A diffraction metasurface full stokes polarimetric spectral detection system and method
By combining diffractive metasurfaces and deep learning algorithms, high-precision recovery of all Stokes parameters under a single exposure was achieved, solving the problems of system complexity and accuracy in traditional polarization spectral acquisition technology, and realizing system miniaturization and real-time monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI INSTITUTE OF TECHNICAL PHYSICS CHINESE ACADEMY OF SCIENCES
- Filing Date
- 2026-04-17
- Publication Date
- 2026-06-23
Smart Images
Figure CN122259033A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of micro-nano optics and polarization spectroscopy detection technology, specifically relating to a detection system and method that uses a diffractive optical metasurface to spatially encode the polarization and spectral information of incident light, and combines deep learning algorithms to recover the full Stokes polarization spectral information. Background Technology
[0002] Polarization and spectroscopy are two core physical dimensions of light waves. Polarization information reflects the surface roughness, morphology, and material properties of an object, while spectral information provides the "fingerprint" of an object's chemical composition and molecular structure. Combining these two to acquire full-Stokes polarization spectral data has extremely important scientific research and engineering application value in fields such as material identification, remote sensing, biochemical analysis, and astronomical observation.
[0003] However, traditional polarization spectral acquisition technology faces a trade-off between size, accuracy, and timeliness: 1. Complex and bulky system: Traditional systems typically require cascading beam-splitting elements (such as gratings and prisms) with polarization elements (such as rotating polarizers and phase retarders). This results in long optical links, complex structures, and makes system integration and miniaturization difficult.
[0004] 2. Multi-time measurement error: The time-division acquisition method (acquiring multiple sets of data through rotating optical elements) has poor ability to capture dynamic targets and is prone to introducing mechanical vibration deviations, making it impossible to achieve real-time monitoring of transient processes.
[0005] 3. Difficulty in coupling polarization and spectral dimensions: In a wide wavelength range, traditional optical elements often exhibit severe dispersion, which causes the modulation efficiency of the polarization state to change drastically with wavelength, greatly increasing the difficulty of accurately decoupling from complex optical signals and recovering the full Stokes spectrum curve.
[0006] 4. Limited data processing accuracy: Existing simplified models often suffer from low reconstruction accuracy and noise sensitivity when processing polarization spectral coupled data, making it difficult to meet the requirements of high-resolution detection.
[0007] In recent years, the rise of metasurface technology has provided a new approach for miniaturized optical sensing. However, existing metasurface sensors mostly focus on spatial imaging, and there is still a lack of an efficient, compact and highly robust system solution for extremely high-precision recovery of incident light polarization spectral information (such as achieving a reconstruction error of less than 1%) and deep learning intelligent extraction of all Stokes parameters.
[0008] Therefore, designing a system that can achieve single-exposure acquisition and, through the collaboration of diffraction coding and deep learning algorithms, accurately recover the full Stokes polarization spectrum information of incident light from complex diffraction fields has become a key technical problem that urgently needs to be solved in the field of micro-nano optics and information detection. Summary of the Invention
[0009] The purpose of this invention is to provide: A diffractive metasurface full Stokes polarization spectroscopy detection system and method are proposed to solve the problem of difficulty in simultaneously achieving full Stokes parameter acquisition, wide-band response, and extremely high reconstruction accuracy in a single exposure of a light spot.
[0010] Terminology Explanation: Unless otherwise defined, all technical terms used herein have the same meanings as commonly understood by one of ordinary skill in the art to which this subject matter pertains. Unless otherwise stated, all patents, patent inventions, and disclosures cited throughout this document are incorporated herein by reference in their entirety. Where multiple definitions exist for terms herein, the definitions provided in this chapter shall prevail.
[0011] It should be understood that the above brief description and the following detailed description are exemplary and for illustrative purposes only, and do not limit the subject matter of the invention in any way. In this invention, the singular is used in conjunction with the plural unless otherwise specifically stated. It should also be noted that, unless otherwise stated, the use of “or” or “or” means “and / or”. Furthermore, the use of the term “comprising” and other forms such as “including,” “containing,” and “contains” are not limiting.
