A polarization-assisted reconfigurable metasurface visual secret sharing method and system based on holographic neural network
By mapping the visual shared key to orthogonal polarization channels and generating a phase hologram through a holographic neural network, and loading it onto a dual-polarization reconfigurable metasurface, the problems of high complexity and insufficient security of existing visual secret sharing technologies are solved, realizing low-complexity, high-imaging-quality, and dynamically switching visual secret sharing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHEJIANG UNIV CITY COLLEGE
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-21
AI Technical Summary
Existing visual secret sharing technologies rely on static carriers, their security is limited to the digital domain, they are difficult to integrate with electromagnetic control devices, have a large number of control units, high system complexity, high power consumption and cost, and are difficult to achieve dynamic updates and high imaging quality.
A holographic neural network is used to map the visual shared key to an orthogonal polarization channel. A phase hologram is generated using a deep holographic prior neural network and loaded onto a dual-polarization reconfigurable metasurface. The original secret image is then decrypted and recovered using orthogonal polarized electromagnetic waves.
It significantly reduces the number of control units and system complexity, lowers power consumption and cost, while improving image reconstruction quality and security, supports dynamic switching of multiple secret information, and is suitable for fields such as electromagnetic stealth and information camouflage.
Smart Images
Figure CN122437647A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electromagnetic wave manipulation and information security technology, and more specifically to a polarization-assisted reconfigurable metasurface visual secret sharing method and system based on holographic neural networks. Background Technology
[0002] Currently, visual secret sharing is an encryption technology that can recover information through visual overlay without complex calculations. However, existing technologies have the following main shortcomings: (1) It relies on a static carrier and cannot dynamically update secret information; (2) Security is limited to the digital domain and lacks physical layer protection; (3) It is difficult to integrate with electromagnetic control devices; Metasurface technology enables subwavelength-scale manipulation of electromagnetic waves, providing a new means for physical layer encryption. In particular, dual-polarization metasurfaces can achieve parallel encoding of information. However, existing technologies still suffer from the following drawbacks: a large number of control units, resulting in high system complexity; unit coupling affecting performance; numerous active devices, leading to high power consumption and cost; and traditional holographic algorithms struggling to balance accuracy and complexity. Therefore, how to provide a visual secret sharing method that is low in complexity, high in imaging quality, supports dynamic switching, and is highly secure is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0003] In view of this, the present invention provides a method and system for sharing visual secrets of polarization-assisted reconfigurable metasurfaces based on holographic neural networks, in order to solve the problems existing in the background art.
[0004] To achieve the above objectives, the present invention adopts the following technical solution: A method for sharing visual secrets of polarization-assisted reconfigurable metasurfaces based on holographic neural networks includes: The original secret image is visually secret-shared encoded to obtain multiple visual sharing keys; The visual shared key is mapped to orthogonal polarization channels to form a polarization shared key; A deep holographic prior neural network is used to convert a polarization shared key into a phase hologram. The phase hologram is loaded onto a dual-polarization reconfigurable metasurface; By irradiating a dual-polarization reconstructable metasurface with orthogonally polarized electromagnetic waves, the polarization shared key is reconstructed and superimposed for decryption, thus restoring the original secret image.
[0005] Optionally, obtaining multiple visual shared keys specifically involves: According to the visual secret sharing encoding protocol, each original pixel in the original secret image is decomposed into at least two visual shared keys containing multiple sub-pixels; wherein, for black pixels and white pixels in the original image, a first visual shared key and a second visual shared key with specific sub-pixel arrangement rules are generated respectively.
[0006] Optionally, the formation of the polarization shared key specifically includes: The first visual shared key is mapped to a preset first polarization channel, and the second visual shared key is mapped to a second polarization channel orthogonal to the first polarization channel, thereby forming a polarization shared key. Optionally, the deep holographic prior neural network includes a deep hologram prior generator for synthesizing phase holograms and an electromagnetic diffraction predictor for predicting the performance of the imaging results; the deep hologram prior generator maps Gaussian noise to a two-dimensional phase hologram; the deep hologram prior generator is trained under given degenerate diffraction conditions and iteratively generates the optimal hologram; the electromagnetic diffraction predictor is constructed based on Rayleigh-Sommerfeld diffraction theory and is used to predict the imaging results of the phase hologram.
