Structure pattern generation model training method and device for electromagnetic super surface

Abstract

The application provides a structure pattern generation model training method and device for an electromagnetic metasurface, applied to a server in a structure pattern generation model training system. The method comprises the following steps: taking one of a plurality of initial regions in a to-be-designed pattern of the electromagnetic metasurface as a to-be-generated region; determining a basic structure pattern of the electromagnetic metasurface based on a resonator pattern selected from a plurality of resonator patterns; determining a target size of the basic structure pattern based on a size selected from a preset size range; determining a structure pattern in the to-be-generated region according to the basic structure pattern and the target size; mapping the structure pattern in the to-be-generated region to each initial region to obtain a target structure pattern of the electromagnetic metasurface; simulating the target structure pattern to obtain electromagnetic response data and combining the target structure pattern to obtain a training data set, and training a structure pattern generation model by using the training data set. The accuracy of the electromagnetic metasurface structure pattern design can be improved.

Description

Technical Field

[0001] This application belongs to the field of electromagnetic metasurface design, specifically relating to a method and apparatus for training structural pattern generation models for electromagnetic metasurfaces. Background Technology

[0002] Electromagnetic metasurfaces, as artificial two-dimensional materials capable of precisely manipulating electromagnetic waves, have performance highly dependent on the geometry of their subwavelength units. In metasurface design, "reverse engineering"—deriving the corresponding physical structure from the desired electromagnetic response (such as a specific reflection phase and amplitude)—is the core and most challenging aspect.

[0003] Currently, methods for achieving "reverse design" of electromagnetic metasurfaces include pixel-based design methods. For example, structural pattern generation models can directly map the performance of electromagnetic metasurfaces to their structure. The resulting mapping is a binary pixel matrix, where each pixel "1" represents the placement of metal and "0" represents the absence of metal. This method offers extremely high design freedom and can theoretically express arbitrarily complex two-dimensional topological structures. Typically, training the structural pattern generation model relies on randomly generated pixel patterns. By performing electromagnetic simulations on these randomly generated patterns, the electromagnetic simulation data is correlated with the pixel patterns and used as sample data for the generated model.

[0004] However, the randomly generated patterns by the system contain many physically invalid patterns, specifically those structures that do not conform to electromagnetic laws and processing requirements. For example, many patterns do not meet these requirements. Symmetrical patterns and fragmented or physically meaningless patterns generated by disordered pixel filling on electromagnetic metasurfaces result in low-quality datasets used to train generative models. Consequently, the generative models cannot effectively learn the "structure-electromagnetic response" relationship of electromagnetic metasurfaces, leading to inaccurate output structures of electromagnetic metasurfaces by subsequent structural pattern generation models during the "reverse design" process, which fail to meet the expected electromagnetic response. Summary of the Invention

[0005] This application provides a method and apparatus for training a structural pattern generation model for electromagnetic metasurfaces. The server in the structural pattern generation model training system can determine the basic structural pattern and its dimensions of the electromagnetic metasurface from several resonator patterns and a preset size range. Then, based on this basic structural pattern and its dimensions, a structural pattern can be designed for one region in the design drawing, and the designed structural pattern for that region can be mapped to other regions so that the structural pattern of the electromagnetic metasurface satisfies… The structural pattern generation model trained using the training dataset constructed from the structural pattern of this electromagnetic metasurface, which is symmetrical and has practical physical meaning, can generate more accurate structural patterns in "reverse engineering".

[0006] This application provides a training method for a structural pattern generation model of an electromagnetic metasurface, applied to a server in a structural pattern generation model training system. The method includes: acquiring a design drawing of the electromagnetic metasurface; selecting one of several initial regions from the design drawing as a region to be generated; determining a basic structural pattern of the electromagnetic metasurface based on a resonator pattern selected from several types of resonator patterns; determining a target size of the basic structural pattern based on a size selected from a preset size range; determining a structural pattern within the region to be generated based on the basic structural pattern and the target size; mapping the structural pattern within the region to be generated to each initial region to obtain a target structural pattern of the electromagnetic metasurface; wherein the target structural pattern can coincide with the design drawing after rotating 90° around its center point and is an axisymmetric pattern; simulating the target structural pattern to obtain electromagnetic response data; obtaining a training dataset based on the target structural pattern and the electromagnetic response data; and training a structural pattern generation model using the training dataset, wherein the trained structural pattern generation model is used to generate a structural pattern of the electromagnetic metasurface corresponding to the input reference electromagnetic response data.

[0007] In one possible embodiment, the structural pattern of the region to be generated includes type identification information for each pixel in the region. Determining the structural pattern within the region to be generated based on the basic structural pattern and the target size includes: determining a pixel coverage confirmation strategy corresponding to the basic structural pattern; determining the pixels covered by the basic structural pattern in the region to be generated based on the pixel coverage confirmation strategy and the target size; determining the type identification information of the pixels covered by the basic structural pattern in the region to be generated as metal type identifiers; and determining the type identification information of the pixels not covered by the basic structural pattern as medium type identifiers.

[0008] In one possible embodiment, the basic structure pattern includes a ring resonator pattern, and the target size includes a radius and a width. Determining the pixels covered by the basic structure pattern in the area to be generated, based on a pixel coverage confirmation strategy and the target size, includes: obtaining the distance between each pixel in the area to be generated and the center point, and the difference between each distance and the radius, according to the pixel coverage confirmation strategy; using the difference whose absolute value is less than or equal to the width as a first difference, and the difference whose absolute value is greater than the width as a second difference; determining the pixels corresponding to each first difference as pixels covered by the ring resonator pattern, and determining the pixels corresponding to each second difference as pixels not covered by the ring resonator pattern.

[0009] In one possible embodiment, determining the type identification information of some pixels covered by the basic structure pattern in the area to be generated as metal type identification includes: setting the type identification information of the pixels covered by the basic structure pattern as metal type identification according to a first preset probability; and in response to the existence of pixels covered by the basic structure pattern whose type identification information is not set as metal type identification, setting the type identification information of the pixels that are not set as metal type identification as medium type identification.

[0010] In one possible embodiment, the method further includes: defining the region where the pixel with type identification information of metal type in the target structure pattern of the electromagnetic metasurface is located as a metal region; determining the number and shape of the metal regions in the target structure pattern; and performing a simulation of the target structure pattern to obtain electromagnetic response data in response to the fact that the number of metal regions in the target structure pattern is one and the shape is a connected ring.

[0011] In one possible embodiment, before mapping the structural pattern within the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface, the method further includes: determining at least one target seed pixel from several types of seed pixels, the target seed pixel including multiple reference pixels; sequentially performing an update operation on the structural pattern of the region to be generated using one of the target seed pixels to obtain the updated structural pattern of the region to be generated, the update operation including: selecting a pixel from the latest medium region in the region to be generated as an anchor point, the medium region being the region where the pixel with the type identification information is a medium type identifier; adding the row coordinates of each reference pixel in the target seed pixel to the row coordinates of the anchor point and the column coordinates to the column coordinates of the anchor point to obtain the pixel where each reference pixel is located in the region to be generated; determining the type identification information of each reference pixel in the region to be generated as a metal type identifier.

