A supervised, evolutionary learning algorithm-driven focused metasurface system that closely resembles the human eye.

Abstract

The present invention discloses a human-eye-like focusing metasurface system driven based on a teacher-involved evolutionary learning algorithm, which is applicable to the technical field of intelligent electromagnetic metasurfaces. The system includes a transmissive metasurface, an array probe, a focusing guide module, and an evolutionary learning module. When an external electromagnetic wave signal passes through the transmissive metasurface, the array probe installed behind the transmissive metasurface detects the external electromagnetic wave data, and the focusing guide module and the evolutionary learning module analyze it and output an adjustment strategy for the transmissive metasurface. The state of the transmissive metasurface changes, the array probe collects new data, the focusing guide module and the evolutionary learning module further analyze the intensity and characteristics of the external electromagnetic wave data, output the next adjustment command, and repeat the above process until focusing at the specified position. The present invention can achieve intelligent focusing at any position under multiple electromagnetic environments, does not require artificial adjustment, and can be used flexibly.

Description

[Technical Field]

[0001] The present invention relates to the technology of intelligent electromagnetic metasurfaces, and more specifically to a focused metasurface system that closely resembles the human eye and is driven based on a supervised evolutionary learning algorithm. [Background technology]

[0002] Optical focusing is an ancient topic that has existed for thousands of years and has widespread applications in biology, photonics, and physics, and scientists' research enthusiasm remains undiminished. Conventional optical focusing lenses are typically manufactured based on various heavy substrates, and by rationally designing the microstructure, they form a user-defined beam intensity or shape. The emergence of electromagnetic metasurfaces offers the possibility of miniaturization and integration of optical lenses. Metasurfaces are artificially designed structures composed of a series of subwavelength unit structures. By devising subwavelength elements and spatial layouts, researchers have developed devices with multiple functions such as beam polarization, focusing, and imaging. Of the many functional devices, metalenses are the most widely applied. Compared to conventional heavy lenses, metalenses can focus incident light in a more compact size.

[0003] Over the past decade, scholars have designed numerous metasurface focusing systems (metalens) to achieve features such as broadband, achromaticity, and high efficiency. Achieving adaptive focusing of electromagnetic waves (light) is of paramount importance in areas such as electromagnetic spectral imaging and communication channel enhancement. However, conventional focusing devices can only operate under specific environmental conditions; when the incident environment changes, the structure or unit array needs to be redesigned, lacking adaptive capabilities. While many intelligent optical devices have been created through the combination of artificial intelligence (deep learning) and metasurfaces, their success heavily depends on the quantity and quality of available training data, and requires prior information about the environment. A single deep learning method may fail when faced with the task of focusing in rapidly changing environments. The human eye is a perfect focusing system, highly adaptable to environmental changes, allowing us to perceive over 80% of environmental information with our eyes. Designing a naturally intelligent focusing system that automatically focuses electromagnetic waves (light) in different environments, like the human eye, would significantly simplify equipment design and facilitate applications. Therefore, how to provide a metasurface system that can adapt to intelligent focusing is a problem that those skilled in the art should urgently address. [Overview of the project] [Means for solving the problem]

[0004] In view of this, the present invention provides a focusing metasurface system that closely resembles the human eye, driven by a supervised evolutionary learning algorithm. By utilizing supervised evolutionary learning as the basic algorithmic framework, when electromagnetic waves are randomly incident on the metasurface, the algorithm learns and evolves to adaptively adjust the state of the metasurface and control the trajectory of the transmitted beam. The system can achieve focusing at any specified position in complex electromagnetic environments and can be applied to various applications including electromagnetic spatial imaging, communication signal enhancement, and wireless charging.

[0005] To achieve the above object, the present invention provides the following technical solutions.

[0006] A human-eye-like focusing metasurface system driven based on a supervised-evolutionary learning algorithm includes a transmissive metasurface, an array probe, a focusing guide module, and an evolutionary learning module. When an external electromagnetic wave signal passes through the transmissive metasurface, the array probe installed behind the transmissive metasurface detects the external electromagnetic wave data. The focusing guide module and the evolutionary learning module analyze the external electromagnetic wave and output an adjustment strategy for the transmissive metasurface. The state of the transmissive metasurface changes, the array probe collects new external electromagnetic wave data, the focusing guide module and the evolutionary learning module further analyze the intensity and characteristics of the external electromagnetic wave data, output the next adjustment command, and repeat the above process until focusing on the specified position.

