Metasurface device for high-dimensional light field detection and detection method thereof

Abstract

The invention discloses a metasurface device for high-dimensional light field detection and a detection method of the metasurface device, and belongs to a light field detection technology. According to the metasurface device, a sub-wavelength slit array structure arranged along an Archimedes spiral line is etched on a metal film, and a transmission phase related to frequency and a geometric phase related to polarization are decoupled by utilizing the spatial dispersion and frequency dispersion characteristics of the array structure; therefore, the frequency and the polarization of the incident light are mapped to the topological charge and the composition coefficient of the plasmon vortex light. Through cooperation of the metasurface device and the neural network, the incident frequency and polarization information of the light field to be measured can be accurately extracted from the vortex light interference pattern, and synchronous analysis of multi-dimensional parameters of the light field is realized. The metasurface device not only can cover visible light, near-infrared and terahertz wave bands and even microwave wave bands, but also can extract multi-dimensional information of a light field, can realize accurate regulation and control of the light field on a micron scale through monolithic integration, and provides a solution for a compact optical system.

Description

Technical Field

[0001] The present invention relates to high-dimensional light field detection technology based on deep neural networks, and specifically to a measurement method that uses a metasurface to decouple the frequency-related transmission phase and the polarization-related geometric phase, thereby mapping the frequency and polarization of the incident light to the plasmon vortex light topological charge and composition coefficient, and finally uses a neural network to accurately extract the incident frequency and polarization information from the vortex light interference pattern. Background Art

[0002] In recent years, the rapid development of optical communications, aerospace remote sensing, target recognition, and autonomous driving has placed higher demands on light field perception and detection, as well as the miniaturization and integration of optical systems. Light fields carry information in multiple dimensions, including intensity, polarization, frequency, and phase. Spectral and polarization information can reveal the composition of materials and the surface morphology of objects. Simultaneously capturing multidimensional optical information is crucial for a comprehensive understanding of material properties and information transmission. Furthermore, acquiring more comprehensive optical information can provide data support for intelligent information processing, enabling rapid identification of stealthy targets or weak signals, and helping to expand applications in optical communications, biomedical characterization, remote sensing, and many other fields. Therefore, the detection of high-dimensional light field information is highly desirable in many applications. However, integrating spectral and polarization detection functions into a single device still presents significant process difficulties and performance bottlenecks. Most reported photodetectors can only capture single-dimensional optical information and are unable to accurately detect high-dimensional light field information. This limited functionality severely limits the ability to fully analyze complex light field information. Moreover, the detection capability of existing light field detectors often relies on the integration of polarization or wavelength sensitive elements in time or space. Its detection capability is generally proportional to the size of the space occupied or the length of time consumed. The overall volume of the device to achieve high-dimensional light field measurement will be relatively large.

[0003] To address this issue, numerous studies in recent years have attempted to exploit integrated photonics to achieve high-dimensional information acquisition. Graphene-based two-dimensional devices have successfully achieved full Stokes parameter measurement at a few specific wavelengths or wavelength discrimination between two circular polarization states through a unique photovoltage mapping mechanism. However, these methods are only applicable to a few predetermined discrete frequencies or polarization states and, due to graphene's inherent fixed dispersion, cannot be extended to other wavelengths. This limitation also applies to other two-dimensional material platforms. Unlike natural materials with fixed dispersion properties, the rise of metamaterials and metasurfaces, which use artificial subwavelength structures to tailor light-matter interactions, has created unprecedented multidimensional light field manipulation capabilities in free-space or on-chip integrated devices. Recent research demonstrates that all-silicon metasurfaces with focusing capabilities have achieved efficient photodetection of 10 polarization states at five mid-infrared wavelengths, significantly expanding existing high-dimensional detection capabilities. However, these wavelength decoupling methods still sacrifice some degrees of freedom, making them incapable of accurately detecting polarization and intensity information that varies arbitrarily across a wide spectral range. In reality, in many natural scenarios, the light field to be measured may have a polarization state that changes arbitrarily within a wide spectrum. Therefore, there is an urgent need to develop a new type of detector that can fully characterize any light field in the three-dimensional (or four-dimensional) parameter space of intensity-polarization-wavelength. Summary of the Invention

