Reverse engineering methods, systems, equipment and media for metasurface lenses

Abstract

This invention relates to a method, system, device, and medium for the reverse design of metasurface lenses, comprising: generating multiple structural unit patterns based on input optical response conditions using a generative adversarial network; predicting the optical responses generated by the multiple structural unit patterns using a convolutional neural network; selecting the structural unit pattern with the smallest preset error as the target structural unit pattern of the metasurface lens, thereby completing the reverse design of the metasurface lens. This invention facilitates the rapid reverse design of high-performance metasurface lenses, reduces numerical simulation time and computer hardware resource costs, and accelerates the industrialization of metasurface lenses.

Description

Technical Field

[0001] This invention relates to a reverse design method, system, device, and medium for metasurface lenses, and pertains to the field of nanophotonics. Background Technology

[0002] Lenses are extremely important components, playing a crucial role in microscopes, cameras, mobile phones, and virtual and augmented reality glasses. A single lens focuses incident light by varying its thickness. However, the intrinsic dispersion of the material causes light of different wavelengths to focus at different locations, resulting in significant chromatic aberration and severely impacting the lens's image quality. Therefore, optical aberration correction is necessary when using lenses. Traditional geometric optics achieves chromatic aberration correction by combining multiple lenses with different surface shapes and materials. However, this system is bulky, costly to manufacture, and time-consuming, greatly limiting its use in portable and wearable devices.

[0003] To meet specific needs such as device miniaturization and system integration, metasurfaces have become a multifunctional platform for wavefront shaping. A metasurface is a planar optical element composed of a quasi-periodic array of subwavelength optical nanoantennas (also known as superatoms). By designing the phase, polarization, and amplitude degrees of freedom of the metasurface units, the wavefront can be controlled, thereby achieving various functions such as anomalous refraction, focusing, holographic imaging, and spin / orbital angular momentum loading. For example, an achromatic metasurface lens designed using metasurface design principles can achieve compact optical imaging, and achromatic focusing can be achieved by controlling the phase and dispersion of each structural unit.

[0004] The conventional design method for metasurface lenses involves first calculating the optical performance of individual structural units with different geometric parameters and shapes using numerical simulation, and then selecting the structural units. However, when faced with tens of thousands of structural units, this method consumes a significant amount of computational resources and time, severely hindering the design process. Using neural networks can establish the relationship between input size parameters, material refractive index, and other parameters and the optical response, enabling rapid prediction from large datasets and avoiding tedious numerical calculations. However, the design process still belongs to a forward design from structural units to optical response, lacking a direct design from optical response to structural units. Furthermore, the relatively simple patterns and limited degrees of freedom further restrict the performance of the resulting metasurface lenses. Summary of the Invention

[0005] The present invention aims to at least solve one of the technical problems existing in the prior art. Therefore, in response to the above-mentioned problems, the object of the present invention is to provide a method, system, device, and medium for reverse design of metasurface lenses, enabling the reverse design of metasurface lenses from demand optical response to structural unit design.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0007] In a first aspect, the reverse engineering method for metasurface lenses provided by the present invention includes:

[0008] Multiple structural unit patterns are generated based on the input optical response conditions using a generative adversarial network.

[0009] The optical response generated by multiple structural unit patterns is predicted using a convolutional neural network;

[0010] The structural unit pattern with the smallest preset error is selected as the target structural unit pattern for the metasurface lens, and the reverse design of the metasurface lens is completed.

[0011] Furthermore, it also includes establishing an initial performance database for metasurface structural units, including:

[0012] Patterns are randomly generated based on the basic properties of metasurface structural units to form structural units;

[0013] For randomly generated structural units, the optical responses of different structural units are obtained through numerical simulation, and an initial performance database is established, wherein the initial performance database includes an initial database pattern matrix and an optical response matrix.

[0014] Furthermore, the convolutional neural network is trained using the initial database pattern and the corresponding optical response in the initial database, wherein the initial database pattern serves as the input layer and the optical response serves as the output layer, and the convolutional neural network is obtained through training a certain number of iterations.