[0012] Unless specifically defined herein, the use of various commercially available products herein employs standard techniques. For example, it may be carried out using the manufacturer's instructions for use, or in accordance with methods known in the art or the description of this invention. The techniques and methods described herein can generally be implemented according to conventional methods well known in the art, based on the descriptions in the various general and more specific documents cited and discussed in this specification.
[0013] The terms “optional / arbitrary” or “optionally / arbitrarily” mean that the event or situation described below may or may not occur, including both the occurrence and non-occurrence of the event or situation.
[0014] The term "metasurface" as used in this article refers to an ultrathin artificial interface composed of a two-dimensional arrangement of subwavelength-scale artificial structural units (superatoms). Its core value lies in the arbitrary and precise control of the amplitude, phase, polarization, and propagation direction of electromagnetic waves (light / microwave / terahertz) within an extremely thin thickness.
[0015] The term “polarization” as used in this article refers to the direction of the electric field vibration of light, which is the fourth dimension of light.
[0016] The term "Poincaré sphere" as used in this article refers to a three-dimensional geometric visualization tool specifically used to map the polarization state of light.
[0017] In a first aspect, the present invention provides: A diffractive metasurface all-Stokes polarization spectroscopy detection system includes: An optical coding module includes a diffractive metasurface containing nanopillars arranged according to a preset rule, which is used to perform polarization multiplexing phase modulation and spectral dispersion coding on incident light, converting incident light information of different polarization states and wavelengths into spatial diffraction distributions. The preset rule is as follows: six polarization states are selected, namely linear polarization at 0°, 45°, 90°, and 135°, left-handed circular polarization, and right-handed circular polarization; the target phase distribution of the six polarization states is as follows:
[0018] in In coordinates The amount of phase delay that needs to be introduced at that point. Let the polar angle of the exit direction be the polar angle corresponding to each incident polarization state. The azimuth angle of the exit direction corresponding to each incident polarization state; 0° linear polarization, the phase gradient direction corresponding to the stripe is in-plane along the 60° direction; 45° linear polarization, the phase gradient direction corresponding to the stripes is in-plane along the 300° direction; 90° linear polarization, the phase gradient direction corresponding to the stripes is in-plane along the 120° direction; 135° linear polarization, the phase gradient direction corresponding to the stripes is in-plane along the 240° direction; Left-handed circular polarization, the phase gradient direction corresponding to the fringes is along the 180° direction in the plane; Right-handed circular polarization, the phase gradient direction corresponding to the fringes is along the 0° direction in the plane; The specific geometric parameters of the nanopillars are determined based on the target phase distribution; by using a preset arrangement rule of nanopillars to match the target phase distribution, incident light of a specific polarization state is induced to deflect in a preset diffraction direction. A photoelectric conversion module is located on the diffraction path of the diffraction metasurface and is used to record the diffraction distribution in a single exposure to generate a two-dimensional original intensity image. An information recovery module, connected to the photoelectric conversion module, has a pre-trained deep learning model built in, used to decouple and recover the full Stokes polarization spectral information of the incident light from the two-dimensional original intensity image.
[0019] Based on further solutions to the technical problems of the present invention, or simultaneous solutions to multiple technical problems, the preferred solution in the technical solution provided in the first aspect of the present invention includes: First preferred option: The diffractive metasurface also includes a substrate, and the nanopillars are integrated on the surface of the substrate; the substrate has a thickness of 500 μm and the nanopillars have a height of 1 μm; the nanopillars are rectangular in shape.
[0020] Second preferred option: The height of the nanopillars is 1 μm, and the center-to-center distance between adjacent nanopillars is 560 nm; the length and width of the nanopillars are both 138 nm to 426 nm.
[0021] Third preferred option: The diffractive metasurface full Stokes polarization spectroscopy detection system also includes six polarization diffraction channels, each corresponding to one of the six feature points on the Poincaré sphere. The diffraction directions of the six polarization diffraction channels are as follows: 0-degree linear polarization: (α,β) = (7.62°, 60°); 90° linear polarization: (α,β) = (7.62°,120°); 135° linear polarization: (α,β) = (7.62°, 240°); 45° linear polarization: (α,β) = (7.62°, 300°); Left-handed circular polarization: (α,β) = (8.12°, 180°); Right-hand circular polarization: (α,β) = (8.12°, 0°); in, Polar angle, It is the azimuth angle.
