obtain the optimized phase change unit arrangement, and construct an optimal metasurface based on the optimized phase change unit arrangement, wherein the optimized phase change unit arrangement includes metasurface design parameters.

[0086] The method of the present invention can be understood as follows:

[0087] A free-form metasurface needs to be constructed as a stealth shield, participating in the sensing, reflection, and scattering of electromagnetic waves before the object being stolen. Simultaneously, by having each phase transition unit participate in different phase transformations at different positions and heights, they collectively form a near-field and far-field distribution distinct from the stolen object. The near-field electric field distribution and far-field radar RCS distribution are obtained using a microwave detection module. This measured data is then processed as input for the next step. This data is transmitted to a deep learning module to calculate and optimize the unit arrangement of the phase transition module. First, the design parameters of each phase transition unit in the phase transition module are changed according to relevant characteristics to alter its generated electromagnetic response. Second, based on the comparison of the obtained electromagnetic response with the electromagnetic response of the surrounding ground, the unit arrangement of the phase transition module is adjusted to make the measured field distribution similar to that of the ground. Finally, continuous optimization leads to an optimal concave-convex stealth metasurface that achieves an electromagnetic response similar to that of the ground.

[0088] Therefore, it is necessary to break with convention. This invention does not impose any restrictions on the shape of the metasurface. It can be a free-form metasurface with multiple convex and concave structures to replace the traditional single convex metasurface. Furthermore, it will introduce a deep learning system with artificial intelligence algorithms to solve the previously incalculable problems of multiple scattering and unit coupling. It makes breakthroughs in performance in terms of shape constraints, incident electromagnetic wave angle, and bandwidth.

[0089] In step S600, obtaining the optimized phase change unit arrangement and constructing the optimal metasurface based on the optimized phase change unit arrangement includes the following steps:

[0090] S610. Based on relevant characteristics, change the design parameters of the phase transition unit, and change the electromagnetic response generated by the metasurface by changing the design parameters to obtain the corresponding electromagnetic response;

[0091] S620. Based on the electromagnetic response, compare it with the actual electromagnetic response of the reference plane. Adjust the arrangement of the phase change units according to the design parameters of the phase change units so that the measured near-field distribution data and far-field distribution data are similar to the field distribution data of the reference plane. The corresponding design parameters of the phase change units are the design parameters of the phase change units in the optimal metasurface.

[0092] In one embodiment, the cascaded neural network model includes an inverse neural network unit and a forward neural network unit, wherein the output layer of the inverse neural network unit is the input layer of the forward neural network unit;

[0093] The input to the inverse neural network unit is the actual electric field distribution data, electromagnetic wave phase distribution data, and matrix tensor; the output of the inverse neural network unit is the phase transition unit arrangement parameters.

[0094] The input to the forward neural network unit is the phase transition unit arrangement parameters, and the output of the forward neural network unit is the predicted electric field distribution data and electromagnetic wave phase distribution data.

[0095] The inverse neural network unit includes a first input layer and a second input layer. The first input layer includes three channels, and the input for each channel is: actual electric field distribution data, incident frequency data, and incident angle data. The second input layer inputs electromagnetic wave phase distribution data.

[0096] The output layer of the forward neural network unit includes three channels, each channel outputting predicted electric field distribution data, predicted incident frequency data, and predicted incident angle data. Furthermore, to improve robustness, the cascaded neural network model is configured as follows:

[0097] The gradient of the loss value propagating in the inverse neural network unit is defined by a composite loss function, which is as follows: Loss = α·(1-PCC) 2 )+β·MSE

[0098] in, Indicates mean square error.

[0099] This represents the Pearson correlation coefficient.

[0100] Y i Indicates the truth value. α represents the predicted value, μ represents the average value, n represents the total number of elements in each data matrix, and α and β represent the weights of each loss function, which are used to optimize the inverse neural network unit to obtain the optimization result.

[0101] In other words, the relevant data obtained through data preprocessing is input into a cascaded neural network model. The output layer of the inverse neural network unit serves as the input layer of the forward neural network unit. The output result is essentially a verification of the input data. Subsequently, based on the optimization results of the neural network and using the electromagnetic response of the real ground as a reference value, the expected metasurface output is obtained. The output of the forward neural network model is used as the arrangement of the target metasurface. By verifying the learned arrangement result, its stealth performance can be determined.