[0007] Optionally, the training method for the deep holographic prior neural network is as follows: A phased loss optimization strategy is adopted, which sequentially introduces mean squared error loss and Wasserstein distance to form the final loss function. In the early stage of training, the optimization process is dominated by mean squared error loss, which provides an evaluation criterion for holographic optimization by forcing pixel-by-pixel consistency between the reconstructed image and the target image. In the later stage of training, Wasserstein distance is introduced as a complementary loss term. Wasserstein distance measures the minimum amount of shift required to transform one distribution into another. Wasserstein distance enables the network to capture statistical features more effectively.
[0008] Optionally, the mean squared error loss expression is:
[0009] in, This represents the electric field predicted by the electromagnetic diffraction predictor. Represents the target electric field. This represents the total number of sampling points on the imaging plane.
[0010] Optionally, the Wasserstein distance is defined as:
[0011] in, express and The set of all possible joint distributions between them, symbol This represents the p-norm.
[0012] Optionally, restoring the original secret image specifically includes: After obtaining the optimal phase hologram, a metasurface loading step is performed. The phase distribution of the phase hologram is mapped onto the joint unit structure on the dual-polarization reconfigurable metasurface through an FPGA control platform to set the bias state of the adjustable elements in each metasurface unit. Finally, in the decryption stage, a decryption step is performed. Incident waves of X-polarization and Y-polarization states are generated sequentially through a transmitting antenna to illuminate the reconfigurable metasurface. The electric field intensity distribution under the corresponding polarization channel is collected and recorded on a preset imaging plane to reconstruct the corresponding polarization shared key. Due to the encoding rules set in the encryption process, the original secret image can only be restored and revealed after obtaining the correct pair of polarization shared keys at the same time and performing incoherent intensity superposition. Any detection result under a single polarization channel or an incorrect key combination will only present as a meaningless random noise pattern.
[0013] A polarization-assisted reconfigurable metasurface visual secret-sharing system based on a holographic neural network includes: The shared key generation module performs visual secret sharing encoding on the original secret image to obtain multiple visual shared keys; The polarization mapping module maps the visual shared key to orthogonal polarization channels to form a polarization shared key. The fully optimized module utilizes a deep holographic prior neural network to convert the polarization shared key into a phase hologram; The metasurface loading module loads the phase hologram onto a dual-polarization reconfigurable metasurface. The image decryption module uses orthogonally polarized electromagnetic waves to irradiate a dual-polarized reconstructable metasurface, reconstructs the polarization shared key, and superimposes it for decryption to recover the original secret image.
[0014] As can be seen from the above technical solution, compared with the prior art, this invention discloses a method for visual secret sharing based on a holographic neural network and a polarization-assisted reconfigurable metasurface. This method involves visually sharing and encoding the original secret image to obtain multiple visual shared keys; mapping these visual shared keys to orthogonal polarization channels to form polarization shared keys; using a deep holographic prior neural network to convert the polarization shared keys into phase holograms; loading the phase holograms onto a dual-polarization reconfigurable metasurface; and irradiating the dual-polarization reconfigurable metasurface with orthogonal polarized electromagnetic waves to reconstruct the polarization shared keys and superimpose them for decryption, thus restoring the original secret image. This invention not only significantly reduces the number of control units, system complexity, power consumption, and manufacturing costs; but also improves image reconstruction quality, edge fidelity, and security; furthermore, it supports dynamic switching of multiple secret information and can be extended to fields such as electromagnetic stealth and information camouflage. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0016] Figure 1 This is a schematic diagram of the method architecture provided by the present invention.