[0022] Fourth preferred option: The deep learning model is a polarization spectral reconstruction neural network. The process by which the polarization spectral reconstruction neural network processes image data is as follows: Downsampling path: Through a series of convolutional and pooling layers, the spatial size of the feature map is gradually reduced, and the number of channels is continuously increased to 512; Skip connection: The input feature map is copied and concatenated with the output of a certain stage in the downsampling path and the output after passing through the downsampling path to obtain a concatenated feature map, which is then output to the upsampling path; Upsampling path: The spliced feature map output by skip connections is upsampled, and the spatial size of the spliced feature map is gradually increased, while the number of channels is first reduced and then increased.
[0023] Fifth preferred option: The input data for the polarization spectral reconstruction neural network is a 512×512×1 original diffraction grayscale image containing the intensity distribution of 6 diffraction channels.
[0024] Sixth preferred option: The output data of the polarization spectral reconstruction neural network is an N×4 multidimensional data matrix, where N is the number of spectral samples; the number 4 corresponds to the total Stokes parameters for each sample point. S 0, S 1, S 2, S 3}.
[0025] Seventh preferred option: The operating wavelength range of the diffractive metasurface full Stokes polarization spectroscopy detection system is 1150 nm to 1650 nm.
[0026] Secondly, this invention provides: a method for detecting a diffractive metasurface using full Stokes polarization spectroscopy, comprising: S1, Optical Field Modulation: The incident light field generates spatially encoded diffraction through the metasurface; through the spatial arrangement of the metasurface, the spectrum and total Stokes parameters of the incident light are mapped to the spatial coordinates and intensity relationship on the detector plane; S2, Image Acquisition: The encoded original grayscale image is acquired through a single exposure of the detector; S3. Information Reconstruction: Input the original grayscale image into the trained neural network model; S4. Output Results: The neural network model outputs full Stokes polarization spectral data.
[0027] The present invention has at least the following beneficial effects: This invention utilizes the diffraction properties of metasurfaces to map complex polarization-spectral multidimensional information to two-dimensional pixel intensities, achieving spatial multiplexing at the physical level. Deep learning algorithms (such as residual networks) effectively handle the effects of nonlinear dispersion and polarization crosstalk of metasurfaces over a wide wavelength range, significantly reducing the spectral reconstruction error (RMSE). Attached Figure Description
[0028] Figure 1 This is a schematic diagram of the preset arrangement of the metasurface in this invention; Figure 2 This is the target phase distribution diagram of the metasurface nanopillars in this invention; Figure 3 This is a schematic diagram of the directional diffraction direction of the metasurface; Figure 4 This is a schematic diagram of metasurface diffraction polarization encoding. Figure 5 A schematic diagram of metasurface diffraction wavelength encoding; Figure 6 Architecture diagram of a neural network for polarization spectrum reconstruction; Figure 7 A schematic diagram illustrating the changes in data input and output of a neural network for polarization spectrum reconstruction. Figure 8 This is a schematic diagram of a metasurface polarization-spectroscopy testing system; Figure 9 Illustration of spectral reconstruction effect Figure 1 ; Figure 10 Illustration of spectral reconstruction effect Figure 2 ; Figure 11 This is a schematic diagram of the polarization reconstruction effect.
[0029] Explanation of reference numerals in the attached figures: 1. Diffractive metasurface; 2. Near-infrared camera; 3. Beam splitter; 4. Quarter wave plate; 5. Half wave plate; 6. Linear polarizer; 7. Acousto-optic tunable filter; 8. Supercontinuum light source; 9. Narrowband bandpass filter; 10. Polarimeter; 11. Spectrometer. Detailed Implementation
[0030] The following non-limiting embodiments are intended to enable those skilled in the art to gain a more comprehensive understanding of the present invention, but do not limit the invention in any way. The following content is merely an exemplary description of the scope of protection claimed by the present invention, and those skilled in the art can make various changes and modifications to the present invention based on the disclosed content, and such changes should also fall within the scope of protection claimed by the present invention.