[0102] In one embodiment, the preset metasurface can be configured as at least two convex shapes, each convex shape having multiple phase transition units, wherein the size of the phase transition units is selected according to the phase transition requirements of their location, and the phase transition units face the microwave band.

[0103] Furthermore, the electromagnetic wave data refracted and reflected by the preset metasurface is generated by a microwave detection module. This module includes a microwave generating unit and a microwave detection unit. The microwave generating unit generates the electromagnetic wave data refracted and reflected by the preset metasurface, and the microwave detection unit detects the electromagnetic wave data to obtain near-field and far-field distribution data. This microwave detection module is actually a very conventional one, and is not limited to it in this invention. Any module that can achieve the corresponding functions of this invention can be used.

[0104] Example 2:

[0105] A system for constructing free-form metasurfaces, such as Figure 2 As shown, it includes a data acquisition module 100, a data preprocessing module 200, a data processing module 300, and a parameter optimization module 400;

[0106] The data acquisition module 100 is used to acquire electromagnetic wave data refracted and reflected by a preset metasurface. The electromagnetic wave data includes near-field distribution data, far-field distribution data, incident frequency data, and incident angle data. The near-field distribution data is electric field distribution data, and the far-field distribution data is radar cross-sectional area distribution.

[0107] The data preprocessing module 200 preprocesses the near-field distribution data and the far-field distribution data respectively to obtain the actual electric field distribution data of the metasurface center plane and the electromagnetic wave phase distribution data in the same plane in the far field. The incident frequency data and incident angle data are discretized to obtain the frequency and angle matrix tensor.

[0108] The data processing module 300 trains and verifies the constructed tandem neural network model using actual electric field distribution data, electromagnetic wave phase distribution data, and matrix tensors to obtain the tandem neural network model; it uses the actual electric field distribution data to be measured, electromagnetic wave phase distribution, incident frequency, and incident angle as inputs to the tandem neural network model to obtain initial phase transition unit arrangement parameters; it verifies the initial phase transition unit arrangement parameters by combining the actual electric field distribution data and electromagnetic wave phase distribution, and uses the initial phase transition unit arrangement parameters as the phase transition unit arrangement parameters of the target metasurface;

[0109] The parameter optimization module 400 is used to optimize the initial phase change unit arrangement parameters to obtain the optimized phase change unit arrangement, and construct an optimal metasurface based on the optimized phase change unit arrangement, wherein the optimized phase change unit arrangement includes metasurface design parameters.

[0110] In one embodiment, the concatenated neural network model is configured as follows:

[0111] The gradient of the loss value propagating in the inverse neural network unit is defined by a composite loss function, which is as follows: Loss = α·(1-PCC) 2 )+β·MSE

[0112] in, Indicates mean square error.

[0113] This represents the Pearson correlation coefficient.

[0114] Y i Indicates the truth value. α represents the predicted value, μ represents the average value, n represents the total number of data in each data matrix, and α and β represent the weights of each loss function used to optimize the inverse neural network unit.

[0115] In one embodiment, the preset metasurface includes at least two convex shapes, each convex shape having multiple phase transition units, wherein the size of the phase transition units is selected according to the phase transition requirements of their location, and the phase transition units face the microwave band.

[0116] In addition, the entire system also includes a microwave detection module;

[0117] The microwave detection module includes a microwave generating unit and a microwave detection unit. The microwave generating unit is used to generate electromagnetic wave data that is refracted and reflected by a preset metasurface.

[0118] The microwave detection unit is used to detect the electromagnetic wave data to obtain near-field distribution data and far-field distribution data.

[0119] Based on the above methods and systems, the present invention can have the following specific embodiments:

[0120] It is necessary to design a metasurface of arbitrary shape, and then adjust the relevant parameters according to artificial intelligence algorithms to finally obtain the design parameters of the optimal metasurface. Based on the optimal metasurface design, an optimal metasurface that can generate field distribution data similar to the reference plane is designed.

[0121] like Figure 4 or Figure 6 As shown, in this embodiment, the two steeper internal peaks of the metasurface structure are rounded semi-rhomboids with a slope angle α of 72°, and the two valleys are semi-circular with a radius of 12mm. The length L along the x-direction is 210mm, the height H is designed to be 50mm, and the period along the y-direction is 10mm.