[0012] In one possible embodiment, selecting a pixel from the latest media region in the region to be generated as an anchor point includes: dividing the latest media region into a first media region and a second media region, wherein each pixel in the first media region is not adjacent to any pixel in the metal region of the region to be generated, and each pixel in the second media region is adjacent to at least one pixel in the metal region of the region to be generated, wherein the metal region of the region to be generated is the region where the pixel whose type identification information is metal type identifier is located; selecting a pixel from the first media region according to a second preset probability, wherein the second preset probability is greater than 50%; in response to selecting a pixel from the first media region according to the second preset probability, using the selected pixel as an anchor point; or, in response to not selecting a pixel from the first media region according to the second preset probability, selecting a pixel from the second media region as an anchor point.

[0013] In one possible embodiment, determining the type identification information of each reference pixel in the region to be generated as a metal type identifier includes: determining whether each reference pixel has a corresponding pixel in the region to be generated; and in response to the fact that each reference pixel has a corresponding pixel in the region to be generated, performing the step of determining the type identification information of each reference pixel in the region to be generated as a metal type identifier.

[0014] In one possible embodiment, mapping the structural pattern within the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface includes: mapping the structural pattern within the region to be generated to each initial region other than the region to be generated, to obtain the initial structural pattern of the electromagnetic metasurface; determining that there are two metal regions separated from each other in the initial structural pattern, the metal regions being the regions where the pixel points with type identification information of metal type are located; adjusting the structural pattern in the region to be generated; and mapping the adjusted structural pattern in the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface.

[0015] This application also provides a training device for a structural pattern generation model of an electromagnetic metasurface, applied to a server in a structural pattern generation model training system. The structural pattern generation model training device includes: a data acquisition unit, a region selection unit, a pattern selection unit, a pattern generation unit, a mapping unit, a simulation unit, a dataset determination unit, and a training unit. The data acquisition unit is used to acquire the design drawing of the electromagnetic metasurface. The region selection unit is used to select one of several initial regions from the design drawing as the region to be generated. The pattern selection unit is used to determine the basic structural pattern of the electromagnetic metasurface based on a resonator pattern selected from several types of resonator patterns, and to determine the target size of the basic structural pattern based on a size selected from a preset size range. The pattern generation unit… The system is divided into three parts: a first part, a second part, and a third part. The first part is used to determine the structural pattern within the region to be generated based on the basic structural pattern and the target size; a second part is used to map the structural pattern within the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface; the target structural pattern is designed to coincide with the center point of the design drawing after a 90° rotation and is axially symmetric; a third part is used to simulate the target structural pattern to obtain electromagnetic response data; a fourth part is used to determine the training dataset based on the target structural pattern and the electromagnetic response data; and a fifth part is used to train the structural pattern generation model using the training dataset. The trained structural pattern generation model is used to generate the structural pattern of the electromagnetic metasurface corresponding to the input reference electromagnetic response data.

[0016] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements any of the above-described methods for training structural pattern generation models for electromagnetic metasurfaces.

[0017] This application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described methods for training structural pattern generation models for electromagnetic metasurfaces.

[0018] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described methods for training structural pattern generation models for electromagnetic metasurfaces.

[0019] In the technical solution of the training method and apparatus for generating structural patterns of electromagnetic metasurfaces provided in this application, the server first divides the design drawing of the electromagnetic metasurface into regions. Then, it uses a set of preset resonator patterns with controllable dimensional parameters as the basic structural pattern of the electromagnetic metasurface in only one region. This effectively generates a continuous feature pattern on the design drawing. Compared with disordered pixel filling of the entire region of the design drawing, this ensures that the generated target structural pattern of the electromagnetic metasurface has physical meaning and reduces the generation of a large number of broken or meaningless structural patterns. After obtaining the structural pattern of the region to be generated, the structural pattern of the region is mapped to other regions so that the target structural pattern of the electromagnetic metasurface satisfies... Symmetry, achieved through a two-stage structural pattern design method, reduces the degree of freedom in randomly designing the overall design drawing, resulting in an electromagnetic metasurface structural pattern that satisfies... The electromagnetic response data obtained from electromagnetic simulation based on the structure of the electromagnetic metasurface is of high quality, which enables the structural pattern generation model to effectively learn the "structure-electromagnetic response" relationship of the electromagnetic metasurface. Consequently, the structure of the electromagnetic metasurface output by the subsequent structural pattern generation model in the "reverse design" process is accurate, and the generated structural pattern of the electromagnetic metasurface can meet the expected electromagnetic response. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1This is one of the flowcharts illustrating a method for training a structural pattern generation model for electromagnetic metasurfaces provided in this application.

[0022] Figure 2 This is a schematic diagram of the communication between the server and the client in the structural pattern generation model training system provided in this application.

[0023] Figure 3 This is a schematic diagram showing the location of the area to be generated in the design drawing provided in this application.

[0024] Figure 4 This is the second flowchart of a method for training a structural pattern generation model for electromagnetic metasurfaces provided in this application.

[0025] Figure 5 This is a schematic diagram of the pixels covered by the basic structure in the design drawing provided in this application.

[0026] Figure 6 This is a schematic diagram of the pixels covered by the basic structure in the area to be generated provided in this application.

[0027] Figure 7 This is the third flowchart of a method for training a structural pattern generation model for electromagnetic metasurfaces provided in this application.

[0028] Figure 8 This is the fourth flowchart of a method for training a structural pattern generation model for electromagnetic metasurfaces provided in this application.

[0029] Figure 9 This is one of the functional unit block diagrams of a structural pattern generation model training device for electromagnetic metasurfaces provided in this application.

[0030] Figure 10 This is the second functional unit block diagram of a training device for generating structural patterns of electromagnetic metasurfaces provided in this application.

[0031] Figure 11 This is a schematic diagram of the structure of an electronic device provided in this application. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0033] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0034] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0035] Given that the current training dataset generation process often involves generating structural patterns from scattered pixels of the design drawing of the electromagnetic metasurface, the generated structural patterns often contain fragmented or physically meaningless structures, resulting in low quality training datasets. As a result, the structural pattern generation model cannot effectively learn the correlation between the structural pattern and the electromagnetic response data, thus making it impossible to output accurate structural patterns during the reverse engineering process.

[0036] To address this technical problem, this application proposes a two-stage method for generating target structural patterns of electromagnetic metasurfaces during the construction of the training dataset. The server determines the basic structural pattern and its dimensions of the electromagnetic metasurface from several resonator patterns and a preset size range. Then, based on the basic structural pattern and its dimensions, a structural pattern can be designed for a region in the design drawing. The structural pattern designed for that region is then mapped to other regions so that the structural pattern of the electromagnetic metasurface satisfies symmetry and has practical physical meaning. The structural pattern generation model obtained after training the training dataset constructed using the structural pattern of the electromagnetic metasurface can generate more accurate structural patterns in the "reverse design".

