A Neural Network-Based Achromatic Broadband Metasurface Holographic Imaging Method

Abstract

This invention discloses a neural network-based achromatic broadband metasurface holographic imaging method, belonging to the field of metasurface holographic imaging. The method includes designing a dual-ring polarization conversion unit structure; constructing a deep learning network model that adaptively optimizes structural and spatial dispersion to generate an achromatic broadband phase hologram; and filling the corresponding holographic metasurface using the dual-ring polarization conversion unit based on the achromatic broadband phase hologram generated by the neural network. This invention aims to consider dispersion relations and combine them with neural network calculation methods to achieve broadband holographic imaging. It can rapidly generate a broadband effective holographic metasurface based on the target image and the determined dispersion relations. The fabrication process is simple, highly practical, and suitable for future broadband applications.

Description

Technical Field

[0001] This invention relates to the field of metasurface holographic imaging technology, and more specifically to an achromatic broadband metasurface holographic imaging method based on neural networks. Background Technology

[0002] Metasurfaces are ultrathin two-dimensional arrays of metamaterials that can effectively manipulate the phase, amplitude, and polarization of electromagnetic waves at the subwavelength scale. In recent years, they have become a hot topic in metamaterials and nanoscience research. Compared to traditional metamaterials, metasurfaces are much thinner than the wavelength, exhibiting low loss and ease of fabrication. Their unique physical properties and flexible control over electromagnetic waves make them promising for applications in antenna technology, microwave and terahertz devices, and many other fields.

[0003] Holographic imaging is a technique that uses the principles of interference and diffraction to record and reproduce objects. It can record not only the amplitude but also the phase of the object's light waves. Under appropriate conditions, when a hologram is illuminated with light, all the information of the object can be reconstructed. Observers can view the object from different angles, creating a three-dimensional effect. With the advent of the information age, computational holography, through the powerful computing capabilities of computers, digitizes the entire holographic process, thus obtaining all the information of the holographic surface. Compared to traditional optical holography, computational holography not only avoids introducing phase differences and noise but also enables the display of virtual objects, greatly improving accuracy and flexibility and significantly expanding the application range of holographic technology. Currently, holographic imaging technology is widely used in information storage, remote sensing, and biomedicine.

[0004] Due to their excellent controllability, metasurfaces are very suitable for designing computational holographic devices that require control over parameters such as amplitude and phase. The combination of metasurfaces and holographic imaging technology is also one of the current research hotspots in nanotechnology, optics, and electromagnetics. However, holographic devices designed using metasurface holography often only have an imaging effect at a certain frequency point due to dispersion phenomena, and cannot achieve imaging in a certain frequency band, which undoubtedly greatly limits the application of metasurface holography technology.

[0005] Therefore, proposing a metasurface holography method to quickly calculate a broadband effective phase metasurface hologram for the imaging target and realize broadband holographic imaging is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a dispersion-agnostic broadband metasurface holographic imaging method based on neural networks, which aims to combine dispersion relations with neural network calculation methods to achieve broadband holographic imaging.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A neural network-based achromatic broadband metasurface holographic imaging method, comprising the following steps:

[0009] Design a dual-ring polarization conversion unit and determine the phase response curve and structural dispersion relationship of the dual-ring polarization conversion unit;

[0010] A neural network model is constructed, an imaging target image is input, and an achromatic broadband phase hologram is generated; the achromatic broadband phase hologram is output with the phase response at a certain frequency point;

[0011] The achromatic broadband phase hologram generated by the neural network model and the dual-ring polarization conversion unit are used to fill and form the corresponding metasurface hologram.

[0012] Preferably, the dual-ring polarization conversion unit includes a dielectric substrate, a back reflective metal plate, and a front dual-ring structure. Phase modulation at a certain frequency point is achieved by changing the opening angle and radius of the dual rings in the front dual-ring structure. The opening angle and radius of the dual rings corresponding to a certain phase are determined by simulation using CST electromagnetic simulation software. At the same time, the structural dispersion relation is directly derived from the CST electromagnetic simulation software.