[0007] Optionally, to stop the iteration after focusing on the specified position, the following judgment conditions need to be simultaneously satisfied:

[0008]

Number

[0009]

Number

[0010]

Number

[0011]

number

[0012]

number

[0013] Selectively, the transparent metasurface is composed of an array of multiple unit structures with different switching states, each unit structure consisting of three layers of metal separated by two layers of F4B medium, with two PIN switching diodes welded to the surface of the first layer of metal, and the switching state of the PIN switching diodes is controlled by an externally applied voltage.

[0014] Selectively, the unit structure of the transparent metasurface can allow current to flow in both forward and reverse directions when positive and negative voltages are applied externally, can achieve binarized phase under electromagnetic wave incidence of -50° to 50°, and has a transmittance greater than 95% at the operating frequency.

[0015] Selectively, the focusing guide module is a focusing directional convolutional neural network, and the focusing directional convolutional neural network is an electric field e t Compensation phase Δψ generated by all unit structures in the transparent metasurface t It consists of mappings to

[0016]

number

[0017]

number

[0018] Selectively, the evolutionary learning module uses the compensation phase Δψ t and collected electric field e t Based on the output time t+1, the voltage of each unit state of the transparent metasurface is adjusted, and the voltage update scheme is as follows:

[0019]

number

[0020]

number

[0021]

number

[0022] Selectively, m t and v t The calculation method is as follows:

[0023]

number

[0024]

number

[0025] As can be seen from the above technical solutions, compared to the prior art, the present invention discloses and provides a focused metasurface system that is similar to the human eye and driven based on a supervised evolutionary learning algorithm, and has the following beneficial effects.

[0026] 1. The adaptive focusing system designed in this invention can achieve intelligent focusing at any position under multiple electromagnetic environments, does not require manual adjustment, and can be used flexibly.

[0027] 2. The present invention constructs a framework for supervised-evolutionary learning algorithms that can adaptively and intelligently adjust output schemes under different environments, thereby compensating for the shortcomings of conventional machine learning in its inability to adapt to different environments.

[0028] 3. The subwavelength metasurface structures designed in this invention are small in volume, easy to manufacture, and easy to integrate and realize.

[0029] 4. The high-transmittance metasurface unit designed in this invention achieves a transmittance of 95% or more under different incidence angles, resulting in high transmittance and stable performance.

[0030] To more clearly illustrate embodiments of the present invention or technical solutions of the prior art, the drawings that may be used in describing embodiments or prior art will be briefly described below. As will be apparent, the drawings described below are merely embodiments of the present invention, and those skilled in the art can obtain other drawings based on the provided drawings without requiring any creative work. [Brief explanation of the drawing]

[0031] [Figure 1] Figure 1 is a schematic diagram of the present invention's focusing metasurface system that closely resembles the human eye. [Figure 2] Figure 2 is a schematic diagram of the unit structure of the transparent metasurface of the present invention. [Figure 3] Figure 3 shows the transmission amplitude and phase response curves of the unit structure of the transmission metasurface of the present invention. [Figure 4] Figure 4 is a flowchart of the supervised-evolutionary learning algorithm of the present invention. [Figure 5] Figure 5 is a three-dimensional schematic diagram of the electric field in single-source electromagnetic data of the present invention. [Figure 6] Figure 6 is a schematic diagram of one-dimensional electric field data from single-source electromagnetic data according to the present invention. [Figure 7] Figure 7 is a schematic diagram of the single-source electromagnetic focusing result of the present invention. [Figure 8] Figure 8 is a three-dimensional schematic diagram of the electric field in the double-source electromagnetic data of the present invention. [Figure 9] Figure 9 is a schematic diagram of one-dimensional electric field data using double-source electromagnetic data according to the present invention. [Figure 10] Figure 10 is a schematic diagram of the double-source electromagnetic focusing result of the present invention. [Figure 11] Figure 11 is a three-dimensional schematic diagram of the electric field in random scattering electromagnetic data according to the present invention. [Figure 12] Figure 12 is a schematic diagram of the one-dimensional electric field data in random scattering electromagnetic data according to the present invention. [Figure 13] Figure 13 is a schematic diagram of the random scattering electromagnetic focusing results of the present invention. [Modes for carrying out the invention]