[0037] Please see Figure 1 , Figure 1 This is one of the flowcharts illustrating a method for training a structural pattern generation model for electromagnetic metasurfaces provided in this application. For example... Figure 2As shown, the method for training a structural pattern generation model for electromagnetic metasurfaces is applied to the server 100 in the structural pattern generation model training system 10. The structural pattern generation model training system 10 may also include a client 200 that is communicatively connected to the server 100. The client 200 can be used to receive model training requests, wherein the model training request may include, but is not limited to, the number of target structural patterns contained in the training dataset. The method for training a structural pattern generation model for electromagnetic metasurfaces may include the following steps:

[0038] S101, Obtain the design drawing of the electromagnetic metasurface.

[0039] Electromagnetic metasurfaces, as artificial two-dimensional materials capable of precisely manipulating electromagnetic waves, rely heavily on the geometry of their subwavelength units. The unit cell on an electromagnetic metasurface is the smallest repeating structural unit and the core basic unit for realizing the metasurface's electromagnetic wave manipulation function. Unit cells can include metallic and dielectric units. The dimensions of each unit cell in an electromagnetic metasurface can range from 9mm to 11mm in width and length, and from 0.8mm to 0.9mm in thickness.

[0040] S102, select one of the initial regions in the design drawing as the region to be generated.

[0041] The design drawing can be a square image, for example, divided into 18×18 pixels, with each pixel corresponding to a unit on an electromagnetic metasurface. One of several initial regions in the design drawing is selected as the region to be generated. This involves dividing the design drawing into eight initial regions. Then, any one of these eight initial regions is selected as the region to be generated. Specifically, the design drawing is divided into eight equal-sized regions using a star-shaped pattern, allowing for the selection of any one region as the region to be generated. Alternatively, the initial region with the smallest sum of row and column coordinates among all initial regions can be selected as the region to be generated; that is, the region to be generated must have the smallest sum of row and column coordinates among all initial regions.

[0042] Optionally, if the design drawing contains 18×18 pixels, the type identifier of the outermost pixel can be set to the media type identifier. For example, the media type identifier can have a pixel value of 0. Based on this, the number of pixels in the design drawing that need to generate the structural pattern is 16×16. Within these 16×16 pixels, the coordinates of the top-left pixel are set to (0,0), and the coordinates of the bottom-right pixel are (15,15). Figure 3As shown, the initial area filled with diagonal lines in the upper left corner can be used as the area to be generated. The sum of the row coordinates and column coordinates of each pixel point in the area to be generated is the smallest among the pixel points in each initial area. The range of the row coordinate r of the area to be generated is: 0 ≤ r < N / 2. In this embodiment, N is the size of the area (16×16) where the structural pattern needs to be generated in the to-be-designed figure, specifically 16, that is: 0 ≤ r < 8. The range of the column coordinate c of the area to be generated is: r ≤ c < N / 2. In this embodiment, the column coordinate c of the area to be generated is: 0 ≤ c < 8.

[0043] S103. Based on the resonator pattern selected from several types of resonator patterns, determine the basic structural pattern of the electromagnetic metasurface; and based on the size selected from the preset size range, determine the target size of the basic structural pattern.

[0044] Among them, several types of resonator patterns can include, but are not limited to, different types of resonator patterns such as cross-shaped resonator patterns, square-frame resonator patterns, and ring-shaped resonator patterns. Among them, the selected resonator pattern can be directly used as the basic structural pattern of the electromagnetic metasurface, or the resonator pattern can be deformed to obtain the basic structural pattern of the electromagnetic metasurface.

[0045] The preset size range can include the size sub-ranges of each type of resonator pattern. For example, the size sub-range of the ring-shaped resonator pattern can include the radius and the width. Here, the radius can be the average radius, and the width can be the difference between the minimum radius or the maximum radius and the average radius. The average radius is the most important factor determining its fundamental frequency resonance point. Generally, the larger the average radius, the lower the resonance frequency. The width of the ring-shaped resonator pattern affects the quality factor (Q value) and bandwidth of the resonance. A wider ring usually has a wider bandwidth. The size sub-range of the square-frame resonator pattern can include the length, height, and width, etc. The length can be the difference between the column coordinates of the right boundary and the left boundary, the height can be the difference between the row coordinates of the lower boundary and the upper boundary, and the width can be the difference between the row coordinates (or column coordinates) of the pixel points on the outer boundary and the row coordinates (or column coordinates) of the inner boundary. The size sub-range of the cross-shaped resonator pattern can include the length, height, width, etc. of the cross. The length of the cross can be the difference between the column coordinates of the right boundary and the left boundary of the cross, the height of the cross can be the difference between the row coordinates of the lower boundary and the upper boundary of the cross, and the width of the cross can be the width of the metal microstrip line used to form the cross-shaped resonator. Determine the corresponding size sub-range according to the selected resonator pattern, and then the corresponding target size can be randomly selected from the corresponding size sub-range or determined from the preset sizes with fewer selection times according to the number of times each preset size is selected.

[0046] The target size of a basic structural pattern can be determined by directly using the selected size as the target size, or by processing the selected size. For example, the target size can be obtained based on the selected size and the size of the area in the design drawing excluding the outermost pixels. Taking a ring resonator pattern as an example, the size sub-range of the ring resonator pattern includes a radius selection range and a width selection range. The radius can be obtained by randomly selecting a value from the radius selection range and then multiplying it by the size of the area in the design drawing excluding the outermost pixels. For example, if 0.3 is selected from the radius selection range [0.1, 0.45], the size of the design drawing is 20×20, and the size of the area in the design drawing excluding the outermost pixels is 18×18, then the radius is 0.3×18, which is 5.4. The width selection range can include one or more selectable values. For example, the width can be a preset fixed value, such as 1 pixel.

[0047] S104. Determine the structural pattern within the area to be generated based on the basic structural pattern and the target size.

[0048] The number of pixels covered by the basic structural pattern in the area to be generated can be determined based on the basic structural pattern and its target size. Pixels covered by the basic structural pattern and pixels not covered by the basic structural pattern can be distinguished and marked to obtain the structural pattern in the area to be generated. Alternatively, pixels not covered by the basic structural pattern can be randomly marked as pixels covered by the basic structural pattern to obtain the structural pattern in the area to be generated.

[0049] S105, map the structural pattern in the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface.

[0050] Among them, the target structure pattern of the electromagnetic metasurface satisfies Symmetry. Specifically, the target structural pattern should be able to coincide with the design drawing after rotating 90° around its center point, and the target structural pattern should be axially symmetric. The axes of symmetry of the target structural pattern include two centerlines and two diagonals that pass through the center point of the design drawing.

[0051] Optionally, the target structure pattern of the electromagnetic metasurface can be stored in an array, for example, the array stores the region to be generated and the type identification information of each pixel in each initial region.