[0031] The present invention will be further described below by way of specific embodiments. Unless otherwise specified, all instruments, devices, equipment and other equipment used in the embodiments of the present invention are obtained through conventional commercial means.
[0032] Example 1 This embodiment provides a diffractive metasurface full Stokes polarization spectroscopy detection system for single-exposure reconstruction of light spots; its operating wavelength covers the near-infrared band, preferably from 1150 nm to 1650 nm, and specifically the system includes: A. Optical encoding module The optical coding module includes a diffractive metasurface, which is composed of a nanopillar array arranged in a preset pattern. It is used to perform polarization multiplexing phase modulation and spectral dispersion coding on incident light, and to convert incident light information with different polarization states and wavelengths into spatial diffraction distribution. The arrangement of the nanopillar array is as follows: Figure 1 As shown, Figure 1Starting from the left, columns 1, 4, 7, etc., each have two channels, with a fixed rotation angle of 45°, but the length and width of the nanopillars vary; columns 2, 5, 8, etc., each have two channels, and the nanopillars in the same column have the same rotation angle, but the rotation angles between different columns are inconsistent, and the length and width also vary; columns 3, 6, 9, etc., each have two channels, and the rotation angle of the nanopillars in each column is 0°, but the length and width vary.
[0033] The overall layout is based on Figure 2 The target phase distribution under different polarization states shown provides a basis for setting the parameters of the nanopillar array. Figure 2 In this context, the intensity of a color represents its phase value, with higher brightness indicating a higher phase. 0° linear polarization, the phase gradient direction corresponding to the stripe is in-plane along the 60° direction; 45° linear polarization, the phase gradient direction corresponding to the stripes is along 300° in-plane; 90° linear polarization, the phase gradient direction corresponding to the stripes is along 120° in-plane; 135° linear polarization, the phase gradient direction corresponding to the stripes is along 240° in-plane; Left-handed circular polarization, the phase gradient direction corresponding to the fringes is along 180° in-plane; Right-hand circular polarization, the phase gradient direction corresponding to the fringes is along 0° in-plane; Left-handed and right-handed circularly polarized light are vertical fringes perpendicular to x. This corresponds to the geometric phase modulation (Pancharatnam-Berry phase) of circularly polarized light. By rotating the meta-atoms, opposite or independent phase gradients are applied to left-handed and right-handed circularly polarized light, respectively.
[0034] The parameters of the nanopillar array can be deduced from the phase distribution. The specific steps are as follows: 1. Extract the local phase gradient from the target phase distribution (which determines the beam deflection / wavefront shape); 2. Select the corresponding phase modulation mechanism (geometric phase / propagation phase / hybrid mechanism) according to the polarization type (linear polarization / circular polarization); 3. From the required phase response, deduce the geometric parameters (length, width, rotation angle, height, etc.) of the nanopillar.
[0035] The diffractive metasurface, serving as the core encoding unit of the system, is composed of a micro / nano structure array made of a high-refractive-index material (such as single-crystal silicon) integrated onto the surface of a low-refractive-index transparent substrate (such as sapphire or quartz). The substrate thickness is 500 μm, and the height of the micro / nano array is 1 μm. The micro / nano structure can be a nanopillar, with a rectangular shape. The micro / nano structure is fabricated using electron beam lithography and plasma etching processes, and its geometric parameters are spatially arranged according to a preset phase distribution. These geometric parameters include a height of 1 μm, a rectangular cross-sectional shape, a center-to-center spacing of 560 nm, and the length and width of the rectangles ranging from 138 nm to 426 nm. The rotation angles are distributed according to a preset rule.
[0036] The physical encoding principle is as follows: a coupling mechanism is introduced between the geometric phase (PB phase) and the propagation phase. The geometric phase is controlled by the rotation angle of the aforementioned rectangular nanopillars; the propagation phase is controlled by the length and width dimensions of the aforementioned rectangular nanopillars. Therefore, when the rectangular nanopillars are arranged, different length and width dimensions and rotation angles achieve the control and coupling of the geometric phase and the propagation phase. This metasurface can achieve spatial multiplexing phase modulation of the incident light field. This modulation effect simultaneously converts the polarization state and spectral information of the incident light into a diffraction distribution with a specific spatial frequency.