[0013] Preferably, the neural network model includes a convolutional neural network generator, a metasurface layer, and a diffraction layer;

[0014] The image of the target being imaged is input into the convolutional network generator to generate a phase hologram at a certain frequency point within the frequency band. The phase hologram at a certain frequency point within the frequency band is input into the metasurface layer, and a phase hologram at each frequency point within the frequency band is fitted according to the structural dispersion relationship. The phase holograms at each frequency point within the frequency band together form the phase hologram of the corresponding frequency band. The phase hologram of the corresponding frequency band is input into the diffraction layer, and an achromatic broadband phase hologram of the corresponding frequency band is generated according to the spatial dispersion relationship.

[0015] Preferably, the neural network model achieves parameter convergence by inputting a fixed random array and minor disturbances.

[0016] Preferably, the mean square error between the achromatic broadband phase hologram and the imaging target is calculated and used for updating and converging the parameters of the neural network model.

[0017] Preferably, the neural network model is further provided with a parameter update filter to ensure that the physical parameters do not change as the neural network model is trained.

[0018] Preferably, the neural network model automatically saves the phase hologram of the best imaging during the convergence process.

[0019] Preferably, the process of filling the corresponding metasurface hologram with the achromatic broadband phase hologram generated by the neural network model and the dual-ring polarization conversion unit includes:

[0020] The phase magnitude is determined based on the achromatic broadband phase hologram, and the corresponding metasurface hologram is obtained by filling it according to the corresponding double-ring opening angle and radius determined during simulation.

[0021] As can be seen from the above technical solution, compared with the prior art, this invention discloses an achromatic broadband metasurface holographic imaging method based on neural networks. The required phase response characteristics and structural dispersion relationship are determined through simulation using a dual-ring polarization conversion unit, and a neural network model is constructed. A phase hologram at a specific frequency point is generated using a convolutional neural network generator. Based on structural and spatial dispersion characteristics, a metasurface layer and a diffraction layer are designed to optimize the broadband holographic imaging effect. The achromatic broadband phase hologram output by the dual-ring polarization conversion unit and the neural network model is filled to form a complete metasurface hologram for the imaging target. This invention considers dispersion relationships and combines them with neural network calculation methods to achieve broadband holographic imaging. It can quickly generate a broadband effective holographic metasurface based on the target imaging and the determined dispersion relationship. The fabrication process is simple, highly practical, and suitable for future broadband application needs. Attached Figure Description

[0022] 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.

[0023] Figure 1(a) is a schematic diagram of the overall structure of the dual-ring polarization conversion unit provided in an embodiment of the present invention.

[0024] Figure 1(b) is a schematic diagram of the front double-ring structure provided in an embodiment of the present invention.

[0025] Figure 2(a) shows the phase response characteristic curve of the dual-ring polarization conversion unit provided in the embodiment of the present invention.

[0026] Figure 2(b) Dispersion characteristic curve of the dual-ring polarization conversion unit provided in the embodiment of the present invention.

[0027] Figure 3 This is a schematic diagram of the neural network model structure provided in an embodiment of the present invention.

[0028] Figure 4(a) is a schematic diagram of the effectiveness of the 9G frequency point of the present invention.

[0029] Figure 4(b) is a schematic diagram of the effectiveness of the 10G frequency point of the present invention.

[0030] Figure 4(c) is a schematic diagram of the effectiveness of the 11G frequency point of the present invention.

[0031] Wherein, 1-substrate, 2-front double-ring structure; p-period length of unit, h-thickness parameter of unit dielectric substrate, α-opening angle of the ring, d-interval between outer and inner ring radii, and r-size of outer ring radius. Detailed Implementation

[0032] 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.