[0004] This paper proposes an innovative idea of ​​using the spatial dispersion and frequency dispersion characteristics of the metasurface to regulate the polarization and spectral response in the vortex light space. It can map all the information of the high-dimensional light field into the surface plasmon (SPP) field distribution, and decode the polarization and spectral information by combining deep learning methods, thus realizing the detection of high-dimensional light information.

[0005] In the first aspect of the present invention, a metasurface device for high-dimensional light field detection is designed, comprising a substrate and a metal film thereon, wherein an array structure consisting of subwavelength slits is etched on the metal film, wherein the subwavelength slits in the array structure are arranged along an Archimedean spiral line ρ1'(θ)=ρ0+m·λ. SPP0 / 2π·θ arrangement, and set the slit rotation angle to α1(θ)=α0+n0·θ; where ρ0 represents the initial radius in the polar coordinate system (ρs,θ), m is an integer, and λ SPP0 is the surface plasmon wavelength corresponding to a certain frequency f0 in the detection band, θ is the angle between the line connecting the center of the slit and the center of the array and the x-axis, α0 and n0 represent the initial angle and rotation factor of the slit, respectively.

[0006] In the nanoarray formed by the arrangement of subwavelength slits along an Archimedean spiral, the subwavelength slits have a length of L, a width of b, a center-to-center distance of d, and a depth equal to the thickness of the metal film. α1(θ) is the angle between the long side of the slit and the positive direction of the y-axis. The values ​​of α0 and n0 must ensure that the slits in the nanoarray do not overlap. α0 can generally be selected from any suitable angle within the range [0, arccos(b / d)], and n0 can generally be a suitable positive integer. m is a structural parameter of the Archimedean spiral and can generally be a suitable positive integer. In a specific embodiment of the present invention, α0 = π / 4, n0 = 1, and m = 1 are set. The frequency f0 is a frequency selected from the detection band and is 0.3 THz, corresponding to a surface plasmon wavelength of 1 mm.

[0007] Furthermore, the substrate can be a transparent substrate of materials such as SiO2, CaF2, Al2O3, or an opaque substrate such as a Si substrate. The metal used in the metal film is preferably a metal material with low surface plasmon loss such as gold (Au), silver (Ag), and aluminum (Al), and the thickness is preferably 150 to 300 nm. For the measurement of terahertz frequency band light fields, it is preferred to additionally cover the overall structure of the device with a dielectric layer (such as polyimide) with a thickness of 10 to 20 μm to enhance the localization ability of terahertz band SPP on the metal surface.

[0008] In one embodiment of the present invention, the metasurface device uses a 2mm thick quartz substrate as its substrate. A 200nm thick gold film is deposited on the substrate surface, and a precisely designed subwavelength slit array structure is etched into the gold film using micro-nanofabrication techniques. Furthermore, an 18μm thick polyimide layer is applied to the top of the entire device to enhance the localization of terahertz-band SPPs on the metal surface.

[0009] When preparing the metasurface device, first select the appropriate slit size (slit length L and width b) according to the working frequency band, and select the appropriate adjacent slit center spacing size d, and the slit height is the thickness of the metal film. In a specific embodiment of the present invention, the terahertz frequency band (about 0.3-1.1THz) is targeted, so the subwavelength slit length is set to L = 200μm, the width is b = 30μm, and the distance between the centers of adjacent slits is d = 100μm. The slit size is set to the subwavelength order here, so the size is smaller than the wavelength corresponding to the maximum frequency. Here, the slit size can be selected between 10-272.7nm, and the parameters are scanned to select a size that has a relatively strong resonance intensity at all wavelengths.