[0015] Furthermore, the convolutional neural network includes convolutional layers, pooling layers, and fully connected layers. The activation function used is sigmoid, tanh, ReLU, or LeakyReLU. The loss function used during training is root mean square error.

[0016] Furthermore, the generative adversarial network is trained by a generator network and a discriminator network. The generator network generates structural unit patterns based on the input optical response conditions, and the discriminator network distinguishes between the generated structural unit patterns and the real patterns. The generative adversarial network is obtained through training a certain number of iterations.

[0017] Furthermore, the generating network includes an input layer, a transposed convolutional layer, and an output layer. The input layer includes the input optical response conditions and a random Gaussian noise vector, and the output layer is the generated structural unit pattern.

[0018] Furthermore, the discriminant network includes an input layer, a strided convolutional layer, and an output layer. The input layer is a spliced ​​tensor of spectral conditions and real and fake patterns, and the output layer is a numerical value that reflects whether the pattern is close to the real pattern.

[0019] Secondly, the present invention provides a reverse design system for metasurface lenses, the system comprising:

[0020] The structural unit generation unit is configured to generate multiple structural unit patterns through a generative adversarial network based on the input optical response conditions.

[0021] An optical response prediction unit is configured to predict the optical response generated by multiple structural unit patterns through a convolutional neural network;

[0022] The target structural unit is configured to select the structural unit pattern with the smallest preset error as the target structural unit pattern of the metasurface lens, thereby completing the reverse design of the metasurface lens.

[0023] Thirdly, the present invention provides an electronic device comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods.

[0024] Fourthly, the present invention provides a computer-readable storage medium for storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any of the methods.

[0025] Because the present invention adopts the above technical solution, it has the following characteristics:

[0026] 1. This invention generates multiple structural unit patterns based on the input optical response conditions using a generative adversarial network; the optical responses generated by these multiple structural unit patterns are predicted using a convolutional neural network; the structural unit pattern with the smallest preset error is selected as the target structural unit pattern for the metasurface lens, thus completing the reverse design of the metasurface lens. Therefore, this invention is beneficial for realizing the rapid reverse design of high-performance metasurface lenses, reducing numerical simulation time and computer hardware resource costs, and accelerating the industrialization of metasurface lenses.

[0027] 2. Compared with existing machine learning-related metasurface lens designs, this invention can realize the direct design of optical responses to complex patterns, with higher design freedom and better performance.

[0028] 3. The reverse design method of the present invention is not limited to materials, has strong universality, high degree of freedom, short design cycle and high device efficiency.

[0029] In summary, this invention is not limited to the achromatic metasurface lens design described in this embodiment, but can also be widely applied to other types of metasurface designs, which is of great significance to the development of the metasurface industry. Attached Figure Description

[0030] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings:

[0031] Figure 1 This is a schematic diagram of randomly generated structural units in an embodiment of the present invention.

[0032] Figure 2 This is a schematic diagram of a convolutional neural network constructed according to an embodiment of the present invention.

[0033] Figure 3 This is a schematic diagram of the generative network constructed according to an embodiment of the present invention.

[0034] Figure 4 This is a schematic diagram of the discriminant network constructed for an embodiment of the present invention.

[0035] Figure 5 This is a flowchart illustrating the reverse engineering process of an embodiment of the present invention.

[0036] Figure 6 The image shows a two-dimensional layout of an achromatic metasurface cylindrical lens designed for an embodiment of the present invention.

[0037] Figure 7 A three-dimensional model of an achromatic metasurface cylindrical lens designed for an embodiment of the present invention.

[0038] Figure 8 The focusing intensity distribution of the achromatic metasurface cylindrical lens designed for embodiments of the present invention at different frequencies.

[0039] Figure 9 The focal length variation of the achromatic metasurface cylindrical lens designed for an embodiment of the present invention at different frequencies is shown in the figure.

[0040] Figure 10 The graph shows the focusing efficiency variation of the achromatic metasurface cylindrical lens designed for an embodiment of the present invention at different frequencies.

