A real-time control method of micro-nano light field based on three-dimensional tight focusing holographic network

Abstract

The present invention discloses a real-time control method of micro-nano light field based on three-dimensional tight focusing holographic network, constructs and trains 3DCFH-Net based on vector diffraction model: target intensity distribution tensor I tar Input a U‑Net to obtain the initial target complex amplitude, derive the complex amplitude distribution of the objective rear pupil plane through the reverse diffraction propagation model, and superimpose to obtain the initial phase #imgabs0#; #imgabs1# and the incident laser amplitude of the SLM are input into another U‑Net, and the final phase #imgabs2# is output; according to the forward diffraction propagation model, reconstruct the intensity distribution of the tightly focused light field under the X-polarization component and full polarization, and obtain its loss value relative to the actual target light field intensity distribution, and use the gradient descent and back propagation algorithms to adjust the network parameters; the I tar The trained 3DCFH‑Net is input, and the instantly generated #imgabs3# is loaded into the SLM, and the micro-nano light field is obtained through optical system modulation. The present invention can realize large-scale generation of holographic phase with high speed and high precision.

Description

Technical Field

[0001] The present invention relates to the field of micro-nano light field control, and in particular to a real-time control method of micro-nano light field based on a three-dimensional tightly focused holographic network. Background Art

[0002] Light field control technology has penetrated into many scientific and technological frontiers, such as micro-nanostructure manufacturing, optical tweezers, neural light stimulation, optical communication, and optical storage. In these application scenarios, high-speed holographic phase generation capabilities are crucial to achieve rapid manipulation of tightly focused light fields. However, the mainstream phase generation strategies of tight focusing holography (TFH) mostly rely on iterative methods or non-convex optimization methods, which often sacrifice speed in pursuit of high precision, take a long time in the loop or gradient descent process, and are difficult to adapt to real-time operation scenarios.

[0003] In recent years, deep learning has shown great advantages in the field of holographic imaging and display (IDH), with significant improvements in speed and accuracy, providing new opportunities for tightly focused light field modulation. However, TFH aims to construct a complex light field composed of a finite number of discrete focal points. The modulation area is usually concentrated in the central area of ​​the field of view, and the overall data distribution shows strong sparsity and irregularity. Compared with IDH, which relies on high-resolution images, it has weaker constraints on the reconstructed light field, resulting in increased instability in solving the inverse problem and increasing the difficulty of finding the global optimal solution under large-scale data sets. In addition, in TFH, the high numerical aperture objective lens causes rapid changes in the focal field along the propagation direction, and there is a large difference in the light field distribution between the holographic plane and the reconstruction plane, which undoubtedly increases the complexity of network learning. In addition, TFH has more stringent requirements on the energy consistency and utilization between the reconstructed focal points. In summary, the TFH network still faces many challenges in pursuing high reconstruction accuracy and generalization ability. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a real-time control method for micro-nano light fields based on a three-dimensional tight focusing holographic network. It utilizes a deep adaptive three-dimensional tight focusing holographic network framework driven by a vector diffraction model, fully considers the tight focusing characteristics and spherical aberration effects of a high numerical aperture (NA) objective lens, and adopts a layer-based learning strategy to strengthen the global constraints of reconstructing the three-dimensional focusing field. By embedding a deep adaptive module in the holographic network, the flexibility and versatility of its modulated focal field are enhanced, and large-scale generation of high-speed and high-precision holographic phase is achieved.

[0005] The specific technical solutions are as follows:

[0006] A method for real-time control of micro-nano light fields based on a three-dimensional tightly focused holographic network comprises the following steps:

[0007] S1: According to the randomly generated three-dimensional tightly focused target light field intensity distribution, the corresponding target intensity distribution tensor I is obtained tar , the training data set is obtained;

[0008] S2: Construct a three-dimensional tightly focused holographic network 3DCFH-Net, and input the training data set into the network for training; the 3DCFH-Net uses two cascaded U-Net modules, and the training process is:

[0009] Will I tar Input the first U-Net module to establish the mapping relationship between the noise-free target light field intensity and its optimal complex amplitude in the X polarization direction, and obtain the initial amplitude of each target z plane and initial phase ; Then the complex amplitude distribution of the rear pupil plane is derived through the optical reverse diffraction propagation model, and the initial phase on the SLM is obtained by superposition ;Will The incident laser amplitude A of the SLM is input into the second U-Net module, and the final phase is output ; According to the optical forward diffraction propagation model, based on Reconstruct the intensity distribution of the tightly focused light field under the X-polarization component, based on Reconstruct the intensity distribution of the tightly focused light field under full polarization; calculate the loss value between the reconstruction result and the actual target light field, and use the back propagation and gradient descent algorithms to adjust the parameters in the 3DCFH-Net so that the 3DCFH-Net converges to the optimal state and obtains the trained 3DCFH-Net;

[0010] S3: I tar Input into the trained 3DCFH-Net to generate the corresponding final phase ;

[0011] S4: It is synchronously loaded onto the SLM, and then modulated by an optical system including a 4f system and an objective lens to obtain a micro-nano light field, thereby achieving real-time modulation of a large-scale three-dimensional tightly focused light field.

[0012] Furthermore, in S2, the expression of the forward diffraction propagation model is as follows:

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] In the formula, E 0x represents the distribution of X-polarization of the electric field E0 at the entrance pupil plane; H ψ represents the defocus aberration caused by refractive index mismatch, H z represents the phase accumulation along the z axis; k represents the wave number, d represents the detection depth of the objective lens, n1 represents the refractive index of the objective lens, n2 represents the refractive index of the sample, θ1 represents the convergence angle of the objective lens, and θ2 represents the convergence angle of the sample; p x The coefficient of the X-polarization component corresponding to E0, p y The coefficient of the Y polarization component corresponding to E0, p z The coefficient of the Z polarization component corresponding to E0; m 11 、m 21 、m 31 The transformation matrix M=[m wt ] 3×3 The elements in M ​​are related to the electric field at the entrance pupil plane and the vector electric field refraction at the interface, w is the row index of the matrix element, and t is the column index;

[0018] The expression of the reverse diffraction propagation model is as follows:

[0019] ;

[0020] In the formula, Indicates p x The complex conjugate of 2x represents the distribution of X polarization of the target light field, E 2y represents the distribution of the Y polarization of the target light field, E 2z Represents the distribution of Z polarization of the target light field.

[0021] Furthermore, in S2, 3DCFH-Net uses a 3×3 convolution block to transform the target amplitude Mapping into a high-dimensional feature space, the spatial resolution remains unchanged, allowing the network to capture more complex local and global patterns without reducing spatial details, enhancing its feature learning and generalization capabilities; the mapping result is expressed as , C0 represents I tar The number of channels, H represents Height, W represents The width of C1 The number of channels; the 3×3 convolution block includes a 3×3 convolution layer, a BN layer, and a ReLU layer; then, the first U-Net module is used to train the features Encode, decode, and output features , C2 represents The number of channels is , which is then passed through a 3×3 convolutional layer to generate a tensor containing the target amplitude and phase of the C0 layer; then, the channel separation is performed and it is divided into two parts, which are activated by ReLU and Hardtanh functions respectively to obtain the X-polarized complex amplitude of all z planes (including the initial phase on the SLM). ); Connect along the channel direction and A become a new tensor, which passes through a 3×3 convolution block, another U-Net submodule and a Hardtanh activation function to obtain the final phase distribution .

[0022] Furthermore, in S2, the loss function L for training 3DCFH-Net is expressed as follows:

[0023] ; ;

[0024] ;

[0025] Where, L cs represents the cosine similarity loss function, L MSE represents the mean square error loss function; η is L cs The weight factor of L MSE The weight factor of represents the intensity distribution of the tightly focused light field under the reconstructed X-polarization component, represents the reconstructed tightly focused light field intensity distribution under all polarizations; represents the average operation, Ω is the set of all focal points in the target light field, represents the target strength in the set Ω, represents the reconstruction strength in the set Ω.