[0122] To ensure our metasurface cloak operates under full polarization, a c4-symmetric circular resonant cavity is also required as the elementary atom / unit, attached to a substrate with a relative permittivity of 3.7. The unit period is 10 mm, and the substrate thickness is 3 mm. When the radius of the circular resonant cavity is changed from 1 mm to 5 mm, the reflection phase almost covers 2π, while the reflection amplitude remains almost constant, such as... Figure 9 As shown.

[0123] To demonstrate the effectiveness of our stealth strategy, we first employ a traditional method to design this concave metasurface stealth cloak. Based on the generalized Snell's law, the required local reflection phase for each element of the metasurface is obtained using a phase compensation formula. Simulation results for the concave metasurface cloak are shown below. Figure 10 As shown, compared to the results obtained from pure ground and metasurfaces obtained by traditional methods, the stealth effect is obviously much worse due to design flaws.

[0124] To enhance stealth performance, the unit parameters in the metasurface were extended to a global inverse design, and the overall arrangement was calculated to improve the stealth performance of the concave metasurface cloak. Furthermore, stealth performance in the 8GHz to 11GHz band was verified, with the electromagnetic wave incident angle controlled within ±45°.

[0125] To ensure the effectiveness of the cascaded neural network, a dataset of approximately 60,000 samples was obtained and divided into 80%, 10%, and 10% as the training set, validation set, and test set, respectively. The cascaded neural network includes a forward network and a reverse network.

[0126] like Figure 11 As shown, for the forward network, the size of the first output (near-field output, named the first output) is 16×64×3 (each channel consists of a 16×64 matrix, the first channel represents the electric field data, the electric field distribution of the part marked by the dashed box is extracted into the 16×64 image pixel matrix, and the other channels are frequency and incident angle).

[0127] The second output (far-field output, named the second output) has a size of 91×2×3 (each channel consists of a 91×2 matrix, the first channel represents the radar cross section (RCS), and the azimuth angles from 0 to π / 2 and from 3π / 2 to 2π are discretized into 91×2 points, and the rest is the same as the first output).

[0128] For the inverse network, the near-field input (input 1) has a dimension of 16×64×3, the far-field input (input 2) has a dimension of 91×2×3, the data processing is the same as the forward network, and the output layer has 31×16×3 neurons, which are used to arrange the pattern (output3).

[0129] To demonstrate the efficiency and robustness of our smart cloak, the generated pure ground near-field and far-field data were used as input to a cascaded neural network, and the stealth performance was validated by predicting ground-like field distributions using a centered metasurface. Similar to the pre-training data, the pure ground field distribution dataset also included a wide frequency range (8-11 GHz) and a wide incident angle range (0-45°). By observing the final field reconstruction of each predicted permutation, the pattern with the highest similarity in the predicted portion was selected as the optimal solution to demonstrate its stealth performance across the frequency and angle ranges. Figure 12 , Figure 13 and Figure 14 In this study, the arrangement pattern was simulated with 97% predicted similarity under broadband frequency of 8.5 GHz and wide incident angle of 36° to illustrate the performance of the metasurface with a concave center. The simulation results were compared with those of the ground and exposed obstacles. Electromagnetic response analysis of the metasurface stealth system demonstrated that the stealth system can meet the phase compensation requirements and the influence of staggered reflected waves.

[0130] To further evaluate and verify the performance of the entire stealth system, PCC (Pearson correlation coefficient) was used to illustrate the similarity between the ground and metasurface cloak field distributions. PCC is a linear correlation coefficient, also known as the Pearson product-moment correlation coefficient. It is a statistic used to reflect the degree of linear correlation between two variables. The correlation coefficient is denoted by r, where n is the sample size, and r represents the observed values ​​and means of the two variables, respectively. r describes the strength of the linear correlation between the two variables. A larger absolute value of r indicates a stronger correlation.