[0010] The slits are then arranged in an Archimedean spiral distribution, and appropriate Archimedean spiral parameters n0,m are selected. The spatial position distribution ρ1'(θ) of the slits and the corresponding rotation angle α1(θ) are designed as follows:

[0011] ρ1'(θ)=ρ0 + m·λ SPP0 / 2π·θ (1)

[0012] α1(θ)=α0+n0·θ (2)

[0013] Where θ is the angle between the line connecting the center of the slit and the center of the array and the x-axis, ρ0 is the initial radius of the Archimedean spiral, α1(θ) is the angle between the long side of the slit and the positive direction of the y-axis, α0 is the set initial rotation angle, n0 and m are the structural parameters of the Archimedean spiral, and λ SPP0 The assigned value is the surface plasmon wavelength corresponding to a certain frequency f0 in the detection band (for example, the wavelength corresponding to f0=0.3THz is 1mm).

[0014] Finally, a pre-designed sub-wavelength slit array structure is etched in the clean area of ​​the metal film.

[0015] Under the above structural parameter settings, when the incident polarization is (1,ae iξ ), when a wave with a frequency of f is irradiated onto the device, the generated SPP field distribution E z The expression is:

[0016]

[0017] In formula (3), e represents a natural constant, J() is a Bessel function, and the introduced vortex optical topological charge is l-1 and l+1, where l = -m·f / f0, where m and f0 are the same as defined above. represents the observation point position, α0 is the initial angle of the slit set above, is the imaginary unit, k SPP is the wave vector of the surface plasmon along the propagation direction, is the phase difference between the y- and x-direction components of the incident polarization field; a = E 0y / E 0x is the amplitude ratio of the y- and x-direction components of the incident polarization field. The above formula shows that under right-handed (left-handed) light, the metasurface designed in this invention only generates l-1 (l+1)-order plasmon vortex beams, and the topological charge of the vortex light is linearly related to the incident frequency f. This is because by decomposing the orbital angular momentum (OAM) of the vortex light generated by the near-field of the measured light irradiating the metasurface, the frequency and polarization of the incident light can be determined.

[0018] In a second aspect of the present invention, a method for detecting high-dimensional light field information based on a specially designed metasurface device is provided, comprising the following steps:

[0019] The metasurface device is illuminated with the light field to be measured, and the surface plasmon field distribution E on the metasurface is collected by a spectrometer. z , use the following formula to calculate E z OAM decomposition can be performed to obtain the complex coefficients M of each order OAM component j .

[0020]

[0021] In formula (4), ρ is the radius of the observation circle from the center of the device array, is the angle between each point on the observation circle and the line connecting the array center and the x-axis, j is the topological charge of the vortex light mode, Furthermore, the incident frequency f = -f0l / m of the light field to be measured can be determined based on the topological charges corresponding to the two highest-intensity vortex light components in the OAM spectrum (theoretically, the l-1 and l+1 orders).

[0022] Furthermore, since any polarized beam can be regarded as a superposition of LCP and RCP, we can determine the component ratio of LCP and RCP by the relative amplitude and phase of the two main vortex light beams, and then infer the polarization state of the incident wave. Based on the above theoretical derivation, we can obtain ae according to the following formula iξ , and then determine the polarization state of the light field to be measured

[0023]

[0024] Among them, M l+1 / M l-1 is the ratio of the coefficients of the two main vortex optical modes.

[0025] The above theoretical method for obtaining light field information has poor robustness, and the resolution of frequency detection is only f0. For frequencies f where p = f / f0 is a non-integer, the metasurface designed by the present invention will produce fractional-order vortex light, which is equivalent to the superposition of countless integer-order vortex light modes. At this time, the calculation method using OAM decomposition mentioned above fails. However, since the plasmon interference patterns generated by the metasurface in the present invention under different frequencies and polarization incidences are uniquely determined, deep learning technology can be introduced to identify these unique vortex light interference patterns, and at the same time, polarization and frequency information can be quickly and accurately extracted from the field distribution pattern.