[0052] S106, Simulate the target structural pattern to obtain electromagnetic response data.

[0053] The server can invoke simulation tools to simulate the target structural pattern. These simulation tools include, but are not limited to, High Frequency Structure Simulators (HFSS). Electromagnetic response data includes, but is not limited to, phase information, reflectivity, transmittance, and absorptivity.

[0054] S107. Based on the target structure pattern and electromagnetic response data, a training dataset is obtained.

[0055] The target structure pattern and electromagnetic response data are added to the training dataset as a set of associated data.

[0056] S108, using the training dataset, train the structural pattern generation model.

[0057] In this model, electromagnetic response data serves as input to the structural pattern generation model, while the target structural pattern acts as a label. The loss is calculated by examining the difference between the target structural pattern and the output structural pattern of the generation model, and this loss is then used to adjust the parameters of the model. The trained structural pattern generation model is used to generate a structural pattern for an electromagnetic metasurface corresponding to the input reference electromagnetic response data.

[0058] In the above scheme, the server in the structural pattern generation model training system first divides the design drawing of the electromagnetic metasurface into regions. Then, it uses a set of preset resonator patterns with controllable dimensional parameters as the basic structural pattern of the electromagnetic metasurface in only one region. This effectively generates a continuous feature pattern on the design drawing. Compared to randomly filling the entire region of the design drawing with pixels, this ensures that the generated target structural pattern of the electromagnetic metasurface has physical meaning and reduces the generation of a large number of broken or meaningless structural patterns. After obtaining the structural pattern of the region to be generated, the structural pattern of that region is mapped to other regions, so that the target structural pattern of the electromagnetic metasurface satisfies... Symmetry, achieved through a two-stage structural pattern design method, reduces the degree of freedom in randomly designing the overall design drawing, resulting in an electromagnetic metasurface structural pattern that satisfies... The electromagnetic response data obtained from electromagnetic simulation based on the structure of the electromagnetic metasurface is of high quality, thus enabling the structural pattern generation model to effectively learn the "structure-electromagnetic response" relationship of the electromagnetic metasurface. This ensures that the structure of the electromagnetic metasurface output by the subsequent structural pattern generation model in the "reverse design" process is accurate, and the generated structural pattern of the electromagnetic metasurface can meet the expected electromagnetic response.

[0059] In one possible embodiment, the structural pattern of the region to be generated includes type identification information for each pixel in the region. See also... Figure 4 The above S104 may include the following steps:

[0060] S201, determine the pixel coverage confirmation strategy corresponding to the basic structure pattern.

[0061] The pixel coverage confirmation strategy for different basic structure patterns can be the same or different. In some possible embodiments, the corresponding pixel coverage confirmation strategy can be determined based on the shape of the basic structure pattern. For example, if the basic structure pattern is annular, the pixel coverage confirmation strategy can be: align the center point of the annulus with the center point of the drawing to be designed, and determine the pixels covered by the basic structure pattern in the drawing to be designed based on the radius and width of the annulus, thereby determining the pixels covered in the area to be generated. If the basic structure pattern is cross-shaped, the pixel coverage confirmation strategy can be: align the center point of the cross with the center point of the drawing to be designed, and determine the pixels covered by the basic structure pattern in the drawing to be designed based on the length, height, and width of the cross, thereby determining the pixels covered in the area to be generated.

[0062] S202, based on the pixel coverage confirmation strategy and the target size, determine the pixels covered by the basic structure pattern in the area to be generated.

[0063] For example, the target size can be substituted into the pixel coverage confirmation strategy to obtain the pixels covered by the basic structure pattern in the area to be generated.

[0064] In one possible embodiment, the basic structural pattern includes a ring resonator pattern, and the target dimensions include radius and width. The ring resonator pattern can be a solid ring without a hollow core or a hollow ring structure, and the radius of the ring resonator pattern can be the average radius. The width of the ring resonator pattern can be the minimum radius or the difference between the maximum radius and the average radius. By using a structural pattern with a clear physical meaning (resonant ring), the electromagnetic characteristic differences and meaningless fragmented structures caused by generating structural patterns from completely random pixels can be reduced. S202 above may include the following steps: according to a pixel coverage confirmation strategy, obtaining the distance between each pixel in the area to be generated and the center point, and the difference between each distance and the radius; using the difference whose absolute value is less than or equal to the width as the first difference, and the difference whose absolute value is greater than the width as the second difference; determining the pixels corresponding to each first difference as pixels covered by the ring resonator pattern, and determining the pixels corresponding to each second difference as pixels not covered by the ring resonator pattern.

[0065] In other words, when the basic structural pattern is a ring resonator pattern, the distance between each pixel in the area to be generated and the center point of the design pattern is obtained, as well as the difference between each distance and the radius of the ring resonator pattern. If the difference is less than or equal to the width, it means that the pixel is between the inner and outer radii of the ring resonator pattern, that is, the pixel is covered by the ring resonator pattern.

[0066] like Figure 5 and Figure 6 As shown, Figure 6 The image shows the number of pixels covered by the basic structure pattern in the design drawing when the basic structure pattern is a ring resonator pattern. Figure 5 The pixels filled with diagonal lines are the covered pixels. Figure 6 The image shows the pixels covered by the basic structure pattern in the area to be generated; specifically, the pixels filled with diagonal lines are the covered pixels.

[0067] S203, the type identification information of the pixels in the area to be generated that are covered by the basic structure pattern is determined as the metal type identifier, and the type identification information of the pixels that are not covered by the basic structure pattern is determined as the medium type identifier.

[0068] Specifically, the type identification information of all or part of the pixels in the area to be generated that are covered by the basic structure pattern can be determined as metal type identifiers, and the type identification information of the pixels not covered by the basic structure pattern can be determined as medium type identifiers. Determining the type identification information of a pixel as a metal type identifier can be done by setting the pixel value to a first pixel value, and determining the type identification information of a pixel as a medium type identifier can be done by setting the pixel value to a second pixel value. The first pixel value and the second pixel value are different; for example, the first pixel value is 1, indicating metal filling, and the second pixel value is 0, indicating no metal filling, i.e., the second pixel value is empty.

[0069] In the above scheme, by determining the type identification information of the pixels covered by the basic structure pattern in the area to be generated as metal type identification, and determining the type identification information of the pixels not covered by the basic structure pattern as medium type identification, the structural pattern in the area to be generated can be mapped to other initial areas, and the structural patterns in each area can form a resonator with resonance function, which has practical physical meaning.

[0070] In one possible embodiment, S203 may include the following steps: setting the type identification information of pixels covered by the basic structure pattern to a metal type identifier according to a first preset probability; in response to the existence of pixels covered by the basic structure pattern whose type identification information is not set to a metal type identifier, setting the type identification information of pixels that are not set to a metal type identifier to a medium type identifier.