[0037] The metasurface employs a spatial encoding mechanism based on polarization and spectral density for physical encoding. In this embodiment, the spatial arrangement of the metasurface maps the spectrum and total Stokes parameters of the incident light into spatial coordinates and intensity relationships on the detector plane. Specifically, the spatial encoding mechanism based on polarization and spectral density includes polarization channel encoding and spectral dispersion encoding.
[0038] Polarization channel encoding: The system is designed with 6 specific polarization diffraction channels, each corresponding to one of the 6 feature points on the Poincaré sphere. The incident light is directionally diffracted to a polar angle. and azimuth Defined specific spatial orientation, such as Figure 3 and Figure 4 As shown, the specific diffraction direction is designed as follows: 0-degree linear polarization: (α,β) = (7.62°, 60°); 90° linear polarization: (α,β) = (7.62°,120°); 135° linear polarization: (α,β) = (7.62°, 240°); 45° linear polarization: (α,β) = (7.62°, 300°); Left-handed circular polarization: (α,β) = (8.12°, 180°); Right-hand circular polarization: (α,β) = (8.12°, 0°).
[0039] The incident light wave vector of the diffractive metasurface is ,in The emitted light wave vector is The phase gradient on the plane corresponds to the change in the tangential wave vector:
[0040] Integrating the phase gradient yields the phase distribution at each location on the diffractive metasurface as follows:
[0041] in In coordinates The amount of phase delay that needs to be introduced at that point. Let the polar angle of the exit direction be the polar angle corresponding to each incident polarization state. The azimuth angle of the outgoing direction corresponding to each incident polarization state.
[0042] Spectral dispersion coding: Based on the structural dispersion characteristics of silicon nanopillar arrays, the diffraction angles within each polarization channel exhibit wavelength dependence. For example... Figure 5 As shown, the deflection angle of the long-wavelength beam is relatively large, and its imaging position on the detector focal plane is far from the center; while the short-wavelength beam is closer to the center. This radial dispersion effect enables the unfolding of spectral information in the spatial dimension.
[0043] B. Photoelectric conversion module The photoelectric conversion module is located on the diffraction light path of the metasurface and is used to record the diffraction distribution in a single exposure to generate a two-dimensional original intensity image; The photoelectric conversion module is used for photoelectric acquisition and can employ an area array detector (such as an InGaAs focal plane array) positioned behind the diffraction metasurface in the diffraction light path. The area array detector is used to acquire a two-dimensional diffraction intensity-encoded image modulated by the diffraction metasurface in a single exposure, which serves as the raw input to the deep learning model.
[0044] C. Information Recovery Module The information recovery module is connected to the photoelectric conversion module and has a built-in pre-trained deep learning model for decoupling and recovering the full Stokes polarization spectrum information of the incident light from the original two-dimensional intensity image.
[0045] The information recovery module has a built-in polarization spectral reconstruction neural network, whose input-output characteristics are as follows: Figure 7As shown, the input is a 512×512×1 original diffraction grayscale image, containing the intensity distribution of 6 diffraction channels (a single frame of 512×512 pixel original grayscale pattern). The output is an N×4 multidimensional data matrix, where N is the number of spectral samples, for example, N=501. N corresponds to spectral sampling points with a step size of 1nm in the wavelength range of 1150nm to 1650nm; 4 corresponds to the total Stokes parameters of each sampling point. S 0, S 1, S 2, S 3}.
[0046] The architecture diagram of the polarization spectrum reconstruction neural network is as follows: Figure 6 As shown, its image processing procedure includes: Input layer: The input data is image data, i.e., the original diffraction grayscale image; Four residual blocks: Four residual blocks are concatenated, each with the same structure. Each residual block uses the previous level for skip connections. After feature extraction from the residual blocks, the spatial size of the feature map gradually decreases, while the number of channels continuously increases. Figure 6 The channel count shown in the diagram is 64, 128, 256, and 512. By using skip connections, the vanishing gradient problem in deep networks is solved, while also accurately recovering detailed information and avoiding the loss of small target features.