[0026] First, we obtained a dataset for training the network model through numerical simulations and actual experiments. We used simulated and experimental surface plasmon field intensity distributions encoded with different polarization and wavelengths as the training dataset, with comprehensive coverage of the polarization Stokes parameter and frequency values ​​within the bandwidth. To enhance the robustness of the system and obtain a sufficient training dataset in a short period of time, we used data augmentation to process the simulated and measured patterns, including adding noise and performing image transformations. These operations help mitigate the impact of environmental fluctuations, making the neural network more robust in real-world applications.

[0027] Furthermore, the dataset is divided into a training set and a test set. The neural network is trained using the field distribution in the training set, and the loss function of the multi-output regression task is defined using the mean square error:

[0028]

[0029] Among them S i real is the label of the polarization Stokes parameter corresponding to the field distribution, S i pre is the polarization parameter predicted by the neural network, freq real is the label of the frequency corresponding to the field distribution, freq pre is the frequency predicted by the neural network, α1 and α2 represent adjustable weight coefficients. S1-S3 are the three components of the Stokes parameters used to completely describe the polarization of light: S1 represents the difference between the horizontal polarization intensity and the vertical polarization intensity, S2 represents the difference between the +45° polarization intensity and the -45° polarization intensity, and S3 represents the difference between the right-handed circularly polarized light intensity and the left-handed circularly polarized light intensity.

[0030] Furthermore, after the neural network is trained, it is tested with a test set. The network input is the plasmon field distribution generated by the metasurface device, and the network output is the predicted polarization parameters and frequency. If the test results are good, it can be applied to the judgment of high-dimensional light field information.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. The metasurface device proposed in this invention is flexible and simple to design. High-dimensional information control is achieved through a slit array distributed in a specific mathematical form. Compared with other devices that require gradual iterative optimization based on optimization algorithms to achieve specific functions, the design process is simple and fast. Moreover, the metasurface design method in this invention is universal and can ensure migration applications at other operating wavelengths without the need for de novo optimization of new nanostructures. By adjusting the size and spacing of the slit units, it can cover the visible light, near-infrared, terahertz bands, and even the microwave band, breaking through the narrowband response limitations of traditional optical devices.

[0033] 2. The device designed in this invention is highly integrated. Traditional light field information measurement typically relies on the coordinated action of a large number of optical components. Traditional cascade measurement methods result in bulky systems, complex assembly, and problems such as alignment errors and energy loss. In contrast, the metasurface used in this invention achieves precise control of light fields at the micron scale through a monolithically integrated subwavelength structure, providing a new solution for compact optical systems in terms of integration, functionality, and scalability.

[0034] 3. Compared with traditional light field measurement technology, the present invention can achieve higher-dimensional light field measurement. Traditional light field measurement technology can usually only obtain single-dimensional parameter information of the light field, such as intensity, wavelength or polarization state. This low-dimensional measurement method has the problem of missing information and it is difficult to fully characterize the characteristics of complex light fields. In contrast, the high-dimensional light field measurement technology proposed in the present invention realizes the synchronous analysis of multi-dimensional parameters of the light field through the collaborative design of metasurface and neural network. It can simultaneously extract multi-dimensional information such as amplitude, phase, polarization, wavelength, etc. of the light field, construct a complete light field representation, and break through the limitation that traditional detectors can only respond to intensity or low-dimensional light field information. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a schematic diagram of the principle of the present invention using a metasurface combined with a neural network to achieve high-dimensional light field information measurement.

[0036] Figure 2 Schematic diagram of the present invention using a metasurface to generate frequency- and polarization-dependent vortex light.

[0037] Figure 3 It is a top view (xy plane) of the metasurface device in a specific embodiment of the present invention.