[0041] Figure 11 This is a structural diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0042] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0043] For ease of description, spatial relative terms can be used in this text to describe the relationship of one element or feature relative to another element or feature as shown in the figure. These relative terms include, for example, "inside," "outside," "middle," "outer," "below," "above," etc. Such spatial relative terms are intended to include different orientations of the device during use or operation, other than those depicted in the figure. This invention is illustrated only with the design of an achromatic metasurface lens as a specific embodiment; other types of metasurface designs will not be described further without further elaboration.

[0044] Achromatic metasurface lenses suffer from drawbacks such as long design time, high computational resource consumption, and low design freedom. This invention provides a method, system, device, and medium for the reverse design of metasurface lenses, comprising: generating multiple structural unit patterns based on input optical response conditions using a generative adversarial network; predicting the optical responses generated by these multiple structural unit patterns using a convolutional neural network; selecting the structural unit pattern with the smallest preset error as the target structural unit pattern for the metasurface lens, thus completing the reverse design of the metasurface lens. Therefore, this invention enables non-intuitive pattern reverse design, offering greater design freedom. The achromatic metasurface lens designed using this invention can achieve polarization-insensitive, high-efficiency achromatic focusing over a wide frequency range.

[0045] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0046] Example 1: The reverse engineering method for metasurface lenses provided in this example includes:

[0047] S1. Establish the initial performance database of metasurface structural units.

[0048] In this embodiment, a certain number of structural units are formed by randomly generating patterns based on the basic properties of the metasurface structural units; wherein, the basic properties of the structural units include the dielectric constant and / or loss, symmetry, and size parameters of the material.

[0049] In this embodiment, for the randomly generated structural units, the target optical performance of different structural units is obtained by numerical simulation software, and an initial performance database is established.

[0050] Furthermore, the target optical performance of the structural unit includes optical responses such as transmission or reflection amplitude, transmission or reflection phase, real part of transmission or reflection coefficient, and imaginary part of transmission or reflection coefficient.

[0051] Furthermore, the initial performance database is in a data format readable by electronic devices, including the initial database pattern matrix and the optical response matrix, which can be implemented in tabular form.

[0052] S2. By constructing and training a convolutional neural network, the relationship between the structural unit pattern and the optical response it can produce is obtained, thereby realizing the positive prediction of the performance by the structural unit pattern.

[0053] In this embodiment, the constructed convolutional neural network includes convolutional layers, pooling layers, and fully connected layers. The activation functions used are sigmoid, tanh, ReLU, or LeakyReLU. After multiple iterations of training, a neural network with a stable loss value is obtained.

[0054] In this embodiment, the convolutional neural network is trained using the initial database pattern and corresponding optical response in the established initial database. The pattern is used as the input layer, the optical response is used as the output layer, and the loss function used during training is the root mean square error. A trained convolutional neural network can be obtained through a certain number of iterations.

[0055] Furthermore, the loss function of a convolutional neural network is represented by the root mean square error or the absolute mean error:

[0056]

[0057] in, y represents the predicted value from the convolutional neural network. i is the actual value, m is the number of training samples, MSE is the root mean square error, MAE is the mean absolute error, and the magnitude of the loss function reflects the predictive performance of the convolutional neural network.

[0058] S3. By constructing and training a generative adversarial network, the relationship between the optical response and the structural unit pattern is obtained, thereby realizing the reverse design of the structural unit pattern based on the optical response.

[0059] In this embodiment, the generative adversarial network (GAN) can use common GAN models such as cGAN, DCGAN, and WGAN, and the activation function can be sigmoid, tanh, ReLU, or LeakyReLU. The GAN is trained by a generator network and a discriminator network. The generator network generates patterns, and the discriminator network determines whether the generated patterns are real or fake.

[0060] Furthermore, the generative network generates pattern designs based on the input optical response conditions. The basic structure of the generative network includes an input layer, a transposed convolutional layer, and an output layer. The input layer contains the input optical response conditions and a random Gaussian noise vector, and the output layer is the generated pattern shape.