[0032] The technical solutions of the embodiments of the present invention will be described clearly and completely below with reference to the drawings of the embodiments of the present invention, and it will be clear that the embodiments described are only a part of the embodiments of the present invention, not all of them. All other embodiments obtained by those skilled in the art without requiring any creative work based on the embodiments of the present invention are all within the scope of the protection of the present invention.

[0033] Embodiments of the present invention disclose a human-eye-like focusing metasurface system driven on a supervised evolutionary learning algorithm, which includes a transmissive metasurface, an array probe, a focusing guide module, and an evolutionary learning module, as shown in Figure 1. When an external electromagnetic wave signal passes through the transmissive metasurface, the array probe, positioned behind the transmissive metasurface, detects the external electromagnetic wave data. The focusing guide module and the evolutionary learning module analyze the external electromagnetic wave and output an adjustment strategy for the transmissive metasurface. As the state of the transmissive metasurface changes, the array probe collects new external electromagnetic wave data. The focusing guide module and the evolutionary learning module further analyze the intensity and characteristics of the external electromagnetic wave data and output the next adjustment command, repeating the above process until the metasurface is focused to a specified position.

[0034] Specifically, the principle of an adaptive focusing system similar to that of the human eye is as follows: Incident light passes through the lens and reaches photosensitive cells on the retina. At this point, the light signal is converted into an electrical signal and perceived by the human brain. The brain analyzes the signal and then outputs an adjustment strategy for the lens via the nervous system. At this time, the ciliary muscle / lens of the eye expands and contracts based on the system's commands, changing the intensity of the signal received by the photosensitive cells. The human brain further analyzes the signal characteristics at this time and outputs adjustment commands, repeating the above process until the light is completely focused on the retina. In this invention, the transmission metasurface functions as the lens, the array probe functions as the photosensitive cells, and the focusing guide module and evolutionary learning module perform analysis as the human brain.

[0035] Specifically, the framework of the supervised-evolutionary learning algorithm, consisting of a focusing guide module and an evolutionary learning module, a transparent metasurface structure, and an external electromagnetic environment constitute a closed-loop adaptive iterative system of "environmental data collection - algorithm prediction - metasurface adjustment - environmental data collection". The external electromagnetic environment includes a single-source electromagnetic environment with incident light from any direction, a multi-source electromagnetic environment, and an electromagnetic environment with unknown scatterers. The supervised-evolutionary learning algorithm consists of a supervised learning process and an evolutionary learning process. The core of the supervised learning is a convolutional neural network, and the core of the evolutionary learning is an adaptive moment estimation (Adam) gradient descent algorithm.

[0036] Furthermore, in order to stop the repetition after converging to the specified position, the following conditions must be met simultaneously:

[0037]

number

[0038]

number

[0039]

number

[0040]

number

[0041]

number

[0042] Furthermore, the transparent metasurface is constructed by arranging multiple unit structures with different switching states, as shown in Figure 2. Each unit structure consists of three layers of metal, separated from the three layers of metal by two layers of F4B medium. Two PIN switching diodes are welded to the surface of the first layer of metal, and the switching state of the PIN switching diodes is controlled by an externally applied voltage.

[0043] Specifically, the adjustable transparent unit structure consists of three metal layers (Cu) and two dielectric layers (relative permittivity 2.65), with an adhesive layer in between having a relative permittivity of 4.4. Two PIN diodes are welded to the surface of the top metal layer, and the diodes are grounded to a common ground through a hole in the middle. As shown in Figure 3, the dashed and solid lines represent the different states of the two diodes, respectively. When the diode is open, its resistance is 2.1Ω and its capacitance is 0fF. When the diode is closed, its resistance is 0Ω and its capacitance is 50fF. The horizontal axis represents frequency, the vertical axis on the left represents phase, and the vertical axis on the right represents amplitude. It can be seen that when the diodes are in different states, their phase is reversed by 180°, and at the operating frequency (5.85GHz~5.95GHz), the amplitude is always greater than -1dB.