[0038] Figure 4 These are the simulation and experimental measurement results at frequencies of 0.3 THz, 0.6 THz, and 0.9 THz in a specific embodiment of the present invention, as well as the corresponding theoretically predicted polarization information, where (a) shows the polarization states at three frequencies; (b) shows the vortex field distribution formed by the metasurface at six polarization states and three frequencies.

[0039] Figure 5 Schematic diagram of designing a neural network in a specific embodiment of the present invention.

[0040] Figure 6 These are the frequency information measurement results within the frequency range of 0.3-1.2 THz obtained by simulation and experiment in a specific embodiment of the present invention, where (a) is the simulation result and (b) is the experimental result.

[0041] Figure 7These are the polarization information measurement results in the frequency range of 0.3-1.2 THz obtained by simulation and experiment in a specific embodiment of the present invention, where (a) is the simulation result and (b) is the experimental result; the hollow circles represent the actual polarization, and the solid circles represent the polarization parameters predicted by the neural network. DETAILED DESCRIPTION

[0042] The present invention will be described in further detail below with reference to the accompanying drawings through a design example in the terahertz band, so that those skilled in the art can more clearly understand the present invention.

[0043] The core concept of the present invention is to realize a new type of sensor that can simultaneously analyze wavelength and polarization information. The functional schematic diagram of the metasurface device is as follows Figure 1 As shown, when exposed to terahertz waves of different frequencies and polarizations, the metasurface excites surface plasmon vortex light with varying topological charges and scale factors. The topological charge of the vortex light depends on the incident frequency, while the composition coefficients of different vortex light modes depend on the incident polarization. The vortex light interferes to form unique field distribution patterns. These field distribution patterns are fed into a pre-trained residual neural network, which outputs predicted frequency and polarization information for the input light.

[0044] Specifically, if Figure 2 As shown, the metasurface can convert the incident left-handed circularly polarized light (LCP) and right-handed circularly polarized light (RCP) into Bessel vortex beams with topological charges of order l+1 and order l-1, respectively, where the topological charge l = m·f / f0 varies monotonically with the frequency f of the incident light. Incident light of any polarization state can be decomposed into a superposition of LCP and RCP, and the transmitted surface plasmon field distribution appears as an interference superposition of two vortex beams, whose mode purity directly corresponds to the composition coefficients of the LCP and RCP components. Therefore, the polarization state of the incident light can be determined by analyzing the mode purity spectrum extracted from the SPP field distribution, and the topological charge of the dominant orbital angular momentum mode can directly reflect the wavelength information. This dual analysis mechanism realizes high-precision synchronous detection of polarization and wavelength, providing a new solution for compact multi-dimensional spectral sensing.

[0045] The subwavelength slit nanoarray pattern of the designed metasurface is as follows Figure 3As shown, the planar optical device for simultaneously generating polarization- and wavelength-sensitive vortex light fields in the near field is a planar structure. It consists of a SiO2 transparent substrate, a 200nm thick Au film and a rectangular nano-slit array forming a metasurface structure. The subwavelength slit is L = 200μm long, b = 30μm wide, and the center spacing between adjacent slits is d = 100μm. The Au film is on the SiO2 transparent substrate, and the entire structure is covered with an 18μm thick polyimide film. In order to achieve frequency-polarization controllable polariton vortex generation, the subwavelength slit is arranged along the Archimedean spiral line ρ1'(θ) = ρ0 + m·λ. SPP0 / 2π·θ arrangement, and set the slit rotation angle to α1(θ)=α0+n0·θ. Figure 3 As shown, ρ0 represents the initial radius in the polar coordinate system (ρs,θ), which is set to 2000μm; m is an integer 1, and λ SPP0 is the surface plasmon wavelength corresponding to the frequency f0, m·λ SPP0 is the radius increase of the slit array after one rotation in the shape of an Archimedean spiral; α0 and n0 represent the initial angle and rotation factor of the slits, respectively. In this embodiment, α0 = π / 4, n0 = 1, and f0 = 0.3 THz.