[0131]

[0132] The correlation between two variables can be measured using many statistical values, the most commonly used being the Pearson correlation coefficient. For sample data, the Pearson product-moment correlation coefficient is defined as follows: the Pearson product-moment correlation coefficient (generally referred to as the sample correlation coefficient) is the sample covariance divided by the product of the standard deviations of X and Y. The simple correlation coefficient of a sample is generally represented by r, where n is the sample size, and n represents the observed values ​​and means of the two variables, respectively. r describes the strength of the linear correlation between the two variables. r ranges between -1 and +1. If r > 0, it indicates a positive correlation, meaning that the larger the value of one variable, the larger the value of the other variable will be; if r < 0, it indicates a negative correlation, meaning that the larger the value of one variable, the smaller the value of the other variable will be. A larger absolute value of r indicates a stronger correlation, but it's important to note that this does not imply a causal relationship. If r = 0, it indicates that the two variables are not linearly correlated, but there may be other forms of correlation, such as a curvilinear correlation. To infer whether two variables in the population are correlated using the sample correlation coefficient, the null hypothesis that the population correlation coefficient is 0 can be tested using the t-statistic. If the t-test is significant, the null hypothesis is rejected, meaning the two variables are linearly related; if the t-test is not significant, the null hypothesis cannot be rejected, meaning the two variables are not linearly related.

[0133] See Figures 12-17 , Figure 12 , Figure 13 and Figure 14 The actual stealth similarity shown is 93.6%. Figure 14 and Figure 15 The heatmap shows the field recovery similarity between exposed and cloaked obstacles, with a higher correlation coefficient indicating a higher similarity. Since the cloak was designed for specific frequencies and angles, and the cascaded neural network achieved good stealth performance over wide frequency and wide incident angles, this demonstrates the effectiveness of the cloak. Figure 14 and Figure 15 The results show that the stealth performance in the 8.5-10.5 GHz and 0-45° ranges is quite convincing, with an average accuracy exceeding 92.4% (see detailed results for the full simulation space in Supplementary Note S1), effectively illustrating the broadband and wide-angle characteristics of this stealth system. Furthermore, the most reliable stealth broadband frequency is determined to be 8.5 GHz, with an average accuracy of 93.2%; while the most reliable stealth incident angle is 0°, with an average accuracy of 96.5%.

[0134] according to Figure 14 and Figure 15 The incident situation at the dashed line, and the one-dimensional radar cross-section (RCS) diagram of the stealth metasurface and the ground, for example. Figure 17 As shown, the ground is represented by solid lines and the carpet by dashed lines. The height between the solid and dashed lines indicates that the intelligent concave hypersurface is suitable for large incident angles.

[0135] The stealth system of this invention is of free form, comprising a complex-shaped metasurface with multiple concave and convex features, breaking the limits of traditional metasurface cloaks in terms of incident angle and bandwidth. Furthermore, a deep learning module is incorporated, enabling global inverse design to couple the complex internal scattering and elusive metasurface to provide more precise stealth efficiency.

[0136] Since the deep learning modules are located under the same metasurface and the data processing is bundled across the frequency band and angle of incidence, it provides a good alternative for broadband and wide-angle cloaks.

[0137] The intelligent concave metasurface system enables large-scale free-form stealth, with potential applications in all-around stealth on land, ranging from microwave frequencies to optical frequencies.

[0138] This invention leverages the advantages of deep learning systems while combining the universality of freeform metasurface applications. From a non-physical perspective, it utilizes global inverse design to overcome the limitations of metasurface cloaks. The aforementioned freeform metasurfaces with multiple concave and convex features can be mutually beneficial with deep learning networks. When a freeform metasurface cloak with multiple convex and concave configurations is proposed, the ability to generate multiple scattering provides a significant opportunity to reshape the energy distribution with large degrees of freedom. Considering nonlocal effects and complex internal scattering, we establish a link between the metasurface cloak and its electromagnetic response using a global inverse design strategy. Specifically, we employ a cascaded neural network consisting of a forward network and a backward network. The multi-input / output network, with both electric field and far-field constraints within the same framework, contributes to more accurate predictions. Using the Keras application programming interface, the cascaded neural network achieves accuracies of 93.9% and 95.2% for the two sub-networks, respectively. Unlike traditional local metasurface designs, our global landscape can generate electromagnetic responses to multiple frequencies and incident angles more quickly. Therefore, both the incident angle range and bandwidth are broadened. Our work summarizes the mainstream paradigm of metasurface cloaks, with the inverse design model acting as a scattering aid for the smart cloak in dynamics.