[0044] Furthermore, the unit structure of the transparent metasurface can allow current to flow in both forward and reverse directions when positive and negative voltages are connected externally, can achieve binarized phase under electromagnetic wave incidence of -50° to 50°, and has a transmittance greater than 95% at the operating frequency.

[0045] Furthermore, the focusing guide module is a focusing directional convolutional neural network, and the focusing directional convolutional neural network is an electric field e t Compensation phase Δψ generated by all unit structures in the transparent metasurface t It consists of mappings to

[0046]

number

[0047]

number

[0048] Furthermore, the evolutionary learning module uses the compensation phase Δψ t and collected electric field e t Based on the output time t+1, the voltage of each unit state of the transparent metasurface is adjusted, and the voltage update scheme is as follows:

[0049]

number

[0050]

number

[0051]

number

[0052] Furthermore, m t and v t The calculation method is as follows:

[0053]

number

[0054]

number

[0055]

number

[0056] Furthermore, as shown in Figure 4, the supervised-evolutionary learning algorithm as a whole consists of two modules: an operation cycle and an evolutionary cycle. When electromagnetic waves are incident on a tunable metasurface (a), the detection array collects electric field data (b) and transmits the data to a supervised guide network (c), which outputs the tuning phase and transmits it to the control side (d). At this time, the algorithm outputs the tuning scheme for each unit on the metasurface (e), and the state of the metasurface changes. This process is repeated until the focus point task is completed. In the supervised-evolutionary process, the algorithm simultaneously collects traversed data to facilitate acceleration of subsequent iterations (f).

[0057] Furthermore, the focusing results at different positions of the adjustable focusing metasurface under different electromagnetic environments are as follows: Figures 5, 6, and 7 show three-dimensional schematic diagrams of the electric field at the first and end of iterations under a single-source electromagnetic environment, a comparison of one-dimensional electric field data (theoretical, first iteration, end of iteration), and focusing results at different focusing positions, respectively. Figures 8, 9, and 10 show three-dimensional schematic diagrams of the electric field at the first and end of iterations under a double-source electromagnetic environment, a comparison of one-dimensional electric field data (theoretical, first iteration, end of iteration), and focusing results at different focusing positions, respectively. Figures 11, 12, and 13 show three-dimensional schematic diagrams of the electric field at the first and end of iterations under an electromagnetic environment with scattering obstacles, a comparison of one-dimensional electric field data (theoretical, first iteration, end of iteration), and focusing results at different focusing positions, respectively. It can be seen that the supervised-evolutionary learning algorithm proposed in this invention has strong adaptive and learning capabilities under different electromagnetic environments and can achieve focusing at any specified position.

[0058] Each example in this specification is described step by step, with each example focusing on its differences from the others, and any identical or similar parts between examples should be referenced to one another.

[0059] Those skilled in the art can implement or use the present invention based on the above description of the disclosed embodiments. Various modifications to these embodiments are obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the invention. Accordingly, the present invention is not limited to these embodiments shown herein, but rather conforms to the broadest scope that is consistent with the principles and novel features disclosed herein.