[0071] The first preset probability can be a preset value, for example, the first preset probability can be greater than 50%. That is, if the type identification information of the covered pixel is not set to the metal type identifier with the first preset probability, then the type identifier information of the pixel is directly set to the medium type identifier.

[0072] In the above scheme, by setting the type identification information of the covered pixels to metal type identification with a certain probability, it is possible that the type identification information of some covered pixels is medium type identification, so that the area to be generated can generate more diverse structural patterns, and the final electromagnetic metasurface structural patterns are also more diverse.

[0073] In one possible embodiment, to further enrich the target structural pattern of the electromagnetic metasurface, the following steps may be performed before performing S105 above:

[0074] At least one target seed pixel is determined from several types of seed pixels. Optionally, the target seed pixel includes multiple reference pixels arranged in a regular pattern, such as three adjacent reference pixels with equal row coordinates, or three adjacent reference pixels with equal column coordinates. Then, an update operation is performed on the structural pattern of the region to be generated using one of the target seed pixels in sequence to obtain the updated structural pattern of the region to be generated. Optionally, after performing an update operation using one target seed pixel, another target seed pixel can be randomly determined from several types of seed pixels to perform an update operation, thereby updating the structural pattern of the region to be generated in an iterative manner. For example, it can be iterated three times: after performing an update operation using the first target seed pixel, a second target seed pixel is randomly selected to perform an update operation, and then a third target seed pixel is selected to perform an update operation. The target seed pixel includes multiple reference pixels. The first target seed pixel performs an update operation on the structural pattern of the region to be generated obtained in S104 above, while non-first target seed pixels perform an update operation on the updated structural pattern of the region to be generated obtained from the previous target seed pixel. Please refer to [link to relevant documentation]. Figure 7 The update operation may include the following steps:

[0075] S301, Select a pixel from the latest media region in the region to be generated as the anchor point.

[0076] A media region is the area where the pixel whose type identification information is a media type identifier is located. There can be one or more media regions, and each media region can contain at least one pixel.

[0077] During the update operation of the first target seed pixel, the latest medium region in the region to be generated can be the medium region contained in the structural pattern of the region to be generated obtained in S104 above. During the update operation of a non-first target seed pixel, the latest medium region in the region to be generated is the medium region contained in the updated structural pattern of the region to be generated obtained from the update operation of the previous target seed pixel.

[0078] In one possible embodiment, S301 above may include the following steps:

[0079] First, the latest media region is divided into a first media region and a second media region. In the first media region, no pixel is adjacent to any pixel in the metal region of the region to be generated. In the second media region, each pixel is adjacent to at least one pixel in the metal region of the region to be generated. The metal region of the region to be generated is the area where pixels with the type identifier "metal" are located. It can be understood that the second media region can be located between the first media region and the metal region, or the second media region can be surrounded by the metal region. The number of pixels included in the metal region can be at least one.

[0080] Then, a pixel is selected from the first medium region according to a second preset probability, where the second preset probability is greater than 50%. For example, the second preset probability could be 70%. That is, during the update operation, there is a 70% probability of selecting a pixel from the first medium region as an anchor point. Alternatively, a pixel can be selected from the second medium region as an anchor point according to a third preset probability. The third preset probability plus the second preset probability equals 100%, meaning there is a 30% probability of selecting a pixel from the second medium region as an anchor point.

[0081] Finally, in response to selecting a pixel from the first medium region according to the second preset probability, the selected pixel is used as an anchor point; or, in response to not selecting a pixel from the first medium region according to the second preset probability, a pixel is selected from the second medium region as an anchor point.

[0082] If a pixel is selected from the first medium region, then the selected pixel is used as the anchor point. If no pixel is selected from the first medium region according to the second preset probability, or if a pixel is selected from the second medium region according to the third preset probability, then the pixel selected from the second medium region is used as the anchor point.

[0083] In the above scheme, by prioritizing the selection of anchor points from the first medium region, more of the reference pixels in the area to be generated fall into the medium region, ensuring that more pixels with medium type identifiers are updated to metal type identifiers, resulting in a richer structural pattern in the updated area to be generated.

[0084] S302, add the row coordinates of each reference pixel in the target seed pixel to the row coordinates of the anchor point, and add the column coordinates of each reference pixel to the column coordinates of the anchor point to obtain the pixel point where each reference pixel is located in the area to be generated.

[0085] For example, if the target seed pixel consists of three adjacent reference pixels with equal row coordinates (0,0), (0,1), and (0,2), and the anchor point has coordinates (3,3), then the row coordinates of each reference pixel are added to the row coordinates of the anchor point, and the column coordinates are added to the column coordinates of the anchor point, resulting in the corresponding pixel coordinates of (3,3), (3,4), and (3,5). If the target seed pixel consists of three adjacent reference pixels with equal column coordinates (0,0), (1,0), and (2,0), and the anchor point has coordinates (3,3), then the row coordinates of each reference pixel are added to the row coordinates of the anchor point, and the column coordinates are added to the column coordinates of the anchor point, resulting in the corresponding pixel coordinates of (3,3), (4,3), and (5,3). Based on the coordinates of the corresponding pixel of each reference pixel, the pixel located in the area to be generated is selected as the pixel position of each reference pixel in the area to be generated.

[0086] S303, determine the type identification information of each reference pixel in the area to be generated as the metal type identifier.

[0087] For each reference pixel in the area to be generated, if the type identifier of the pixel is a medium type identifier, then the type identifier of the pixel is determined to be a metal type identifier; if the type identifier of the pixel is a metal type identifier, then the type identifier of the pixel is kept as a metal type identifier, thereby obtaining the updated structural pattern of the area to be generated.

[0088] In one possible embodiment, S303 may include the following steps: determining whether each reference pixel has a corresponding pixel in the area to be generated. In response to the condition that each reference pixel has a corresponding pixel in the area to be generated, performing the step of determining the type identification information of the pixel in the area to be generated as a metal type identifier.

[0089] Optionally, in response to the existence of at least one reference pixel that does not have a corresponding pixel in the area to be generated, the update operation is re-performed for the target seed pixel. That is, a pixel is selected again from the latest medium region in the area to be generated as the anchor point, and the row coordinates of each reference pixel in the target seed pixel are added to the row coordinates of the anchor point, and the column coordinates are added to the column coordinates of the anchor point to obtain the pixel position of each reference pixel in the area to be generated. In response to the existence of a corresponding pixel for each reference pixel in the area to be generated, the step of determining the type identification information of the pixel position of each reference pixel in the area to be generated as a metal type identifier is performed.

[0090] In the above scheme, the validity of the update operation is determined by whether each reference pixel has a corresponding pixel in the area to be generated. If the update operation is valid, the type identification information of each reference pixel in the area to be generated is determined as a metal type identifier. This can update the type identification information of the pixels in the medium area of ​​the area to be generated to the metal type identifier as much as possible, thereby improving the richness of the structural pattern of the area to be generated.