[0017] Figure 2 The present invention provides a schematic diagram of the working principle of visual secret sharing encryption and the generation process of polarization shared keys. Among them, (a) shows a schematic diagram of visual secret sharing encryption logic for different secret image pixel attributes, (b) is a sub-pixel sequence set, (c) is a detailed encoding process taking the secret character "C" as an example, (ci) is the original secret image "C", (c-ii) is the X-polarization visual shared key, (c-iii) is the Y-polarization visual shared key, (c-iv) is the ideal effect of the target image "C" recovered by theoretical superposition, (cv) and (c-vi) correspond to the generated X-polarization shared key and Y-polarization shared key, respectively, and (d) and (e) respectively show the visual shared key and the corresponding polarization shared key generated for two other exemplary secret characters "A" and "B". Figure 3 The present invention provides a design architecture and detailed network model for a deep hologram prior neural network; wherein (a) is the design architecture and (b) is the detailed internal hierarchical connection model. Figure 4 This is a comparison chart of simulation results for the polarization shared key corresponding to the exemplary secret character "U" and its decryption and recovery image under different system control complexity conditions provided by the present invention. Figure 5 This is a schematic diagram illustrating the simulation verification results of the present invention, which demonstrates the dynamic encoding and reconstruction of multiple sets of different secret images using the same dual-polarization reconfigurable metasurface under two different system control complexity configurations. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] This invention discloses a method for sharing visual secrets of polarization-assisted reconfigurable metasurfaces based on holographic neural networks, comprising: S1. Visual secret sharing encoding is performed on the original secret image to obtain multiple visual sharing keys; In this step, the original secret image to be encrypted is first received. This original secret image can be visually readable information such as characters, graphics, or QR codes. According to a visual secret sharing encoding protocol, each original pixel in the original secret image is decomposed into at least two visual shared keys containing multiple sub-pixels. Specifically, for black and white pixels in the original image, a first visual shared key and a second visual shared key with specific sub-pixel arrangement rules are generated respectively. A single visual shared key matrix statistically represents a random binary noise pattern. When acquired individually, it does not contain any structured information that identifies the features of the original image, thus ensuring high information concealment under a single key channel. This encoding method utilizes the superposition integral effect of the human visual system, providing a foundation for subsequent simple optical decryption at the physical layer without mathematical calculations.
[0020] S2. Map the visual shared key to the orthogonal polarization channel to form a polarization shared key; After obtaining at least two visual shared keys, a polarization mapping step is further performed to introduce the polarization degree of freedom of the physical layer electromagnetic waves as an additional encryption dimension. In this step, the first visual shared key is mapped to a preset first polarization channel, and the second visual shared key is mapped to a second polarization channel orthogonal to the first polarization channel, thereby forming a polarization shared key. For example, the first visual shared key can be mapped to an X-ray polarization channel, and the second visual shared key can be mapped to a Y-ray polarization channel. This process converts the black-and-white logic states of subpixels in the visual shared key into the strength logic of the electromagnetic wave electric field intensity distribution in a specific polarization direction. By utilizing the natural orthogonality between electromagnetic wave polarization states, this step constructs two parallel and non-interfering encrypted transmission channels within the same physical aperture, significantly improving the parallel capacity of information encoding without increasing the physical size of the metasurface device, and effectively suppressing inter-channel crosstalk during key reconstruction.
[0021] S3. Use a deep holographic prior neural network to convert the polarization shared key into a phase hologram; After generating the polarization shared key, a holographic optimization step is performed to generate a physical phase hologram for driving the dual-polarization reconfigurable metasurface. This step is the core of the invention and is implemented using a deep holographic prior neural network. The deep holographic prior neural network consists of a deep holographic prior generator and an electromagnetic diffraction predictor.