[0033] This invention discloses an achromatic broadband metasurface holographic imaging method based on neural networks, comprising the following steps:

[0034] Design a double-ring polarization conversion unit and determine the phase response curve and structural dispersion relationship of the double-ring polarization conversion unit;

[0035] A neural network model is constructed, the imaging target is input, and an achromatic broadband phase hologram is generated; the achromatic broadband phase hologram is output with the phase response at a certain frequency point;

[0036] The achromatic broadband phase hologram generated by the neural network model and the double-ring polarization conversion unit are used to fill and construct the corresponding metasurface hologram.

[0037] In this embodiment, the overall structure of the double-ring polarization conversion unit is shown in Figure 1(a), including a dielectric substrate 1, a front double-ring structure 2, and a back reflective metal plate (not shown in the figure); the front double-ring structure 2, as shown in Figure 1(b), achieves phase modulation by changing the double-ring opening angle α and radius r, and can provide a phase shift of -π to π. The phase response characteristic curve is shown in Figure 2(a), and the reflectivity is as high as 95%; the double-ring opening angle and radius corresponding to a certain phase are determined by simulation using CST electromagnetic simulation software; at the same time, the dispersion characteristic curve is directly exported by CST electromagnetic simulation software, as shown in Figure 2(b).

[0038] Specifically, the CST electromagnetic simulation software can determine that the change in aperture angle from 60° to 157° plus the change in radius from 3.6 to 4.2 can provide the change from -π to π at 10 GHz. The dispersion relationship at frequencies such as 9 GHz, 10 GHz, and 11 GHz can be directly exported from the CST simulation software.

[0039] The neural network model constructed in this embodiment of the invention includes a convolutional neural network generator, a metasurface layer, and a diffraction layer; wherein, the convolutional neural network generator uses a decoder-encoder network structure to generate a phase hologram of a certain frequency point within a frequency band after inputting an imaging target.

[0040] The specific structure of the neural network model is as follows: Figure 3 As shown, the right side represents the decoder-encoder structure. The target image is processed according to the downsampling functions d1, d2, ..., d... N We obtain a series of corresponding eigenvalues, and then the residual functions s1, s2, ..., s N Each of these is concatenated with its corresponding feature value, and then upsampled using the functions u1, u2, ..., u... N The final phase hologram at 10 GHz was obtained. This phase hologram was then input into the metasurface, and the simulated structural dispersion relations φ(f1), φ(f2), ..., φ(f... N Phase holograms at 9 GHz and 11 GHz are fitted. These phase holograms from 9 GHz, 10 GHz, and 11 GHz together form the phase hologram for the 9 GHz-11 GHz frequency band. This phase hologram is then input into a diffraction layer and processed by spatial dispersion functions coef(f1), coef(f2), ..., coef(f...). N The dissipative broadband phase hologram for this frequency band is obtained. The dissipative broadband phase hologram and the target image are then subjected to MSE calculation for parameter updates and convergence, and for training the neural network model.

[0041] The neural network model is trained unsupervised, which can quickly and adaptively obtain a dispersion-free, wide-band usable phase hologram. The parameters converge by inputting a fixed random array and small disturbances.

[0042] It is worth noting that traditional neural networks converge by using the difference between the reconstructed image and the target image obtained after training the data through the network, and then adjusting the parameters using a backpropagation algorithm. Therefore, traditional network training requires a large amount of diverse sample data (a training sample includes the target image and its corresponding hologram) to form a training set (providing a large number of features) or effective artificial image prior conditions (making it easier to converge to the desired result) to ensure convergence. This is impossible to achieve for the metasurface holographic imaging in this embodiment. The method proposed in this invention only has a target image and no concept of a dataset. Therefore, we designed a generator based on depth image prior conditions. By inputting a fixed random number and adjusting the hologram generated by the generator based on small perturbations, the difference between the reconstructed image and the target image of this hologram is used for parameter adjustment and network convergence. This is equivalent to the original solution space directly pointing to the hologram (without a dataset for training), now constructing a mapping from fixed data to the hologram (small perturbations provide sufficient feature values ​​for training). This eliminates the need for manually designed image prior conditions and a large dataset of diverse data (target image and its corresponding hologram).