[0026] Furthermore, in the S4, the optical system including the 4f system and the objective lens is a laser parallel direct writing device, which includes: a 3D nano-displacement platform and two cameras, and a femtosecond laser, a half-wave plate, a beam expander, an aperture stop, an SLM, a 4f system, a beam splitter, an objective lens, and a sample arranged in sequence along the optical axis; the femtosecond laser is the light source, the half-wave plate is used to manipulate the polarization direction of the laser, and the beam expander and the aperture stop are used to amplify and adjust the diameter of the incident light beam; the 4f system is used to scale the output light beam from the SLM and propagate it to the back pupil of the objective lens; the beam splitter is used to divide the incident light beam into three paths, one of which is input into the objective lens, and the objective lens is used to focus the light beam onto the sample; another path is input into camera one, and camera one is used to observe the sample and the laser direct writing process in real time; the third path is input into camera two, and camera two is used to obtain the intensity distribution of the focal plane at different depths.

[0027] Furthermore, in the S4, the optical system including the 4f system and the objective lens is a fluorescence dynamic display device, which includes: a 3D translation stage and a camera, and a continuous laser, a fiber collimator, a polarizer, a half-wave plate, an aperture stop, a right-angle reflection prism, an SLM, a 4f system, a long-wave pass dichroic mirror, an objective lens, and a sample arranged in sequence along the optical axis; the excitation light beam emitted by the continuous laser is collimated by the fiber collimator, and then passes through the polarizer and the half-wave plate in sequence, and the right-angle reflection prism is used to make the light beam enter the SLM at an angle less than 10°; the 4f system is used to scale the output light beam from the SLM and propagate it to the rear pupil plane of the objective lens; the long-wave pass dichroic mirror is used to divide the incident light beam into two paths, one path is input to the objective lens, and the objective lens is used to focus the light beam on the sample; the other path is input to the camera, and the camera is used to observe the sample and the fluorescence dynamic display process in real time.

[0028] Furthermore, each sample tensor in the training data set of S1 contains several focuses, the intensity and position of the focuses are randomly generated, and the focuses are distributed in the region of interest of the entire tensor, and the position and range of the region of interest are determined by the optical aperture and focusing range of the actual optical path system device.

[0029] A micro-nano light field real-time control device based on a three-dimensional tightly focused holographic network is used to implement the micro-nano light field real-time control method based on a three-dimensional tightly focused holographic network, comprising: a training data set acquisition module, a three-dimensional tightly focused holographic network training module, a target light field phase calculation module, a phase loading module, and a three-dimensional tightly focused light field modulation system; the three-dimensional tightly focused light field modulation system is an optical system including a 4f system and an objective lens;

[0030] The training data set acquisition module is used to obtain the corresponding target intensity distribution tensor according to the randomly generated target light field intensity distribution, and integrate it to obtain the training data set; the three-dimensional tightly focused holographic network training module is used to optimize the parameters of 3DCFH-Net through back propagation and gradient descent algorithms; the target light field phase calculation module is used to input the target light field into the trained 3DCFH-Net, and the final holographic phase is loaded into the SLM through the phase loading module, and then modulated by the three-dimensional tightly focused light field modulation system to obtain the micro-nano light field.

[0031] An electronic device includes a memory and one or more processors, wherein the memory stores executable codes, and when the one or more processors execute the executable codes, they are used to implement the real-time control method of micro-nano light field based on three-dimensional tightly focused holographic network.

[0032] A computer-readable storage medium stores a program, which, when executed by a processor, implements the real-time control method of micro-nano light field based on a three-dimensional tightly focused holographic network.

[0033] The beneficial effects of the present invention are:

[0034] (1) The method provided by the present invention fully considers the spherical aberration effect of the optical system and strengthens the constraints on the reconstructed focal field.

[0035] (2) The method provided by the present invention can directly generate an aberration-free multifocal array with higher precision and control flexibility.

[0036] (3) The method provided by the present invention can achieve high-speed and high-precision large-scale generation of holographic phases. Compared with traditional methods, the calculation speed is shortened by 2 to 3 orders of magnitude. It can calculate focusing fields with arbitrary spatial, intensity and axial spacing distributions on a large scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the basic optical system for SLM to modulate the tightly focused light field.