[0022] For polarization-assisted vector nonlocal encryption schemes, this neural network aims to optimize the corresponding phase hologram for any pair of polarization state key pairs (PSKs) in the secret image. To achieve this goal while maintaining physical feasibility, a deep holographic prior neural network is introduced, comprising two components: a deep hologram prior generator for synthesizing the phase hologram, and an electromagnetic diffraction predictor for estimating the performance of the imaging result. The deep hologram prior generator incorporates Gaussian noise... Mapped to a two-dimensional phase hologram , is represented as:
[0023] in, This represents the parameters of the neural network. The generator operates under given degenerate diffraction conditions. The process involves training and iteratively generating the optimal hologram. Formally, the inverse hologram design problem can be reduced to the following energy minimization process:
[0024] in, This represents data fidelity items for a specific task, while It is a regularization term that captures general prior information. For the inverse optimization problem of a single-image hologram, the regularization term... This can be replaced by the implicit priors learned by the neural network. This leads to the following formula:
[0025] in, The phase hologram is obtained through a gradient-based optimization process starting from random initialization. Furthermore, adaptive pooling and upsampling layers are introduced into the generator to adjust the joint unit pattern of the generated phase hologram and the complexity of system control. An electromagnetic diffraction predictor, based on rigorous Rayleigh-Sommerfeld diffraction theory, is used to predict the imaging results of this phase hologram.
[0026] The electromagnetic diffraction predictor is constructed based on Rayleigh-Sommerfeld diffraction theory, and its mathematical expression is as follows:
[0027] in, This represents the complex amplitude field distribution on the aperture plane (or metasurface). For the operating wavelength, It represents the angle between the direction of propagation and the normal to the aperture plane.
[0028] Compared to the Fresnel or Fraunhofer approximations, the Rayleigh-Sommerfeld diffraction formula does not rely on paraxial approximation assumptions, and maintains higher computational accuracy, especially in applications involving short propagation distances, large numerical apertures, or strong phase modulation. This characteristic makes Rayleigh-Sommerfeld diffraction theory particularly suitable for accurate modeling and numerical simulation of reconfigurable metasurfaces, near-field holographic imaging, and high-resolution electromagnetic wave manipulation systems.
[0029] It is worth noting that the implicit prior conditions enable this neural network to converge directly from random noise without relying on a pre-collected training dataset, thus achieving fully unsupervised hologram synthesis. Furthermore, the completely decoupled architecture between the generator and the electromagnetic diffraction predictor provides high flexibility. Specifically, the depth and structure of the generator, the underlying physical electromagnetic model, and the function expression of the predictor can all be independently adjusted and adapted for different tasks (such as electromagnetic illusions and invisibility cloaks). This flexibility allows the network to capture implicit electromagnetic features and achieve strong optimization capabilities, thereby broadening the applicability of this deep hologram prior neural network framework.
[0030] To guide the iterative update of the parameters of the aforementioned deep hologram prior neural network, this embodiment employs a staged loss optimization strategy.
[0031] The loss function quantifies the difference between the network output and the target, and provides direction for parameter updates during backpropagation. In the proposed deep holographic prior neural network, a staged optimization strategy is employed, sequentially introducing mean squared error loss and Wasserstein distance to construct the final loss function. In the early stages of training, the optimization process is dominated by the mean squared error loss, whose expression is:
[0032] in, This represents the electric field predicted by the electromagnetic diffraction predictor. Represents the target electric field. The total number of sampling points on the imaging plane. Mean squared error loss provides an intuitive and effective criterion for holographic optimization by forcing pixel-by-pixel consistency between the reconstructed image and the target image. This characteristic allows the network to converge quickly in the early stages of training and ensures accurate recovery of the overall intensity distribution and main structural features, thus reliably preserving the global contour and contrast of the reconstructed image. However, as a point-by-point error metric, mean squared error loss is inherently insensitive to higher-order statistical correlations. Optimization based solely on mean squared error often leads to overly smooth reconstruction results, blurred edges, and suppression of high-frequency details, which are crucial for resolving sharp boundaries and fine structures in visually secret shared images.
[0033] To address the aforementioned limitations, Wasserstein distance is introduced as a complementary loss term in the later stages of training. Wasserstein distance measures the loss of a distribution... Transform into another distribution The minimum amount of movement required is defined as:
[0034] in, express and The set of all possible joint distributions between them, symbol This represents the p-norm. The Wasserstein distance provides a more suitable metric for quantifying the difference between two distributions, enabling networks to capture statistical features more effectively. This distribution-level constraint significantly enhances the robustness of reconstructed images to noise, suppresses non-physical artifacts, and exhibits a clear advantage in edge preservation.