[0046] Under the above-mentioned metasurface design, the metasurface is illuminated from the back with a light field of the frequency and polarization to be measured, and then the two-dimensional electric field distribution of the detection surface is collected line by line using a terahertz near-field scanning system to obtain different vortex light interference patterns under different polarizations and wavelengths. Figure 4 (b) shows the vortex field distribution formed by the metasurface under six polarization states and three frequencies; (a) shows the OAM decomposition of this field distribution according to The polarization parameters (S1-S3) obtained in simulation and experiment are in good agreement with the theoretical results.

[0047] In addition to theoretically solving the light field information by performing OAM decomposition on the light field, a residual neural network can also be introduced to directly predict the input light field information based on the distribution of the vortex light interference field generated by the metasurface. The neural network architecture used in the present invention is as follows: Figure 5 As shown in the figure, the classic framework of residual neural network is adopted, including convolutional layers, pooling layers, activation functions, residual blocks, etc. The network input is the field intensity distribution of the light field to be measured irradiated on the metasurface device, and the output is the predicted polarization and frequency parameters. During the network training process, the mean square error is used to define the loss function of the multi-target regression task:

[0048]

[0049] After the neural network is trained, the test set is used to test the neural network. The network input is the plasmon field distribution generated by the metasurface, and the network output is the predicted polarization parameters and frequency. Figure 6 and Figure 7 The image compares the network's predicted light field information with the actual light field information. The image demonstrates that the neural network combined with the metasurface can accurately predict the frequency and polarization of the incident light. The frequency prediction accuracy is at least 25 GHz within the 0.3-1.1 THz frequency range, and the Stokes parameters S1-S3, which represent polarization, are accurately predicted, demonstrating the reliability of this method. Figure 7 (a) and (b) are the polarization information test results of simulation and experiment, respectively. Different frequencies and polarization states were selected for testing. The color of the circle represents the incident frequency, the hollow circle represents the actual polarization, and the solid circle represents the polarization parameter predicted by the neural network.

[0050] The present invention combines neural networks for the first time to realize high-dimensional light field measurement in the terahertz band. The high-dimensional light field measurement technology proposed in the present invention realizes the synchronous analysis of multi-dimensional parameters of the light field through the collaborative design of metasurfaces and neural networks. It can simultaneously extract multi-dimensional information such as amplitude, phase, polarization, wavelength, etc. of the light field to construct a complete light field representation, breaking through the limitation that traditional detectors can only respond to intensity or low-dimensional light field information. At the same time, the metasurface used realizes precise control of the light field at the micron scale through a monolithic integrated sub-wavelength structure, and the invented method can be easily migrated to any wavelength in the visible light and near-infrared bands. This invention has opened up a new path for ultra-compact, high-dimensional information detection and imaging detection, and has huge application value in the fields of optical communications, remote sensing, industrial detection, etc.

[0051] Finally, it should be noted that the purpose of disclosing the embodiments is to facilitate a further understanding of the present invention. Those skilled in the art should understand that various substitutions and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the contents disclosed in the embodiments; the scope of protection claimed by the present invention shall be determined by the scope defined in the claims.

Claims

1. A metasurface device for high-dimensional light field detection, comprising a substrate and a metal film thereon, characterized in that: An array structure consisting of sub-wavelength slits is etched on a metal film, wherein the sub-wavelength slits in the array structure are arranged along an Archimedean spiral line ρ1'(θ)=ρ0+m·λ. SPP0 / 2π·θ arrangement, and set the slit rotation angle to α1(θ)=α0+n0·θ; Where ρ0 represents the initial radius in the polar coordinate system (ρs,θ), m is an integer, and λ SPP0 is the surface plasmon wavelength corresponding to a certain frequency f0 in the detection band, θ is the angle between the line connecting the center of the slit and the center of the array and the x-axis, α0 and n0 represent the initial angle and rotation factor of the slit, respectively.