[0061] Furthermore, the discriminant network distinguishes between generated images and real images. The basic structure of the discriminant network includes an input layer, a strided convolutional layer, and an output layer. The input layer is a tensor of spectral conditions and spliced ​​real and fake images, and the output layer is a numerical value used to distinguish between real and fake patterns.

[0062] S4. Combining convolutional neural networks and generative adversarial networks to complete the reverse design, the desired target structural unit is obtained by inputting the required optical response, and the achromatic metasurface lens design is completed through optimization.

[0063] In this embodiment, the above process includes:

[0064] S51, Input Requirements: Optical Response to Reverse Design Network

[0065] The reverse design network includes generative adversarial networks (GANs) and convolutional neural networks (CNNs). The GANs are used to generate multiple possible pattern designs for a given input optical response; the CNNs are used to predict the optical response of multiple possible pattern designs.

[0066] S52. Select the target structural unit and complete the design of the achromatic metasurface lens.

[0067] Specifically, the selection of target structural units includes: selecting the structural unit with the smallest preset error as the desired target structural unit to complete the design of the achromatic metasurface lens. The final selected structural unit not only has a good phase match with the required phase, but also has a high transmission amplitude. The specific settings can be made according to the actual application, which will not be elaborated here.

[0068] The following detailed embodiments illustrate how the present invention achieves the reverse design of achromatic metasurface lens structure units using convolutional neural networks and generative adversarial networks.

[0069] The reverse design method for achromatic metasurface lenses provided in this embodiment includes:

[0070] The metasurface operates in the 400-700 THz frequency band (3 THz step), the substrate material is fused silica, the dielectric pillar material is titanium dioxide, the period of the structural unit is 320 nm, the height of the dielectric pillar is 600 nm, and the operating frequency band is set to S. 21 The real and imaginary parts of the parameter (transmission coefficient) are used as the target optical response (a total of 202 values).

[0071] First, 6000 different cross-sectional patterns are randomly generated (each cross-sectional pattern is 64*64 pixels, and each pixel represents a 5*5nm area), such as Figure 1 The image shows a partially randomly generated pattern, where black pixels represent the presence of a medium and white pixels represent air.

[0072] Secondly, electromagnetic field simulation software was used to simulate 6000 randomly generated patterns to obtain the operating frequency band S. 21 The real and imaginary parts of the parameters are used to establish an initial performance database, which includes the initial database pattern matrix and the optical response matrix.

[0073] Next, construct and train a convolutional neural network.

[0074] Specifically, the convolutional neural network constructed in this embodiment is as follows: Figure 2 As shown, the input image is a 64*64 pattern. After passing through four 3*3 convolutional layers and a 2*2 pooling layer, a 128*4*4 tensor is obtained. The ReLU activation function is used. This 128*4*4 tensor is flattened into 2048 neurons, which are then connected to fully connected layers containing 512 and 202 neurons respectively. The output is S at 101 frequencies. 21 The real and imaginary parts of the parameter.

[0075] Convolutional neural networks are trained using an initial performance database. The loss function used during training is the root mean square error. A well-trained convolutional neural network can be obtained through a certain number of iterations.

[0076] Then, a generative adversarial network is constructed and trained.

[0077] Specifically, generative adversarial networks are trained by generating networks and discriminative networks.

[0078] like Figure 3As shown, the input conditions for constructing the generator network are the cosine and sine values ​​of the phase at 101 frequencies (a total of 202 values). The input layer is obtained by concatenating it with a random Gaussian noise vector of dimension 50. After four transposed convolutions (using ReLU as the activation function), the output is a 64*64 pixel generated image.

[0079] like Figure 4 As shown, a discriminant network is constructed. The input layer is a tensor of spectral conditions and spliced ​​real and fake images. A 512*4*4 tensor is obtained through 4 strided convolutions (using LeakyReLU as the activation function), and finally a numerical value is output.