[0035] S4. Load the phase hologram onto a dual-polarization reconfigurable metasurface; The optimal phase hologram is obtained by solving the above steps. Next, the metasurface loading step is performed. This step uses an FPGA control platform to map the phase distribution of the phase hologram onto the various joint unit structures on the dual-polarization reconfigurable metasurface, so as to set the bias state of the adjustable elements in each metasurface unit.
[0036] S5. Irradiate the dual-polarized reconstructable metasurface with orthogonally polarized electromagnetic waves, reconstruct the polarization shared key and superimpose it for decryption to restore the original secret image.
[0037] During the decryption phase, decryption steps are performed. Incident waves in X-polarized and Y-polarized states are sequentially generated via a transmitting antenna to illuminate the reconfigurable metasurface. The electric field intensity distribution under the corresponding polarization channel is acquired and recorded on a preset imaging plane to reconstruct the corresponding polarization shared key. Due to the encoding rules set during the encryption process, the original secret image can only be restored and revealed after simultaneously acquiring the correct pair of polarization shared keys and performing incoherent intensity superposition; any detection result under a single polarization channel or an incorrect key combination will only present a meaningless random noise pattern.
[0038] In summary, this invention maps visual secret sharing codes to orthogonal physical channels of dual-polarization metasurfaces and introduces a deep hologram prior neural network with complexity control capabilities for hologram inversion solving. This significantly reduces system control complexity while achieving high-security, high-fidelity, and dynamically switchable multi-secret switching physical layer encryption.
[0039] Specifically, Figure 1The diagram shows the overall architecture and unit structure of the dual-polarization reconfigurable metasurface visual encryption system based on a deep holographic prior network proposed in this invention.
[0040] like Figure 1 As shown in (a), the encrypted communication framework constructed by this invention relates to a typical end-to-end secure transmission scenario. In this scenario, the information sender (e.g., user John) intends to transmit confidential information to the information receiver (e.g., user Robert). To ensure the security and integrity of the secret message during transmission, the secret message is first decomposed according to a preset polarization-assisted visual secret sharing encoding protocol. Specifically, each secret character (e.g., "A", "B", "C" as exemplarily shown in the figure) is decomposed into a pair of visual shared keys, namely a first shared key and a second shared key. Subsequently, in order to introduce a physical layer security barrier, the two shared keys are respectively associated with two orthogonal polarization channels, namely the X-polarization channel and the Y-polarization channel, thereby forming corresponding polarization shared keys. Through this encoding and mapping mechanism, any third party that has not obtained the correct polarization pairing information can only observe an irregular random noise pattern under a single polarization channel and cannot obtain effective information content.
[0041] like Figure 1 As shown in (b), after generating the polarization shared key, this invention employs a deep hologram prior neural network to perform hologram optimization calculations. This neural network possesses complexity control capabilities, enabling it to generate phase holograms with different system control complexities based on preset hardware control constraints. The generated phase holograms are re-encoded and mapped onto the various control units of the dual-polarization reconfigurable metasurface via a field-programmable gate array (FPGA)-based control platform. During the decryption phase, the dual-polarization reconfigurable metasurface is sequentially irradiated with X-polarized and Y-polarized incident electromagnetic waves, thereby reconstructing the corresponding polarization shared keys on a preset imaging plane. When the receiver obtains the correct pair of polarization shared keys and performs an incoherent intensity superposition operation, the original secret information is successfully recovered. It should be noted that any incorrect illumination strategy or mismatched key superposition operation will result in the output of only a meaningless, chaotic image.