[0139] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0140] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0141] This invention is described with reference to flowchart illustrations and / or block diagrams of the method, terminal device (system), and computer program product according to the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0142] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0143] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0144] It should be noted that:

[0145] The phrase "an embodiment" or "an embodiment" used in this specification means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the invention. Therefore, the phrase "an embodiment" or "an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0146] Furthermore, it should be noted that the shapes and names of the parts and components described in the specific embodiments described in this specification may differ. All equivalent or simple variations made to the structure, features, and principles described in this patent concept are included within the protection scope of this patent. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to replace them, as long as they do not depart from the structure of this invention or exceed the scope defined in these claims, they should all fall within the protection scope of this invention.

Claims

1. A method for constructing a free-form metasurface, characterized in that, Includes the following steps: Electromagnetic wave data refracted and reflected by a preset metasurface is acquired. The electromagnetic wave data includes near-field distribution data, far-field distribution data, incident frequency data, and incident angle data. The near-field distribution data is electric field distribution data, and the far-field data is radar cross-sectional area distribution. Preprocessing the near-field and far-field distribution data respectively yields the actual electric field distribution data of the metasurface center plane and the electromagnetic wave phase distribution data in the same plane in the far field. Discretizing the incident frequency and incident angle data yields the frequency and angle matrix tensors. The tandem neural network model was trained and validated using actual electric field distribution data, electromagnetic wave phase distribution data, and matrix tensors to obtain the tandem neural network model. The actual electric field distribution data to be measured, the electromagnetic wave phase distribution, the incident frequency and the incident angle are used as inputs to the series neural network model to obtain the initial phase transition unit arrangement parameters. The initial phase transition unit arrangement parameters were verified by combining actual electric field distribution data and electromagnetic wave phase distribution, and the initial phase transition unit arrangement parameters were used as the phase transition unit arrangement parameters of the target metasurface. The initial phase change unit arrangement parameters are optimized to obtain the optimized phase change unit arrangement. An optimal metasurface is constructed based on the optimized phase change unit arrangement, wherein the optimized phase change unit arrangement includes metasurface design parameters. The cascaded neural network model includes a reverse neural network unit and a forward neural network unit. The output layer of the reverse neural network unit is the input layer of the forward neural network unit. The input of the reverse neural network unit consists of actual electric field distribution data, electromagnetic wave phase distribution data, and a matrix tensor. The output of the reverse neural network unit is the phase transition unit arrangement parameters. The input of the forward neural network unit is the phase transition unit arrangement parameters. The output of the forward neural network unit is the predicted electric field distribution data and electromagnetic wave phase distribution data. (The reverse neural network unit is also mentioned.) The inverse neural network unit includes a first input layer and a second input layer. The first input layer includes three channels, and the input for each channel is: actual electric field distribution data, incident frequency data, and incident angle data. The second input layer inputs electromagnetic wave phase distribution data. The output layer of the forward neural network unit includes three channels, and the output for each channel is predicted electric field distribution data, predicted incident frequency data, and predicted incident angle data. The cascaded neural network model is configured such that the gradient of the loss value propagating in the inverse neural network unit is defined by a composite loss function, which is as follows: in, Indicates mean square error. This represents the Pearson correlation coefficient. Indicates the truth value. Indicates the predicted value. This represents the average value. This represents the total number of items in each data matrix. and The weights of each loss function are used to optimize the inverse neural network units, resulting in the optimization results.

2. The method for constructing a free-form metasurface according to claim 1, characterized in that, The process of obtaining the optimized phase transition unit arrangement and constructing the optimal metasurface based on this arrangement includes the following steps: The design parameters of the phase transition unit are changed based on relevant characteristics, and the electromagnetic response generated by the metasurface is changed by changing the design parameters to obtain the corresponding electromagnetic response. Based on the electromagnetic response comparison with the actual electromagnetic response of the reference plane, the arrangement of the phase change units is adjusted according to the design parameters of the phase change units so that the measured near-field distribution data and far-field distribution data are similar to the field distribution data of the reference plane. The corresponding design parameters of the phase change units are the design parameters of the phase change units in the optimal metasurface.