[0091] In this process, the last target seed pixel is updated, and the updated structural pattern of the area to be generated is mapped onto other initial areas in the design drawing. These other initial areas are those outside the area to be generated. This mapping can be achieved based on the axial symmetry relationship of each initial area, thus simplifying the design and reducing its complexity.

[0092] In the above scheme, by using target seed pixels to iteratively evolve and optimize the structural pattern in the region to be generated, more complex structural changes are explored, making the target structural pattern of the electromagnetic metasurface obtained later more diverse.

[0093] In one possible embodiment, please refer to Figure 8 The above S105 may include the following steps:

[0094] S401, map the structural pattern within the region to be generated to each of the several initial regions other than the region to be generated, to obtain the initial structural pattern of the electromagnetic metasurface.

[0095] S402, it was determined that there are two metal regions that are separated from each other in the initial structural pattern.

[0096] The metallic region refers to the area where pixels with metal type identifiers are located. The initial structural pattern of the electromagnetic metasurface may still have some issues, such as the lack of closed loops. If the resonator lacks closed loops, there may be risks of functional failure. The separation of two metallic regions can be considered as each pixel in one metallic region being non-adjacent to any pixel in the other metallic region.

[0097] S403, Adjust the structural pattern in the area to be generated.

[0098] One way to adjust the structural pattern in the area to be generated is to change the type identification information of some pixels in the medium region of the structural pattern to a metal type identification. Specifically, this can be done by selecting pixels from the second medium region of the area to be generated, or by selecting at least one pixel from both the first and second medium regions and adjusting it to a metal type identification.

[0099] S404: Map the adjusted structural pattern of the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface.

[0100] In the above scheme, when there are mutually separated metal regions in the initial structural pattern of the electromagnetic metasurface, the structural pattern in the region to be generated is adjusted, and the target structural pattern of the electromagnetic metasurface obtained by mapping the adjusted structural pattern has a unique metal region, so that the function of the electromagnetic metasurface is guaranteed.

[0101] In one possible embodiment, considering that after mapping the structural pattern of the region to be generated to other initial regions, the structural patterns of each region may be disconnected or broken, the following step can also be performed: simulation is performed only if it is determined that the metal regions in the structural patterns of each region can be connected to form a ring and the number of such rings is one.

[0102] First, the region where the pixel with the type identifier information of metal is located in the target structure pattern of the electromagnetic metasurface is regarded as the metal region.

[0103] Next, the number and shape of the metal regions in the target structural pattern are determined. If there is only one metal region, the shape can be a connected ring or a disconnected ring. If there are multiple metal regions, the shape can be multiple separate metal regions. After processing steps S401 to S404, the target structural pattern of the electromagnetic metasurface has exactly one metal region, meaning that multiple separate metal regions form a single, interconnected, and unique overall metal region.

[0104] The steps involve simulating the target structural pattern to obtain electromagnetic response data, given that the target structural pattern contains only one metal region of a connected ring shape.

[0105] If the target structural pattern contains only one metal region and its shape is a non-connected ring, a non-connected ring cannot form an effective resonant current path. Therefore, the target structural pattern can be discarded and not simulated. Alternatively, the structural pattern in the region to be generated can be adjusted and then mapped to other initial regions to obtain the final structural pattern of the electromagnetic metasurface. Then, the final structural pattern of the electromagnetic metasurface can be simulated.

[0106] In the above scheme, by determining that the number of metal regions in the target structural pattern is one and that the shape is a connected ring before conducting the simulation, the quality of the training dataset can be further guaranteed.

[0107] In one possible embodiment, the structural pattern generation model can be used for the automated design of various electromagnetic devices, such as reflective / transmissive metasurface unit design, specifically for designing superlenses, holographic imaging devices, polarization control devices, and electromagnetic absorbing / shielding materials. The structural pattern generation model can also be applied to antenna design, specifically for designing microstrip patch antennas and reflective / transmissive array antennas to achieve high gain, multi-beam, and beamforming capabilities. The structural pattern generation model can also be applied to RF / microwave circuit design, specifically for designing filters, power dividers, and couplers. Furthermore, the structural pattern generation model can be applied to antenna array decoupling design, specifically for designing decoupling structures in compact antenna arrays to improve array isolation.

[0108] In one possible embodiment, the step of generating the target structural pattern of the electromagnetic metasurface is performed by the unit generation model. The generation process of the training dataset in this scheme may include the following steps: receiving a model training request, which includes the number N of target structural patterns to be generated in the training sample set. Then, an empty list is initialized, where the type identification information of each pixel is empty. This empty list can be the design drawing of the electromagnetic metasurface. The unit generation model sequentially executes steps S101 to S105 to obtain the target structural pattern of the electromagnetic metasurface. Then, by connecting to the interface library of the simulation tool, the simulation tool is called to simulate the target structural pattern of the electromagnetic metasurface to obtain electromagnetic response data. Specifically, the target structural pattern can be input into the simulation script to obtain the corresponding electromagnetic response data. The target structural pattern and each electromagnetic response data are paired, and the paired target structural pattern and electromagnetic response data constitute a training sample. The training sample is added to the list storing the training dataset. After obtaining N training samples, the list storing the training dataset is saved as a persistent file, for example, as a file with the .csv or .npy extension.

[0109] In the above scheme, the target structural pattern of the electromagnetic metasurface is generated first by using the unit generation model, and then the corresponding interface is called to simulate the target structural pattern using simulation tools. Based on the target structural pattern and electromagnetic response data, a training dataset is constructed, realizing the fully automated process of constructing the structural pattern of the electromagnetic metasurface, simulation, and training dataset construction.

[0110] The structural pattern generation model can be a generative adversarial network (GAN), a variational autoencoder (VAE) model, etc. The method of training the structural pattern generation model using the training dataset can be as follows: use the electromagnetic response data in the training dataset as data, calculate the loss based on the difference between the structural pattern output by the structural pattern generation model and the target structural pattern in the training dataset, and then adjust the parameters in the structural pattern generation model based on the loss to obtain the trained structural pattern generation model.

[0111] The following describes a training device for generating structural patterns for electromagnetic metasurfaces, which corresponds to the method described above.

[0112] Please see Figure 9 , Figure 9This is one of the functional unit block diagrams of a structural pattern generation model training device for electromagnetic metasurfaces provided in this application. The structural pattern generation model training device 500 is applied to the server in the structural pattern generation model training system. The structural pattern generation model training device 500 includes: a data acquisition unit 501, a region selection unit 502, a pattern selection unit 503, a pattern generation unit 504, a mapping unit 505, a simulation unit 506, a dataset determination unit 507, and a training unit 508. The data acquisition unit 501 is used to acquire the design drawing of the electromagnetic metasurface; the region selection unit 502 is used to select one of the initial regions in the design drawing as the region to be generated; the pattern selection unit 503 is used to determine the basic structural pattern of the electromagnetic metasurface based on the resonator pattern selected from several types of resonator patterns; and the pattern selection unit 503 is used to determine the basic structural pattern of the electromagnetic metasurface based on the size selected from a preset size range. The system comprises the following components: a target size for the basic structural pattern; a pattern generation unit 504 for determining the structural pattern within the region to be generated based on the basic structural pattern and the target size; a mapping unit 505 for mapping the structural pattern within the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface; wherein the target structural pattern can coincide before and after rotating 90° around the center point of the design drawing and is an axisymmetric pattern; a simulation unit 506 for simulating the target structural pattern to obtain electromagnetic response data; a dataset determination unit 507 for obtaining a training dataset based on the target structural pattern and electromagnetic response data; and a training unit 508 for training the structural pattern generation model using the training dataset, wherein the trained structural pattern generation model is used to generate the structural pattern of the electromagnetic metasurface corresponding to the input reference electromagnetic response data.