[0047] Pooling and Fully Connected Classification Region: The 512-dimensional high-dimensional features output by the residual block are globally downsampled into a one-dimensional feature vector, which is then connected to the fully connected layer. The first fully connected layer has 512 neurons and performs the first global mapping of the features; the second fully connected layer has 2004 neurons and further performs high-dimensional feature mapping, ultimately outputting a 501×4 multi-dimensional data matrix.
[0048] The polarization spectral reconstruction neural network implements the above image processing based on residual blocks. The internal features of the polarization spectral reconstruction neural network include: Batch Normalization (BN): A BN layer is added after each convolutional layer to normalize the mean and variance of the intermediate feature maps. This effectively solves the internal covariate shift problem in deep training, enabling the network to maintain consistent sensitivity to spectral signals of different intensity levels, accelerating convergence and improving generalization ability.
[0049] Residual Blocks: By using a skip connection structure, the original light intensity distribution information is not lost when extracting high-order semantic features, thereby ensuring the smoothness of the reconstructed spectral curve.
[0050] Training and representation methods for polarization spectral reconstruction neural networks: To ensure reconstruction accuracy, the system needs to undergo the following training and calibration process using a testing system, such as... Figure 8 As shown, it includes: 1) Dataset building unit: It is used to realize the construction of high-throughput training datasets. It mainly consists of a supercontinuum light source 8 and an acousto-optic tunable filter (AOTF) 7, which can realize multi-channel wavelength switching (supports up to 8 wavelength channels for synchronous output in a single operation). 2) Polarization modulation unit: It consists of a wide-band achromatic linear polarizer 6, a half-wave plate 5 and a quarter-wave plate 4, and uses a piezoelectric rotary table to control the polarization state coverage of the entire surface of the Poincaré sphere.
[0051] Incident light illuminates the diffractive metasurface 1, and the detector pattern is recorded using a near-infrared camera 2. Simultaneously, a beam splitter 3 extracts another incident light beam, which is then split again and passed through a narrowband bandpass filter 9. The polarization spectrum of the incident light is recorded as the true value using a commercially available polarimeter 10 and spectrometer 11. The network parameters are iteratively optimized using an adaptive moment estimation method by minimizing the mean square error (MSE) between the predicted polarization spectrum and the true value.
[0052] like Figure 9 As shown, using a fixed polarization method, eight spectra with different peak wavelengths are output simultaneously, and detection and reconstruction are completed. The true and predicted wavelength values match perfectly. Figure 10 As shown, using fixed polarization conditions, eight single-peak spectra with different peak wavelengths were output sequentially. Each spectrum was tested and reconstructed in turn, and the true and predicted wavelength values matched perfectly. Figure 11 As shown, using a fixed wavelength, 30 different polarization states were generated sequentially, and each was detected and reconstructed. The true and predicted values of the Stokes parameters matched perfectly. Figures 9 to 11 It can be demonstrated that the present invention can significantly reduce spectral reconstruction errors.
[0053] Example 2 This embodiment provides a method for detecting full Stokes polarization spectroscopy on a diffractive metasurface using the aforementioned full Stokes polarization spectroscopy detection system, specifically including: S1, Optical Field Modulation: The incident light field generates spatially encoded diffraction through the metasurface; through the spatial arrangement of the metasurface, the spectrum and total Stokes parameters of the incident light are mapped to the spatial coordinates and intensity relationship on the detector plane.
[0054] S2, Image Acquisition: The encoded original grayscale image is acquired through a single exposure of the detector; S3. Information Reconstruction: Input the original grayscale image into the trained neural network model; S4. Output Results: The neural network model outputs full Stokes polarization spectral data.
[0055] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention do not depart from the essence and scope of the technical solution of the present invention.