[0042] like Figure 1 As shown in (c) Figure 1 (c) Specifically, a top view and a three-dimensional schematic diagram of the dual-polarization tunable metasurface unit proposed in this invention are presented. This metasurface unit originates from a simplified 2×2 unit topology. Its top-layer metal pattern structure is designed to provide enhanced electromagnetic modulation stability while significantly reducing the number of active lumped elements, thereby effectively reducing fabrication complexity and hardware material costs. In one specific embodiment, the duty cycle length of this metasurface unit is... The design is approximately two-thirds of the operating wavelength (approximately 22 mm at 10 GHz). Further electromagnetic simulation results show that, over a wide bandwidth from 9.9 GHz to 10.9 GHz, this metasurface unit structure can achieve an independent phase modulation dynamic range of over 320 degrees for both X-polarized and Y-polarized reflected electromagnetic waves.
[0043] Figure 2 The working principle of visual secret sharing encryption and the generation process of polarization shared key are explained in detail.
[0044] like Figure 2 As shown in (a), a schematic diagram of the visual secret sharing encryption logic for different secret image pixel attributes is illustrated. This invention introduces a polarization-assisted sharing encoding strategy that combines polarization multiplexing with visual secret sharing. According to this encoding rule, white pixels in the original secret image are encoded as being represented by two sub-pixel sequences with orthogonal polarization states pointing in opposite directions; black pixels in the original secret image are encoded as being represented by two sub-pixel sequences with parallel polarization states pointing in opposite directions. This encoding rule embeds an inherent polarization-assisted access condition: a single horizontal or vertical sub-pixel sequence is only sufficient to decode the logical information corresponding to the black pixel, while correct reconstruction of the white pixel requires the simultaneous participation of two orthogonal polarization channels. Thus, each original secret image is decomposed into a pair of visual shared keys consisting of randomly distributed horizontal and vertical sub-pixel sequences.
[0045] like Figure 2 As shown in (c), the figure further illustrates the detailed encoding process using the secret character "C" as an example. Figure 2 (ci) represents the original secret image "C". By replacing... Figure 2 The sub-pixel sequence set shown in (b) yielded the following: Figure 2 The X-polarization vision shared key shown in (c-ii) and Figure 2 The Y-polarized vision shared key shown in (c-iii) is a key for vision. Figure 2 (c-iv) demonstrates the ideal result of the target image “C” recovered by theoretical superposition. Figure 2 (cv) and Figure 2 (c-vi) correspond to the generated X-polarization shared key and Y-polarization shared key, respectively. Similarly, Figure 2 (d) and Figure 3 (e) Visual shared keys and corresponding polarization shared keys generated for two other exemplary secret characters “A” and “B” are shown respectively. It can be clearly observed that each secret image is encoded as a pair of independent polarization shared keys, and the hidden information (e.g., the character “C”) only emerges at the visual perception level when the two are incoherently superimposed.
[0046] Figure 3 The design architecture and detailed network model of the deep hologram prior neural network are shown.
[0047] like Figure 3 As shown in (a), to convert the aforementioned polarization shared key into a physical phase hologram capable of practically driving metasurface devices, this invention designs a deep hologram prior neural network. This network architecture consists of two core components: a deep hologram prior generator for synthesizing phase holograms, and an electromagnetic diffraction predictor for estimating the performance of the imaging results. In the specific workflow, the deep hologram prior generator receives a Gaussian noise vector. As input, it is mapped to output as a two-dimensional phase hologram. The process can be described as follows: ,in The neural network weight parameters within the generator are characterized. The generator is iteratively trained under given degenerate diffraction constraints and gradually converges to produce the optimal hologram distribution. .
[0048] To achieve physical fabrication feasibility and reduce control overhead, this invention introduces a complexity control mechanism into the generator network structure. Specifically, an adaptive pooling layer and an upsampling layer are integrated at the end of the generator. This mechanism can enforce constraints on adjacent elements in the generated phase hologram. Each pixel shares the same phase value, thus physically corresponding to the... The structure comprises a joint unit cell composed of metasurface units. Furthermore, the electromagnetic diffraction predictor is built upon rigorous Rayleigh-Sommerfeld diffraction theory. Its function is to calculate the diffraction field propagation of the generated candidate phase holograms in each iteration cycle to predict their electric field intensity distribution on the imaging plane. Notably, the implicit prior conditions utilized in this invention enable the neural network to directly converge from random noise to the target distribution without any pre-collected training dataset, achieving fully unsupervised hologram synthesis.