3. The method for constructing a free-form metasurface according to claim 1, characterized in that, The preset metasurface includes at least two convex shapes, each convex shape having multiple phase transition units. The size of the phase transition units is selected according to the phase transition requirements of their location, and the phase transition units face the microwave band.

4. The method for constructing a free-form metasurface according to claim 1, characterized in that, The electromagnetic wave data refracted and reflected by the preset metasurface is generated by the microwave detection module; The microwave detection module includes a microwave generating unit and a microwave detection unit. The microwave generating unit is used to generate electromagnetic wave data that is refracted and reflected by a preset metasurface. The microwave detection unit is used to detect the electromagnetic wave data to obtain near-field distribution data and far-field distribution data.

5. A system for constructing free-form metasurfaces, characterized in that, It includes a data acquisition module, a data preprocessing module, a data processing module, and a parameter optimization module; The data acquisition module is used to acquire electromagnetic wave data refracted and reflected by a preset metasurface. The electromagnetic wave data includes near-field distribution data, far-field distribution data, incident frequency data, and incident angle data. The near-field distribution data is electric field distribution data, and the far-field distribution data is radar cross-sectional area distribution. The data preprocessing module preprocesses the near-field distribution data and the far-field distribution data respectively to obtain the actual electric field distribution data of the metasurface center plane and the electromagnetic wave phase distribution data in the same plane in the far field. The incident frequency data and incident angle data are discretized to obtain the frequency and angle matrix tensor. The data processing module trains and verifies the constructed tandem neural network model using actual electric field distribution data, electromagnetic wave phase distribution data, and matrix tensors to obtain the tandem neural network model. It then uses the actual electric field distribution data to be measured, electromagnetic wave phase distribution, incident frequency, and incident angle as inputs to the tandem neural network model to obtain initial phase transition unit arrangement parameters. Finally, it verifies the initial phase transition unit arrangement parameters using the actual electric field distribution data and electromagnetic wave phase distribution, and uses these initial phase transition unit arrangement parameters as the phase transition unit arrangement parameters for the target metasurface. The parameter optimization module is used to optimize the initial phase change unit arrangement parameters to obtain the optimized phase change unit arrangement, and construct an optimal metasurface based on the optimized phase change unit arrangement, wherein the optimized phase change unit arrangement includes metasurface design parameters. The cascaded neural network model includes a reverse neural network unit and a forward neural network unit. The output layer of the reverse neural network unit is the input layer of the forward neural network unit. The input of the reverse neural network unit consists of actual electric field distribution data, electromagnetic wave phase distribution data, and a matrix tensor. The output of the reverse neural network unit is the phase transition unit arrangement parameters. The input of the forward neural network unit is the phase transition unit arrangement parameters. The output of the forward neural network unit is the predicted electric field distribution data and electromagnetic wave phase distribution data. (The reverse neural network unit is also mentioned.) The inverse neural network unit includes a first input layer and a second input layer. The first input layer includes three channels, and the input for each channel is: actual electric field distribution data, incident frequency data, and incident angle data. The second input layer inputs electromagnetic wave phase distribution data. The output layer of the forward neural network unit includes three channels, and the output for each channel is predicted electric field distribution data, predicted incident frequency data, and predicted incident angle data. The cascaded neural network model is configured such that the gradient of the loss value propagating in the inverse neural network unit is defined by a composite loss function, which is as follows: in, Indicates mean square error. This represents the Pearson correlation coefficient. Indicates the truth value. Indicates the predicted value. This represents the average value. This represents the total number of items in each data matrix. and The weights of each loss function are used to optimize the inverse neural network units, resulting in the optimization results.

6. The method for constructing a free-form metasurface according to claim 5, characterized in that, The preset metasurface includes at least two convex shapes, each convex shape having multiple phase transition units. The size of the phase transition units is selected according to the phase transition requirements of their location, and the phase transition units face the microwave band.

7. The method for constructing a free-form metasurface according to claim 5, characterized in that, It also includes a microwave detection module; The microwave detection module includes a microwave generating unit and a microwave detection unit. The microwave generating unit is used to generate electromagnetic wave data that is refracted and reflected by a preset metasurface. The microwave detection unit is used to detect the electromagnetic wave data to obtain near-field distribution data and far-field distribution data.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 4.

9. An apparatus for constructing a free-form metasurface, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 4.

Images