[0080] Generative adversarial networks (GANs) use the Wasserstein distance to measure the magnitude of the loss function. The loss function expression for the discriminant network D is as follows:

[0081]

[0082] The loss function expression for the generator network G is as follows:

[0083]

[0084] in To generate an image, x ~ P r These are real photos. Random interpolation between the generated image and the real image:

[0085]

[0086] λ is the penalty coefficient, which is set to 10 in this example. Let E represent the 2-norm of the discriminant network gradient, E represent the expected value, and ε be a random number in the range [0,1]. To generate an image, a score is determined by the output of the discrimination network. To generate an image, x represents the actual image.

[0087] Finally, the inverse design is completed by combining convolutional neural networks and generative adversarial networks. By inputting the required optical response, the desired target structural unit is obtained, thus completing the design of the achromatic metasurface lens.

[0088] Specifically, such as Figure 5 As shown, the phase response spectrum of the structural unit at each position is given according to the ideal formula that the achromatic metasurface lens needs to satisfy. Then, the phase is converted into cosine and sine values, which are input into the generative adversarial network to obtain 100 candidate pattern designs. The transmission spectrum response of the 100 candidate pattern designs is predicted by the convolutional neural network. In this embodiment, the structure with the smallest phasor error is selected for the final achromatic metasurface lens.

[0089] Furthermore, the relevant parameters of the achromatic metasurface lens are: the required phase at the center angular frequency ω0. Group latency Group delay dispersion The expression is:

[0090]

[0091] In this context, mod represents the modulo operation, r is the radial position coordinate of the structural unit, F is the focal length of the designed achromatic metasurface lens, R is the radius of the designed achromatic metasurface lens, A, B, and D are given constants, and c is the speed of light.

[0092] At this point, the expression for the required phase Φ(r,ω) is:

[0093]

[0094] Phasor error is defined as the design phasor predicted by a convolutional neural network. With demand phasor e iΦ(r,ω) The distance between them, at which point the total phasor error at r is:

[0095]

[0096] Wherein, ω1~ω k To design the frequency band sampling angular frequency, t(r,ω) is the transmission amplitude of the superatom at the radial position r. It is the phase of the superatom at the radial position r.

[0097] like Figure 6 The image shows a two-dimensional layout of the designed 10-micron diameter achromatic metasurface cylindrical lens, as shown. Figure 7 The image shown is a 3D model. Figure 8 The figure shows the results of numerical simulation calculations of the focused light intensity of a cylindrical lens at different frequencies. From this, the focal length variation at different frequencies can be obtained as follows: Figure 9 As shown, the average focal length is 17.7 micrometers, and the average deviation is 0.73 micrometers. Figure 10 The graph shows the focusing efficiency variation at different frequencies. The average focusing efficiency is as high as 68%, indicating that the achromatic metasurface lens obtained by reverse design using the present invention has excellent performance.

[0098] Example 2: Following the reverse design method for achromatic metasurface lenses provided in Example 1, this example provides a reverse design system for achromatic metasurface lenses. The system provided in this example can implement the reverse design method for achromatic metasurface lenses of Example 1. This system can be implemented through software, hardware, or a combination of both. For ease of description, this example is described by dividing the functions into various units. Of course, in implementation, the functions of each unit can be implemented in one or more software and / or hardware components. For example, the system may include integrated or separate functional modules or units to execute the corresponding steps in the methods of Example 1. Since the system in this example is basically similar to the method example, the description process of this example is relatively simple. Relevant details can be found in the description of Example 1. The example of the reverse design system for achromatic metasurface lenses provided by this invention is merely illustrative.

[0099] Specifically, the reverse engineering system for metasurface lenses provided in this embodiment includes:

[0100] The structural unit generation unit is configured to generate multiple structural unit patterns through a generative adversarial network based on the input optical response conditions.

[0101] An optical response prediction unit is configured to predict the optical response generated by multiple structural unit patterns through a convolutional neural network;

[0102] The target structural unit is configured to select the structural unit pattern with the smallest preset error as the target structural unit pattern of the metasurface lens, thereby completing the reverse design of the metasurface lens.