[0113] In one possible embodiment, the structural pattern of the region to be generated includes type identification information for each pixel in the region to be generated. The pattern generation unit 504 determines the structural pattern in the region to be generated based on the basic structural pattern and the target size, including: determining a pixel coverage confirmation strategy corresponding to the basic structural pattern; determining the pixels covered by the basic structural pattern in the region to be generated based on the pixel coverage confirmation strategy and the target size; determining the type identification information of the pixels covered by the basic structural pattern in the region to be generated as metal type identification, and determining the type identification information of the pixels not covered by the basic structural pattern as medium type identification.

[0114] In one possible embodiment, the basic structure pattern includes a ring resonator pattern, and the target size includes a radius and a width. The pattern generation unit 504 determines the pixels covered by the basic structure pattern in the area to be generated according to the pixel coverage confirmation strategy and the target size, including: obtaining the distance between each pixel in the area to be generated and the center point, and the difference between each distance and the radius, according to the pixel coverage confirmation strategy; taking the difference whose absolute value is less than or equal to the width as the first difference, and the difference whose absolute value is greater than the width as the second difference; determining the pixels corresponding to each first difference as pixels covered by the ring resonator pattern, and determining the pixels of each second difference as pixels not covered by the ring resonator pattern.

[0115] In one possible embodiment, the pattern generation unit 504 determines the type identification information of some pixels covered by the basic structure pattern in the area to be generated as metal type identification, including: setting the type identification information of the pixels covered by the basic structure pattern as metal type identification according to a first preset probability; and in response to the existence of pixels covered by the basic structure pattern whose type identification information is not set as metal type identification, setting the type identification information of the pixels that are not set as metal type identification as medium type identification.

[0116] In one possible embodiment, the pattern generation unit 504 is further configured to: take the region where the pixel with type identification information of metal type identification is located in the target structure pattern of the electromagnetic metasurface as the metal region; determine the number and shape of the metal regions in the target structure pattern; and, in response to the fact that the number of metal regions in the target structure pattern is one and the shape is a connected ring, perform the step of simulating the target structure pattern to obtain electromagnetic response data.

[0117] In one possible embodiment, before mapping the structural pattern within the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface, the pattern generation unit 504 is further configured to: determine at least one target seed pixel from several types of seed pixels, the target seed pixel including multiple reference pixels; sequentially perform an update operation on the structural pattern of the region to be generated using one of the target seed pixels to obtain the updated structural pattern of the region to be generated, the update operation including: selecting a pixel from the latest medium region in the region to be generated as an anchor point, the medium region being the region where the pixel with the type identification information is a medium type identifier; adding the row coordinates of each reference pixel in the target seed pixel to the row coordinates of the anchor point and the column coordinates to the column coordinates of the anchor point to obtain the pixel where each reference pixel is located in the region to be generated; and determining the type identification information of each reference pixel in the region to be generated as a metal type identifier.

[0118] In one possible embodiment, the pattern generation unit 504 selects a pixel from the latest medium region in the region to be generated as an anchor point, including: dividing the latest medium region into a first medium region and a second medium region, wherein each pixel in the first medium region is not adjacent to any pixel in the metal region of the region to be generated, and each pixel in the second medium region is adjacent to at least one pixel in the metal region of the region to be generated, wherein the metal region of the region to be generated is the region where the pixel whose type identification information is metal type identifier is located; selecting a pixel from the first medium region according to a second preset probability, wherein the second preset probability is greater than 50%; in response to selecting a pixel from the first medium region according to the second preset probability, using the selected pixel as an anchor point; or, in response to not selecting a pixel from the first medium region according to the second preset probability, selecting a pixel from the second medium region as an anchor point.

[0119] In one possible embodiment, the pattern generation unit 504 determines the type identification information of each reference pixel in the area to be generated as a metal type identifier, including: determining whether each reference pixel has a corresponding pixel in the area to be generated; and in response to the fact that each reference pixel has a corresponding pixel in the area to be generated, performing the step of determining the type identification information of each reference pixel in the area to be generated as a metal type identifier.

[0120] In one possible embodiment, the mapping unit 505 maps the structural pattern within the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface, including: mapping the structural pattern within the region to be generated to each initial region other than the region to be generated, to obtain the initial structural pattern of the electromagnetic metasurface; determining that there are two metal regions separated from each other in the initial structural pattern, the metal regions being the regions where the pixel points with type identification information of metal type are located; adjusting the structural pattern in the region to be generated; and mapping the adjusted structural pattern in the region to be generated to each initial region to obtain the target structural pattern of the electromagnetic metasurface.

[0121] It is understood that since the method embodiments and the device embodiments are different presentations of the same technical concept, the content of the method embodiment section in this application should be adapted to the device embodiment section in a synchronous manner, and will not be repeated here.

[0122] In the case of using integrated units, please refer to Figure 10 , Figure 10 This is the second functional unit block diagram of a structural pattern generation model training device for electromagnetic metasurfaces provided in this application embodiment. The structural pattern generation model training device 500 is used as a server in a structural pattern generation model training system. Figure 10The structural pattern generation model training device 500 includes a processing module 512 and a communication module 511. The processing module 512 controls and manages the actions of the structural pattern generation model training device 500, for example, executing the steps of the data acquisition unit 501, region selection unit 502, pattern selection unit 503, pattern generation unit 504, mapping unit 505, simulation unit 506, dataset determination unit 507, and training unit 508, and / or other processes for performing the techniques described herein. The communication module 511 is used for interaction between the structural pattern generation model training device 500 and other devices. Figure 10 As shown, the structural pattern generation model training device 500 may also include a storage module 513, which is used to store the program code and data of the structural pattern generation model training device 500.

[0123] The processing module 512 can be a processor or controller, such as a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an ASIC, an FPGA, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The communication module 511 can be a transceiver, RF circuitry, or a communication interface, etc. The storage module 513 can be a memory.

[0124] All relevant content in each scenario involved in the above method embodiments can be referenced from the functional descriptions of the corresponding functional modules, and will not be repeated here. The above-mentioned structural pattern generation model training device 500 can perform the above-mentioned... Figure 1 The method shown is for training a structural pattern generation model for electromagnetic metasurfaces.