[0043] Furthermore, to ensure that the physical parameters do not change during training, a parameter update filter is set during the training of the neural network model to guarantee the effectiveness of imaging. In addition, the neural network model automatically saves the phase hologram of the best imaging during the convergence process.

[0044] By inputting the smiley face imaging target into a neural network model, a broadband effective phase hologram usable in the 9-11 GHz range can be automatically optimized. CST full-wave simulation of this model yields imaging results for 9 GHz, 10 GHz, and 11 GHz. Figures 4(a)-4(c) As shown.

[0045] In this embodiment, the achromatic broadband phase hologram output by the neural network model is output based on the phase response at 10G. Once the phase magnitude is determined, the metasurface hologram can be obtained by filling the model with the corresponding parameters (aperture angle / radius size parameters) determined during simulation.

[0046] The method proposed in this invention can be directly processed using PCB technology, which is simple. At the same time, the designed network structure can achieve broadband imaging effect and has wide practical value.

[0047] 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.

[0048] 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 neural network-based achromatic broadband metasurface holographic imaging method, characterized in that, The steps are as follows: Design a dual-ring polarization conversion unit and determine the phase response curve and structural dispersion relationship of the dual-ring polarization conversion unit; A neural network model is constructed, an imaging target image is input, and an achromatic broadband phase hologram is generated; the achromatic broadband phase hologram is output with the phase response at a certain frequency point; The achromatic broadband phase hologram generated by the neural network model and the dual-ring polarization conversion unit are used to fill and form the corresponding metasurface hologram.

2. The achromatic broadband metasurface holographic imaging method based on neural networks according to claim 1, characterized in that, The dual-ring polarization conversion unit includes a dielectric substrate, a back reflective metal plate, and a front dual-ring structure. Phase modulation is achieved by changing the opening angle and radius of the dual rings in the front dual-ring structure. The opening angle and radius of the dual rings corresponding to a certain phase are determined by simulation using CST electromagnetic simulation software. At the same time, the structural dispersion relation is directly derived from the CST electromagnetic simulation software.

3. The achromatic broadband metasurface holographic imaging method based on neural networks according to claim 1, characterized in that, The neural network model includes a convolutional neural network generator, a metasurface layer, and a diffraction layer; The imaging target is input into the convolutional network generator to generate a phase hologram at a certain frequency point within the frequency band; the phase hologram at a certain frequency point within the frequency band is input into the metasurface layer, and a phase hologram at each frequency point within the frequency band is fitted according to the structural dispersion relationship. The phase holograms at each frequency point within the frequency band together form the phase hologram of the corresponding frequency band; the phase hologram of the corresponding frequency band is input into the diffraction layer, and an achromatic broadband phase hologram of the corresponding frequency band is generated according to the spatial dispersion relationship.

4. The achromatic broadband metasurface holographic imaging method based on neural networks according to claim 1, characterized in that, The neural network model achieves parameter convergence by inputting a fixed random array and minor disturbances.

5. The achromatic broadband metasurface holographic imaging method based on neural networks according to claim 1, characterized in that, The construction of the neural network model, inputting the imaging target image and generating an achromatic broadband phase hologram, also includes: The mean square error between the achromatic broadband phase hologram and the image of the target is calculated and used for updating and converging the parameters of the neural network model.

6. The achromatic broadband metasurface holographic imaging method based on neural networks according to claim 3, characterized in that, The neural network model also includes a parameter update filter.

7. The achromatic broadband metasurface holographic imaging method based on neural networks according to claim 5, characterized in that, The neural network model automatically saves the phase hologram of the best imaging during the convergence process.

8. The achromatic broadband metasurface holographic imaging method based on neural networks according to claim 1, characterized in that, The achromatic broadband phase hologram generated by the neural network model and the dual-ring polarization conversion unit are used to fill and form the corresponding metasurface hologram, including: The phase magnitude is determined based on the achromatic broadband phase hologram, and the corresponding metasurface hologram is obtained by filling it according to the corresponding double-ring opening angle and radius determined during simulation.

Images