[0103] Example 3: This example provides an electronic device corresponding to the reverse design method of the achromatic metasurface lens provided in Example 1. The electronic device can be an electronic device for the client, such as a mobile phone, laptop, tablet computer, desktop computer, etc., to execute the method of Example 1.

[0104] like Figure 11As shown, the electronic device includes a processor, memory, communication interface, and bus. The processor, memory, and communication interface are connected via the bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. The memory stores a computer program that can run on the processor. When the processor runs the computer program, it executes the method described below. The implementation principle and technical effects are similar to those in Embodiment 1, and will not be repeated here. Those skilled in the art will understand that... Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computing device on which the present application is applied. The specific computing device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0105] In a preferred embodiment, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), and optical discs.

[0106] In a preferred embodiment, the processor can be any type of general-purpose processor such as a central processing unit (CPU) or a digital signal processor (DSP), and is not limited thereto.

[0107] Example 4: This example provides a computer program product. The computer program product may include a computer program stored on a computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by the computer, the computer can execute the method provided in Example 1 above. Its implementation principle and technical effects are similar to those in Example 1, and will not be repeated here.

[0108] In a preferred embodiment, the computer-readable storage medium may be a tangible device for holding and storing instructions used by an instruction execution device, such as, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. The computer-readable storage medium stores computer program instructions that cause a computer to perform the method provided in Embodiment 1 above.

[0109] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In the description of this specification, the reference to terms such as "a preferred embodiment" indicates that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the embodiments in this specification. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0110] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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 apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0111] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function 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.

[0112] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable 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.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A reverse design method for metasurface lenses, characterized in that... include: Multiple structural unit patterns are generated based on the input optical response conditions using a generative adversarial network. The optical response generated by multiple structural unit patterns is predicted using a convolutional neural network; The structural unit pattern with the smallest preset error is selected as the target structural unit pattern for the metasurface lens, and the reverse design of the metasurface lens is completed. The generative adversarial network (GAN) is trained through a generator network and a discriminator network. The generator network generates the structural unit pattern based on the input optical response conditions. The generator network includes an input layer, a transposed convolutional layer, and an output layer. The input layer includes the input optical response conditions and a random Gaussian noise vector, and the output layer is the generated structural unit pattern. The discriminator network distinguishes between the generated structural unit pattern and the real pattern. The GAN is obtained through training a certain number of iterations. The discriminator network includes an input layer, a strided convolutional layer, and an output layer. The input layer is the spectral conditions and the spliced ​​tensor of the real and fake patterns, and the output layer is a value indicating whether the pattern closely approximates the real pattern.

2. The reverse design method for metasurface lenses according to claim 1, characterized in that, It also includes establishing an initial performance database for metasurface structural units, including: Patterns are randomly generated based on the basic properties of metasurface structural units to form structural units; For randomly generated structural units, the optical responses of different structural units are obtained through numerical simulation, and an initial performance database is established, wherein the initial performance database includes an initial database pattern matrix and an optical response matrix.

3. The reverse design method for metasurface lenses according to claim 2, characterized in that, The convolutional neural network is trained using the initial database pattern and the corresponding optical response in the initial database, wherein the initial database pattern is used as the input layer and the optical response is used as the output layer, and the convolutional neural network is obtained through a certain number of iterations.

4. The reverse design method for metasurface lenses according to claim 3, characterized in that, The convolutional neural network includes convolutional layers, pooling layers, and fully connected layers. The activation functions used are sigmoid, tanh, ReLU, or LeakyReLU. The loss function used during training is the root mean square error.

5. A reverse design system for a metasurface lens used to implement the reverse design method for a metasurface lens according to any one of claims 1 to 4, characterized in that, The system includes: The structural unit generation unit is configured to generate multiple structural unit patterns through a generative adversarial network based on the input optical response conditions. An optical response prediction unit is configured to predict the optical response generated by multiple structural unit patterns through a convolutional neural network; The target structural unit is configured to select the structural unit pattern with the smallest preset error as the target structural unit pattern of the metasurface lens, thereby completing the reverse design of the metasurface lens.

6. An electronic device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 4.

7. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described in claims 1 to 4.

Images