[0038] Figure 2 This is a theoretical diagram of the Debye vector diffraction integral.

[0039] Figure 3 It is a flow chart of a method for real-time control of micro-nano light fields based on a three-dimensional tightly focused holographic network in an embodiment of the present invention.

[0040] Figure 4 3D C F H-Net model in an embodiment of the present invention.

[0041] Figure 5 It is a schematic diagram of the modulation results of 3DCFH-Net on the target light field with five depths in an embodiment of the present invention.

[0042] Figure 6 It is a schematic diagram of the modulation results of 3DCFH-Net on a dense focal array data set in an embodiment of the present invention.

[0043] Figure 7 It is a feasibility verification device of 3DCFH-Net in laser parallel direct writing in an embodiment of the present invention.

[0044] Figure 8 It is a schematic diagram of the effect of processing a hologram generated by 3DCFH-Net in samarium-doped sodium aluminate glass in an embodiment of the present invention.

[0045] Fig. 9It is a schematic diagram of the modulation results of 3DCFH-Net in laser parallel direct writing in an embodiment of the present invention, wherein (a) is the modulation effect of the three-layer target light field arranged in an array, (b) is the modulation effect of the three-layer target light field arranged randomly, (c) is the dynamic evolution process of the holographic focal field generated by 3DCFH-Net under the binary intensity distribution mode, and (d) is the dynamic evolution process of the holographic focal field generated by 3DCFH-Net under the variable intensity distribution mode.

[0046] Fig.10 It is a feasibility verification device of 3DCFH-Net in holographic fluorescent dynamic display in an embodiment of the present invention.

[0047] Fig.11 3DCFH-Net is a schematic diagram of holographic fluorescence dynamic display based on 3DCFH-Net in an embodiment of the present invention, wherein (a) is a two-dimensional code focal array modulated by 3DCFH-Net and its 520nm photoluminescence image generated on perovskite nanocrystal glass, and (b) is a schematic diagram of holographic dynamic display of photoluminescence images (520nm, 580nm, 600nm, 640nm) of different materials under 405nm excitation focal field.

[0048] Fig.12 Schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0050] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0051] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0052] Before further describing the embodiments of the present invention in detail, the nouns and terms involved in the embodiments of the present invention are described. The nouns and terms involved in the embodiments of the present invention are subject to the following explanations.

[0053] (1) Spatial Light Modulator (SLM) is an optical element that can dynamically change the phase, amplitude, polarization or wavelength of the light wavefront, and control the distribution of the light field through electrical signals or other forms of input. In an optical system, a spatial light modulator can be used as a programmable optical element to control the distribution of the light field through computer-generated images or data, and has extremely high flexibility and programmability. Combined with computational holography technology, SLM can achieve high-quality holographic projection, providing important support for virtual reality and augmented reality applications.

[0054] (2) Micro-nano light field control: By designing and manufacturing micron- or nanometer-level optical elements, such as gratings, lenses, and prisms, we can achieve fine control of light propagation, distribution, phase, amplitude and other characteristics, thereby achieving high-resolution imaging and detection.

[0055] (3) Tight Focusing Holography (TFH) is a technology that uses laser tight focusing technology to enhance the resolution and depth perception of holographic imaging. It can create three-dimensional images with extremely high resolution by precisely controlling the light field through the computationally generated hologram, especially by achieving fine manipulation of the light beam in a very small space. Its technical principles include: laser tight focusing - using a high NA lens or a special optical system to tightly focus the laser beam on a very small spatial scale to form a light spot with high energy density. Computational holography - using computer algorithms to simulate the light propagation process and generate digital holographic patterns based on the desired optical effects. These patterns can be displayed in real time through modulation devices (SLM is used in this invention) to produce the desired light field distribution.