[0049] like Figure 4 As shown in (b), the figure illustrates the detailed internal hierarchical connection model of the deep holographic prior neural network. This network architecture employs a design paradigm of complete decoupling between the generator and the electromagnetic diffraction predictor.
[0050] Figure 5 This paper presents a comparison of simulation results for the polarization shared key corresponding to the exemplary secret character "U" and its decryption and recovery image under different system control complexities. The figures show the results under different system control complexity parameters. and Under two configurations, the X-polarization shared key and Y-polarization shared key were extracted through simulation calculations, and the secret image was finally recovered through incoherent superposition. The figures clearly show that, under both control complexities, the simulated X-polarization and Y-polarization shared keys exhibit binarized random noise distribution characteristics consistent with theoretical coding expectations. When the corresponding pair of polarization shared keys are incoherently superimposed, the outline of the original secret character "U" becomes clearly visible. (Comparison) and The simulation results show that although reducing control complexity (i.e. increasing the size of the joint unit structure) leads to a slight degradation of recognizable details in the reconstructed image, the overall shape and readability of the secret character are still well preserved.
[0051] Figure 5 Simulation results demonstrate the dynamic encoding and reconstruction of multiple sets of different secret images using the same dual-polarization reconfigurable metasurface under two different system control complexity configurations. As shown, thanks to the independent dual-polarization tunable characteristics of the supercell designed in this invention, a single reconfigurable metasurface hardware platform can support the dynamic switching and encoding display of multiple independent secret images. The figure focuses on showcasing the system control complexity at various levels. and Under certain conditions, after applying seven different key sequences sequentially, different visual secret images are reconstructed through simulation calculations. Exemplary secret images include, but are not limited to, the characters "J", "U", and "1". Simulation results show that under two different control complexity configurations, all reconstructed secret images maintain clear contrast and recognizable character shapes. This demonstrates that the method of this invention can significantly reduce hardware control complexity while still ensuring high-quality recovery of multi-channel, multi-image encrypted information.
[0052] A polarization-assisted reconfigurable metasurface visual secret-sharing system based on a holographic neural network includes: The shared key generation module performs visual secret sharing encoding on the original secret image to obtain multiple visual shared keys; The polarization mapping module maps the visual shared key to orthogonal polarization channels to form a polarization shared key. The fully optimized module utilizes a deep holographic prior neural network to convert the polarization shared key into a phase hologram; The metasurface loading module loads the phase hologram onto a dual-polarization reconfigurable metasurface. The image decryption module uses orthogonally polarized electromagnetic waves to irradiate a dual-polarized reconstructable metasurface, reconstructs the polarization shared key, and superimposes it for decryption to recover the original secret image.
[0053] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0054] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for sharing visual secrets of a polarization-assisted reconfigurable metasurface based on a holographic neural network, characterized in that, include: The original secret image is visually secret-shared encoded to obtain multiple visual sharing keys; The visual shared key is mapped to orthogonal polarization channels to form a polarization shared key; A deep holographic prior neural network is used to convert a polarization shared key into a phase hologram. The phase hologram is loaded onto a dual-polarization reconfigurable metasurface; By irradiating a dual-polarization reconstructable metasurface with orthogonally polarized electromagnetic waves, the polarization shared key is reconstructed and superimposed for decryption, thus restoring the original secret image.
2. The method for sharing visual secrets of a polarization-assisted reconfigurable metasurface based on a holographic neural network according to claim 1, characterized in that, The specific steps for obtaining multiple visual shared keys are as follows: According to the visual secret sharing encoding protocol, each original pixel in the original secret image is decomposed into at least two visual shared keys containing multiple sub-pixels; wherein, for black pixels and white pixels in the original image, a first visual shared key and a second visual shared key with specific sub-pixel arrangement rules are generated respectively.