2. The metasurface device according to claim 1, wherein The width of the subwavelength slit is b, the distance between the centers of adjacent slits is d, and the depth is the thickness of the metal film; α1(θ) is the angle between the long side of the slit and the positive direction of the y-axis; α0 is selected as a suitable angle in the range of [0, arccos(b / d)], and n0 is a positive integer. The values ​​of α0 and n0 ensure that the slits in the array structure do not overlap.

3. The metasurface device according to claim 2, wherein: Set α0 = π / 4, n0 = 1, m = 1; f0 is 0.3 THz, λ SPP0 =1mm。 4. The metasurface device according to claim 1, wherein The substrate is a transparent substrate or a non-transparent substrate; the metal film is selected from a metal material with low surface plasmon loss; and the metal film is covered or not covered with a dielectric layer.

5. The metasurface device according to claim 4, wherein: The substrate material is selected from SiO2, CaF2, Al2O3, and Si; the metal film is selected from gold, silver, and aluminum, and has a thickness of 150 to 300 nm.

6. The metasurface device according to claim 4, wherein: The dielectric layer covering the metal film is a polyimide layer with a thickness of 10 to 20 μm.

7. A method for detecting high-dimensional light field information using the metasurface device according to any one of claims 1 to 6, comprising the following steps: 1) Irradiating the metasurface device with the light field to be measured to collect the surface plasmon field distribution E on the metasurface z , using formula (4) to calculate E z Perform OAM decomposition to obtain the complex coefficients M of each order OAM component j ; In formula (4), polar coordinates represents the position of the observation point, where ρ is the radius of the observation circle from the center of the device array, is the angle between each point on the observation circle and the line connecting the array center and the x-axis; e represents a natural constant; j is the topological charge of the vortex light mode, and i is an imaginary unit; 2) Determine the incident frequency f = -f0l / m of the light field to be measured based on the topological charges corresponding to the two highest-intensity vortex light components in the OAM spectrum, i.e., the l-1 and l+1 orders; 3) According to formula (5), we can get ae iξ , and then determine the polarization state of the light field to be measured Among them, M l+1 / M l-1 is the ratio of the coefficients of the two main vortex optical modes, J() is the Bessel function, ρ is the radius of the observation circle from the center of the device array, i is the imaginary unit, k SPP is the wave vector of the surface plasmon along the propagation direction, is the phase difference between the y- and x-direction components of the incident polarization field, a = E 0y / E 0x is the amplitude ratio of the y- and x-direction components of the incident polarization field.

8. A method for detecting high-dimensional light field information using the metasurface device according to any one of claims 1 to 6, characterized in that: The input light field information is predicted directly based on the distribution of the vortex light interference field generated by the metasurface by training a neural network. The training process of the neural network includes the following steps: 1) illuminating the metasurface device with light fields of different polarizations and wavelengths, and obtaining surface plasmon field distributions through numerical simulation and experiments as a data set for training a neural network model; 2) The dataset is divided into a training set and a test set. The surface plasmon field distribution in the training set is used to train the neural network. The mean square error is used to define the loss function of the multi-output regression task: Among them, S i real is the label of the polarization Stokes parameter corresponding to the field distribution, S i pre is the polarization parameter predicted by the neural network, freq real is the label of the frequency corresponding to the field distribution, freq pre is the frequency predicted by the neural network, α1 and α2 represent adjustable weight coefficients; S1-S3 are the three components of the Stokes parameters used to completely describe the polarization of light: S1 represents the difference between the horizontal polarization intensity and the vertical polarization intensity, S2 represents the difference between the +45° polarization intensity and the -45° polarization intensity, and S3 represents the difference between the right-handed circularly polarized light intensity and the left-handed circularly polarized light intensity; 3) Use the test set to verify the trained neural network. The network input is the plasmon field distribution generated by the metasurface device, and the network output is the predicted polarization parameters and frequency.

9. The method according to claim 8, wherein The neural network is a residual neural network.

10. The method according to claim 7 or 8, characterized in that The light field is in the terahertz band.

Images