[0056] Below, in order to explain the present application more clearly, firstly, forward diffraction propagation and reverse diffraction propagation are briefly introduced based on vector diffraction theory:

[0057] like Figure 1 As shown in Figure 1, the optical system for tight focus light field modulation uses a 4f optical device to effectively relay and scale the laser beam on the SLM plane to the back pupil plane of the high NA objective. Therefore, 3D tight focus holography can be described as searching for the best phase pattern on the back pupil plane of the objective according to the target intensity distribution. The complex field at the back pupil plane can be expressed as , where A is the incident laser amplitude, φ0 is the phase map loaded on the SLM, and i represents the imaginary unit. Considering the spherical aberration effect of the high NA objective and the change of the polarization component at the entrance pupil, the following is adopted: Figure 2 The Debye vector diffraction integral shown in the figure is used as the theoretical core to achieve fast and accurate simulation and calculation of three-dimensional tightly focused light fields. At the same time, the aberration effect caused by refractive index mismatch is considered in the vector diffraction calculation. The optical axis of the objective lens is the z-axis, and the x-axis and y-axis are located on the vertical plane of the z-axis and are perpendicular to each other. When the laser passes through the high NA objective lens and passes through the interface of two media (i.e., the objective lens and the sample), the Fourier form of the expression of its tightly focused electric field (i.e., the target light field) is:

[0058] (1)

[0059] Where c0 is a constant, F( ) represents the Fourier transform function; E0=[E 0x ,E 0y ,E 0z ] T represents the electric field distribution at the entrance pupil plane, M=[m wt ] 3×3 is a transformation matrix related to the electric field at the entrance pupil plane and the vector electric field refraction at the interface, w is the row index of the matrix element, t is the column index; n1 represents the refractive index of the objective lens, n2 represents the refractive index of the sample, θ1 represents the convergence angle of the objective lens, θ2 represents the convergence angle of the sample; k represents the wave number, and d represents the detection depth of the objective lens, that is, the distance from the theoretical focusing point of the objective lens to the interface of the two media.

[0060] Since SLM requires horizontally polarized light to be incident, the electric field at the entrance pupil plane of the objective lens is E0=[E 0x ,0,0] T , when certain constant terms are ignored, formula (1) can be further simplified as:

[0061] (2)

[0062] In the formula, the vector , p x The coefficient of the X-polarization component of the electric field E0 corresponding to the entrance pupil plane, p y The coefficient of the Y polarization component corresponding to E0, p z The coefficient of the Z polarization component corresponding to E0. 0xrepresents the distribution of X polarization of E0; the first exponential term H ψ =exp[ikd(n2cosθ2-n1cosθ1)] represents the defocus aberration caused by refractive index mismatch. The second exponential term H z =exp[ikn2cosθ2z] represents the phase accumulation along the z-axis.

[0063] When the refractive index of the objective lens is the same as or infinitely close to that of the sample, the convergence angle θ2 is also the same as or infinitely close to θ1. At this time, the exponential term H of the refractive index mismatch ψ Can be ignored.

[0064] Formula (2) is used as a forward diffraction propagation model to reconstruct the target light field from the hologram plane and provide physical guidance for unsupervised convolutional neural networks. Correspondingly, through the reverse diffraction propagation model, the complex amplitude distribution of the entrance pupil plane can also be inferred from the electric field distribution of the target focal field. The expression of the reverse diffraction propagation model is as follows:

[0065] (3)

[0066] In the formula, Indicates p x The conjugate complex number of , and the other parameters with conjugate symbols are the same; E 2x represents the distribution of X polarization of the target light field, E 2y represents the distribution of the Y polarization of the target light field, E 2z Represents the distribution of Z polarization of the target light field.

[0067] Based on the above architecture, the embodiment of the present invention proposes a real-time control method of micro-nano light field based on a three-dimensional tightly focused holographic network, such as Figure 3 As shown, the specific steps include:

[0068] S1: According to the randomly generated three-dimensional tightly focused target light field intensity distribution, the corresponding target intensity distribution tensor I is obtained tar , the training data set is obtained.

[0069] S2: Build a 3D Coarse-to-Fine Holographic Network (3DCFH-Net) framework to transform the target intensity distribution tensor I corresponding to the training dataset tar Input into 3DCFH-Net, train the 3DCFH-Net network parameters, and obtain the trained 3DCFH-Net.