Claims
1. A full Stokes polarization spectroscopy detection system for diffractive metasurfaces, characterized in that, include: An optical coding module includes a diffractive metasurface containing nanopillars arranged according to a preset rule, which is used to perform polarization multiplexing phase modulation and spectral dispersion coding on incident light, converting incident light information of different polarization states and wavelengths into spatial diffraction distributions. The preset rule is as follows: six polarization states are selected, namely linear polarization at 0°, 45°, 90°, and 135°, left-handed circular polarization, and right-handed circular polarization; the target phase distribution of the six polarization states is as follows: in In coordinates The amount of phase delay that needs to be introduced at that point. Let the polar angle of the exit direction be the polar angle corresponding to each incident polarization state. The azimuth angle of the exit direction corresponding to each incident polarization state; The specific geometric parameters of the nanopillars are determined based on the target phase distribution; by using a preset arrangement rule of nanopillars to match the target phase distribution, incident light of a specific polarization state is induced to deflect in a preset diffraction direction. A photoelectric conversion module is located on the diffraction path of the diffraction metasurface and is used to record the diffraction distribution in a single exposure to generate a two-dimensional original intensity image. An information recovery module, connected to the photoelectric conversion module, has a pre-trained deep learning model built in, used to decouple and recover the full Stokes polarization spectral information of the incident light from the two-dimensional original intensity image.
2. The diffractive metasurface full Stokes polarization spectroscopy detection system according to claim 1, characterized in that, The diffractive metasurface also includes a substrate, and the nanopillars are integrated on the surface of the substrate; the substrate has a thickness of 500 μm and the nanopillars have a height of 1 μm; the nanopillars are rectangular in shape.
3. The diffractive metasurface full Stokes polarization spectroscopy detection system according to claim 2, characterized in that, The height of the nanopillars is 1 μm, and the center-to-center distance between adjacent nanopillars is 560 nm; the length and width of the nanopillars are both 138 nm to 426 nm.
4. The diffractive metasurface full Stokes polarization spectroscopy detection system according to claim 1, characterized in that, The diffractive metasurface full Stokes polarization spectroscopy detection system also includes six polarization diffraction channels, each corresponding to one of the six feature points on the Poincaré sphere. The diffraction directions of the six polarization diffraction channels are as follows: 0-degree linear polarization: (α,β) = (7.62°, 60°); 90° linear polarization: (α,β) = (7.62°,120°); 135° linear polarization: (α,β) = (7.62°, 240°); 45° linear polarization: (α,β) = (7.62°, 300°); Left-handed circular polarization: (α,β) = (8.12°, 180°); Right-hand circular polarization: (α,β) = (8.12°, 0°); in, Polar angle, It is the azimuth angle.
5. The diffractive metasurface full Stokes polarization spectroscopy detection system according to claim 1, characterized in that, The deep learning model is a polarization spectral reconstruction neural network. The process by which the polarization spectral reconstruction neural network processes image data is as follows: Downsampling path: Through a series of convolutional and pooling layers, the spatial size of the feature map is gradually reduced, and the number of channels is continuously increased to 512; Skip connection: The input feature map is copied and concatenated with the output of a certain stage in the downsampling path and the output after passing through the downsampling path to obtain a concatenated feature map, which is then output to the upsampling path; Upsampling path: The spliced feature map output by skip connections is upsampled, and the spatial size of the spliced feature map is gradually increased, while the number of channels is first reduced and then increased.
6. The diffractive metasurface full Stokes polarization spectroscopy detection system according to claim 5, characterized in that, The input data for the polarization spectral reconstruction neural network is a 512×512×1 original diffraction grayscale image containing the intensity distribution of 6 diffraction channels.
7. The diffractive metasurface all-Stokes polarization spectroscopy detection system according to claim 6, characterized in that, The output data of the polarization spectral reconstruction neural network is an N×4 multidimensional data matrix, where N is the number of spectral samples; the number 4 corresponds to the total Stokes parameters for each sample point. S 0, S 1, S 2, S 3}.
8. The diffractive metasurface all-Stokes polarization spectroscopy detection system according to claim 1, characterized in that, The operating wavelength range of the diffractive metasurface full Stokes polarization spectroscopy detection system is 1150 nm to 1650 nm.
9. A method for detecting diffractive metasurfaces using full Stokes polarization spectroscopy, comprising: S1, Optical Field Modulation: The incident light field generates spatially encoded diffraction through the metasurface; through the spatial arrangement of the metasurface, the spectrum and total Stokes parameters of the incident light are mapped to the spatial coordinates and intensity relationship on the detector plane; S2, Image Acquisition: The encoded original grayscale image is acquired through a single exposure of the detector; S3. Information Reconstruction: Input the original grayscale image into the trained neural network model; S4. Output Results: The neural network model outputs full Stokes polarization spectral data.