Claims

1. A focused metasurface system, A transparent metasurface composed of multiple unit structures with different switching states arranged in a sequence, An array probe in which elements for detecting incident electromagnetic waves are arranged, A focusing guide module, which is a focusing directional convolutional neural network, It includes an evolutionary learning module consisting of a supervised learning process using a convolutional neural network and an evolutionary learning process using an adaptive moment estimation (ADAM) gradient descent algorithm. The aforementioned transparent metasurface consists of three metal units, separated from each other by two layers of glass-filled PTFE (polytetrafluoroethylene) medium, with two PIN switching diodes welded to the surface of the first layer of metal. The focusing guide module consists of a mapping from the electric field et to the compensation phase Δψt generated by all the unit structures in the transmission metasurface, The evolutionary learning module adjusts the voltage of each unit state of the transmitted metasurface based on the compensation phase Δψt and the output time t+1 of the collected electric field et. The focusing guide module and the evolutionary learning module constitute a supervised evolutionary learning algorithm, including an operational cycle and an evolutionary cycle. When an external electromagnetic wave signal passes through the transparent metasurface, an array probe placed behind the transparent metasurface detects the external electromagnetic wave data. In the aforementioned operating cycle, the focusing guide module and the evolutionary learning module analyze the external electromagnetic waves detected by the array probe and output adjustment commands to the transmission metasurface. The state of the transparent metasurface changes as a result of this adjustment command. The array probe collects new external electromagnetic wave data, In the evolutionary cycle described above, the focusing guide module and the evolutionary learning module further analyze the intensity and characteristics of newly collected external electromagnetic wave data, output the next adjustment command, and update the convolutional neural networks of the focusing guide module and the evolutionary learning module. A focusing metasurface system characterized by repeating the above process until it converges to a specified position.

2. In order to stop the repetition after converging to a specified position, the following conditions must be met simultaneously: [Math 1] [Math 2] [Math 3] where \(r(E\) Ｔｈｅｏｒｙ , E\) Ｔｅｓｔ ) is the effective correlation coefficient between the theoretical electric field \(E\) Ｔｈｅｏｒｙ and the test electric field \(E\) Ｔｅｓｔ , \(C(E\) Ｔｈｅｏｒｙ , E\) Ｔｅｓｔ ) is the covariance between the theoretical electric field \(E\) Ｔｈｅｏｒｙ and the test electric field \(E\) Ｔｅｓｔ , \(Var[E\) Ｔｈｅｏｒｙ is the variance of the theoretical electric field \(E\) Ｔｈｅｏｒｙ , \(Var[E\) Ｔｅｓｔ is the variance of the test electric field \(E\) Ｔｅｓｔ , \(\rho\) ｍ is the main lobe energy ratio of the electromagnetic wave, [Math 4] This is the main lobe energy of electromagnetic waves, [Math 5] ρ is the measured total electric field energy. ｓ E is the ratio of the total side lobe energies of electromagnetic waves. Ｓｉｄｅ＿ｌｏｂｅ The focusing metasurface system according to claim 1, characterized in that is the side lobe energy of an electromagnetic wave.

3. The focusing metasurface system according to Claim 1, characterized in that the switching state of the PIN switching diode is controlled by an externally applied voltage.

4. The focusing metasurface system according to claim 3, characterized in that the unit structure of the transparent metasurface can allow current to flow in forward and reverse directions when positive and negative voltages are connected externally, can achieve binarized phase under the incidence of electromagnetic waves from -50° to 50°, and has a transmittance greater than 95% at the operating frequency.

5. The focused directional convolutional neural network is [Math 6] During the ceremony, 【Number 7】 θ is the focusing compensation phase that the i-th unit structure must satisfy, i ∈ [1, N], and θ is the network parameter when training a focusing directional convolutional neural network, f θ (e ｔ ) is the electric field e ｔ The input is Δψ ｔ The focused metasurface system according to claim 3, characterized in that it is a neural network that is trained to be such as.

6. The evolutionary learning module has the following voltage update scheme: [Number 8] In the formula, u ｔ and u ｔ＋１ α and ε represent the adjustment voltages at time t and time t+1, respectively, and β are learning parameters. １ , β ２ These are the damping coefficients for the first moment and the second moment, respectively. [Number 9] is β １ It is t raised to the power of, [Number 10] is β ２ It is t raised to the power of m ｔ is the first moment of the gradient at time t, and v ｔ The focusing metasurface system according to claim 5, characterized in that is the second moment of the gradient at time t.

7. I understand ｔ and v ｔ The calculation method is as follows: [Math 11] [Math 12] In the formula, g ｔ m is the gradient. ｔ－１ is the first moment of the gradient at time t-1, and v ｔ－１ The focusing metasurface system according to claim 6, characterized in that is the second moment of the gradient at time t-1.

Images