[0070] like Figure 4As shown in the figure, the overall framework of the proposed 3DCFH-Net adopts two cascaded U-Net modules. The first U-Net module establishes the mapping relationship between the noise-free target light field intensity and its optimal complex amplitude in the X-polarization direction, and indirectly generates the initial holographic phase through an inverse transformation operation. Compared with directly learning the nonlinear relationship between the target-phase pair, this strategy emphasizes the inherent constraints of physical laws, provides clearer optimization guidance for network model training, and accelerates the convergence speed. The subsequent second U-Net module plays the role of fine-tuning, improving the prediction accuracy of the network by correcting the initial phase and incorporating the consideration of the vector focused light field. The U-Net module consists of 4 down-convolution blocks and 4 up-convolution blocks by default, in which the upsampling or downsampling of features is achieved through 4×4 convolution or deconvolution operations with a step size of 2, ensuring that key intensity details can be well preserved and reconstructed.

[0071] The overall process of training 3DCFH-Net using the training data set obtained from S1 is as follows: First, the three-dimensional target intensity distribution tensor I tar As the input of the first U-Net module, calculate the initial amplitude of each target z plane and initial phase ; Then, the complex amplitude distribution of the rear pupil plane is derived through the reverse diffraction propagation model, and the initial phase on the SLM is obtained after superposition. ,Right now Next, the phase is fed into the second U-Net module together with the incident laser amplitude A of the SLM, and the final phase is output after a series of convolution operations. .

[0072] In the training phase, according to the optical forward diffraction propagation model, based on the initial phase Reconstruct the intensity distribution of the tightly focused light field under the X-polarization component based on the final phase Reconstruct the intensity distribution of the tightly focused light field under full polarization. Calculate the loss value between these reconstruction results and the actual target light field, and use the back propagation and gradient descent algorithms to continuously adjust the parameters in the 3DCFH-Net so that the entire 3DCFH-Net gradually converges to the optimal state. This multi-level supervised learning framework can effectively speed up the convergence of the network.

[0073] Specifically, 3DCFH-Net uses a 3×3 convolutional block to transform the target amplitude Mapping into a high-dimensional feature space, the spatial resolution remains unchanged, allowing the network to capture more complex local and global patterns without reducing spatial details, enhancing its feature learning and generalization capabilities. The mapping result is expressed as , C0 represents I tar The number of channels, H represents Height, W represents The width of C1 The number of channels is ; the 3×3 convolution block includes a 3×3 convolution layer, a BN (batch normalization) layer, and a ReLU layer. Then, the first U-Net module is used to train the features Encode, decode, and output features , C2 represents The number of channels is , and it passes through a 3×3 convolutional layer to generate a tensor containing the target amplitude and phase of the C0 layer. Then, it is channel-separated and divided into two parts, which are activated by ReLU and Hardtanh functions respectively to obtain the X-polarized complex amplitude of all z planes (including the initial phase on the SLM). ). Connect along the channel direction and A become a new tensor, which passes through a 3×3 convolution block, another U-Net submodule and a Hardtanh activation function to obtain the final phase distribution During the convolution process, the amplitude and phase information are always inseparable, which enhances the interaction between them to a certain extent and helps to improve the calculation accuracy.

[0074] In order to speed up the training of the network and improve the calculation accuracy of the network, the cosine similarity loss function L cs And the mean square error loss function L MSE , as the overall loss function L to train 3DCFH-Net.

[0075] (4)

[0076] Where η is L cs The weight factor of L MSE The weight factor is used to balance the two loss functions in equation (4); represents the intensity distribution of the tightly focused light field under the reconstructed X-polarization component, represents the intensity distribution of the tightly focused light field under full polarization reconstructed. When calculating the loss function, the target intensity and reconstruction strength The total energy is normalized.

[0077] Among them, the cosine similarity loss function L cs What is considered is a relative difference between dimensions, which is highly related to the uniformity of the reconstructed light field. The expression is as follows:

[0078] (5)

[0079] The cosine similarity loss function is calculated by Itar and I rec The cosine value of the angle in the vector space is used to measure the similarity between them. The closer the cosine value is to 1, the more similar the two are. cs The closer to 0.