[0125] Please see Figure 11 , Figure 11 This is a schematic diagram of the structure of an electronic device provided in this application. For example... Figure 11As shown, the electronic device may include a processor 610, a communications interface 620, a memory 630, and a communication bus 640, wherein the processor 610, communications interface 620, and memory 630 communicate with each other via the communication bus 640. The processor 610 can call logic instructions in the memory 630 to execute the aforementioned training method for the structural pattern generation model of electromagnetic metasurfaces. The electronic device may include a server in the aforementioned structural pattern generation model training system.

[0126] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, 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 the present invention. 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), magnetic disks, or optical disks.

[0127] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the structural pattern generation model training method for electromagnetic metasurfaces provided in the above embodiments.

[0128] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described methods for training structural pattern generation models for electromagnetic metasurfaces.

[0129] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0130] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the methods described in the above method embodiments, wherein the computer includes an electronic device.

[0131] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0132] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0134] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0135] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0136] If the aforementioned integrated units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 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 memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0137] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0138] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for training a structural pattern generation model for electromagnetic metasurfaces, characterized in that, A server applied to a structural pattern generation model training system, the method comprising: Obtain the design drawing of the electromagnetic metasurface; One of the initial regions in the design drawing is taken as the region to be generated. The basic structural pattern of the electromagnetic metasurface is determined based on a resonator pattern selected from several types of resonator patterns; and the target size of the basic structural pattern is determined based on a size selected from a preset size range. Based on the basic structural pattern and the target size, determine the structural pattern within the area to be generated; The structural pattern within the region to be generated is mapped to each of the initial regions to obtain the target structural pattern of the electromagnetic metasurface; wherein the target structural pattern can coincide before and after rotating 90° around the center point of the design drawing and the target structural pattern is an axisymmetric pattern; The target structural pattern was simulated to obtain electromagnetic response data; Based on the target structural pattern and the electromagnetic response data, a training dataset is obtained; Using the training dataset, a structural pattern generation model is trained. The trained model generates a structural pattern for an electromagnetic metasurface corresponding to the input reference electromagnetic response data. Before mapping the structural pattern within the region to be generated to each of the initial regions to obtain the target structural pattern of the electromagnetic metasurface, the method further includes: At least one target seed pixel is determined from a number of seed pixels, wherein the target seed pixel includes a plurality of reference pixels; The structure pattern of the region to be generated is updated by sequentially using one of the target seed pixels to obtain the updated structure pattern of the region to be generated. The update operation includes: Select a pixel from the latest media region in the region to be generated as an anchor point. The media region is the region where the pixel with the type identification information is the media type identifier. The row coordinates of each reference pixel in the target seed pixel are added to the row coordinates of the anchor point, and the column coordinates are added to the column coordinates of the anchor point to obtain the pixel point where each reference pixel is located in the region to be generated. The type identification information of each reference pixel in the region to be generated is determined as a metal type identifier.

2. The method according to claim 1, characterized in that, The structural pattern of the region to be generated includes type identification information for each pixel in the region to be generated. Determining the structural pattern within the region to be generated based on the basic structural pattern and the target size includes: Determine the pixel coverage confirmation strategy corresponding to the basic structure pattern; Based on the pixel coverage confirmation strategy and the target size, determine the pixels covered by the basic structure pattern in the area to be generated; The type identification information of the pixels in the area to be generated that are covered by the basic structure pattern is determined as metal type identification, and the type identification information of the pixels that are not covered by the basic structure pattern is determined as medium type identification.

3. The method according to claim 2, characterized in that, The basic structural pattern includes a ring resonator pattern, and the target size includes a radius and a width. Determining the number of pixels covered by the basic structural pattern in the area to be generated, based on the pixel coverage confirmation strategy and the target size, includes: According to the pixel coverage confirmation strategy, the distance between each pixel in the area to be generated and the center point, and the difference between each distance and the radius are obtained; The difference whose absolute value is less than or equal to the width is taken as the first difference, and the difference whose absolute value is greater than the width is taken as the second difference. The pixels corresponding to each of the first differences are determined as pixels covered by the ring resonator pattern, and the pixels corresponding to each of the second differences are determined as pixels not covered by the ring resonator pattern.

4. The method according to claim 2, characterized in that, The step of determining the type identification information of the pixels in the region to be generated that are covered by the basic structure pattern as metal type identification includes: The type identification information of the pixels covered by the basic structure pattern is set as the metal type identifier according to a first preset probability; In response to the existence of pixels whose type identification information is not set to the metal type identifier when the underlying structure pattern is covered, the type identification information of the pixels that are not set to the metal type identifier is set to the medium type identifier.

5. The method according to claim 4, characterized in that, The method further includes: The region where the pixel with the type identification information of the target structure pattern of the electromagnetic metasurface is the metal type identifier is located is taken as the metal region. Determine the number and shape of the metal regions in the target structural pattern; In response to the fact that the number of metal regions in the target structural pattern is one and the shape is a connected ring, the step of simulating the target structural pattern to obtain electromagnetic response data is performed.

6. The method according to claim 1, characterized in that, Selecting a pixel from the latest media region in the region to be generated as an anchor point includes: The latest medium region is divided into a first medium region and a second medium region. Each pixel in the first medium region is not adjacent to any pixel in the metal region of the region to be generated. Each pixel in the second medium region is adjacent to at least one pixel in the metal region of the region to be generated. The metal region of the region to be generated is the region where the pixel whose type identification information is the metal type identifier is located. A pixel is selected from the first medium region according to a second preset probability, wherein the second preset probability is greater than 50%. In response to selecting a pixel from the first medium region according to the second preset probability, the selected pixel is used as the anchor point; Alternatively, in response to the fact that no pixel is selected from the first medium region according to the second preset probability, a pixel is selected from the second medium region as the anchor point.

7. The method according to claim 1, characterized in that, The step of determining the type identification information of each reference pixel in the region to be generated as a metal type identifier includes: Determine whether each of the reference pixels has a corresponding pixel in the area to be generated; In response to the fact that each of the reference pixels has a corresponding pixel in the area to be generated, the step of determining the type identification information of the pixel in the area to be generated as a metal type identifier is performed.

8. The method according to any one of claims 1 to 5, characterized in that, The step of mapping the structural pattern within the region to be generated to each of the initial regions to obtain the target structural pattern of the electromagnetic metasurface includes: The structural pattern within the region to be generated is mapped to each of the initial regions other than the region to be generated, to obtain the initial structural pattern of the electromagnetic metasurface; It was determined that there are two metal regions that are separated from each other in the initial structural pattern. The metal region is the area where the pixel with the type identification information is metal type. Adjust the structural pattern in the area to be generated; The adjusted structural pattern of the region to be generated is mapped to each of the initial regions to obtain the target structural pattern of the electromagnetic metasurface.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for training a structural pattern generation model for electromagnetic metasurfaces as described in any one of claims 1-8.

Images