3. The method for sharing visual secrets of a polarization-assisted reconfigurable metasurface based on a holographic neural network according to claim 2, characterized in that, The formation of the polarization shared key specifically includes: The first visual shared key is mapped to a preset first polarization channel, and the second visual shared key is mapped to a second polarization channel orthogonal to the first polarization channel, thereby forming a polarization shared key.
4. The method for sharing visual secrets of a polarization-assisted reconfigurable metasurface based on a holographic neural network according to claim 1, characterized in that, The deep holographic prior neural network includes a deep hologram prior generator for synthesizing phase holograms and an electromagnetic diffraction predictor for predicting the performance of the imaging results. The deep hologram prior generator maps Gaussian noise to a two-dimensional phase hologram. The deep hologram prior generator is trained under given degenerate diffraction conditions and iteratively generates the optimal hologram. The electromagnetic diffraction predictor is constructed based on Rayleigh-Sommerfeld diffraction theory and is used to predict the imaging results of the phase hologram.
5. The method for sharing visual secrets of a polarization-assisted reconfigurable metasurface based on a holographic neural network according to claim 4, characterized in that, The training method for the deep holographic prior neural network is as follows: A phased loss optimization strategy is adopted, which sequentially introduces mean squared error loss and Wasserstein distance to form the final loss function. In the early stage of training, the optimization process is dominated by mean squared error loss, which provides an evaluation criterion for holographic optimization by forcing pixel-by-pixel consistency between the reconstructed image and the target image. In the later stage of training, Wasserstein distance is introduced as a complementary loss term. Wasserstein distance measures the minimum amount of shift required to transform one distribution into another. Wasserstein distance enables the network to capture statistical features more effectively.
6. The method for sharing visual secrets of a polarization-assisted reconfigurable metasurface based on a holographic neural network according to claim 5, characterized in that, The expression for the mean square error loss is: in, This represents the electric field predicted by the electromagnetic diffraction predictor. Represents the target electric field. This represents the total number of sampling points on the imaging plane.
7. The method for sharing visual secrets of a polarization-assisted reconfigurable metasurface based on a holographic neural network according to claim 5, characterized in that, The Wasserstein distance is defined as follows: in, express and The set of all possible joint distributions between them, symbol This represents the p-norm.
8. The method for sharing visual secrets of a polarization-assisted reconfigurable metasurface based on a holographic neural network according to claim 1, characterized in that, The restoration of the original secret image specifically includes: After obtaining the optimal phase hologram, a metasurface loading step is performed. The phase distribution of the phase hologram is mapped onto the joint unit structure on the dual-polarization reconfigurable metasurface through an FPGA control platform to set the bias state of the adjustable elements in each metasurface unit. Finally, in the decryption stage, a decryption step is performed. Incident waves of X-polarization and Y-polarization states are generated sequentially through a transmitting antenna to illuminate the reconfigurable metasurface. The electric field intensity distribution under the corresponding polarization channel is collected and recorded on a preset imaging plane to reconstruct the corresponding polarization shared key. Due to the encoding rules set in the encryption process, the original secret image can only be restored and revealed after obtaining the correct pair of polarization shared keys at the same time and performing incoherent intensity superposition. Any detection result under a single polarization channel or an incorrect key combination will only present as a meaningless random noise pattern.
9. A polarization-assisted reconfigurable metasurface visual secret sharing system based on a holographic neural network, characterized in that, The method for sharing visual secrets of a polarization-assisted reconfigurable metasurface based on any one of claims 1-8 includes: The shared key generation module performs visual secret sharing encoding on the original secret image to obtain multiple visual shared keys; The polarization mapping module maps the visual shared key to orthogonal polarization channels to form a polarization shared key. The fully optimized module utilizes a deep holographic prior neural network to convert the polarization shared key into a phase hologram; The metasurface loading module loads the phase hologram onto a dual-polarization reconfigurable metasurface. The image decryption module uses orthogonally polarized electromagnetic waves to irradiate a dual-polarized reconstructable metasurface, reconstructs the polarization shared key, and superimposes it for decryption to recover the original secret image.