[0080] Mean square error loss function L MSE The calculation is the target focus I tar and I rec The mean square error between them, which measures an absolute numerical difference, is closely related to the efficiency of reconstructing the light field. The expression is as follows:

[0081] (6)

[0082] In the formula, represents the average operation, Ω is the set of all focal points in the target light field, represents the target strength in the set Ω, represents the reconstruction strength in the set Ω.

[0083] Loss function L cs and L MSE The overall loss of the network is evaluated from different perspectives, and the relationship between them is complementary to a certain extent. Therefore, reasonable weight values ​​η and β need to be set during the training process to improve network performance.

[0084] In this embodiment, in order to fully train the proposed 3DCFH-Net, 20,000 training samples and 2,000 test samples are randomly generated for the tightly focused light field distributions under different sizes and dimensions, wherein half of the training samples and half of the test samples have a binary (0 or 1) target intensity distribution, and the other half have a variable target intensity distribution, with the intensity values ​​randomly distributed between 0.2 and 1. In addition, in order to fully consider the impact of factors such as the effective aperture and beam diameter of the optical system on the training of 3DCFH-Net, some restrictions are implemented when constructing the training and test samples. The sample resolution is set to 640×640, and each sample tensor contains several foci. The intensity and position of the foci are randomly generated, and it is ensured that all foci are concentrated in the region of interest (hereinafter referred to as ROI) at the center of the sample, thereby effectively preventing the target focus from being lost due to edge clipping of the optical system; the position and range of the ROI area are determined by the optical aperture and focusing range of the actual optical path system device. 3DCFH-Net was trained and tested on Nvidia GeForce RTX3090 GPU, and the Adam algorithm was used as the optimizer to optimize the parameters of 3DCFH-Net. During the training process, the initial learning rate was set to 0.005, the batch size was 8, and a total of 60 epochs were trained. 5,000 samples were randomly selected again in each epoch to enhance the generalization ability of the network and prevent overfitting. A multi-stage learning rate adjustment strategy was adopted. A larger learning rate was used in the early stage to accelerate the decline of the loss function, and then the learning rate was gradually reduced to fine-tune the model parameters, so that 3DCFH-Net converged more accurately to the optimal solution. That is, in the first three epochs, the learning rate of each epoch was reduced to 0.5 times, and the learning rate of the remaining epochs was reduced to 0.9 times.

[0085] like Figure 5 As shown, the input size of the target light field of the two examples shown is 5×640×640, and the layer spacing is 10μm, where the left half is the modulation result of the binary target intensity distribution sample, and the right half is the modulation result of the variable target intensity distribution sample, and s represents the magnification of the reconstructed light field. Figure 5 The modulation effect of 3DCFH-Net on the target light field with five depth layers is demonstrated, showing high consistency, and the calculation speed of a single instance can reach 10ms; the shortest calculation time of a single-layer light field is 3.7ms, which is 2~3 orders of magnitude shorter than the traditional method, and the accuracy can reach 0.93.

[0086] like Figure 6As shown, in order to verify the applicability of 3DCFH-Net to dense focus distribution data, the DIV2K dataset was used to generate target light fields with different ROI sizes and focus intervals. The input size is 1×640×640. On the left side of the dotted line, the ROI size of the target intensity distribution is 128×128, and the spacing between adjacent focus centers is 1 pixel; on the right side of the dotted line, the ROI size is 320×320, and the spacing is 2 pixels. Figure 6 The distribution complexity of the modulation focus is increased, and the focus distribution is denser, indicating that the performance and applicability of 3DCFH-Net are better.

[0087] S3: The target intensity distribution tensor I to be modulated tar Input into the trained 3DCFH-Net to generate the corresponding holographic phase (final phase) .

[0088] S4: Phase the holographic The light is synchronously loaded onto the SLM, and then modulated by an optical system including a 4f system and a high NA objective lens to obtain a micro-nano light field, thereby achieving real-time modulation of a large-scale three-dimensional tightly focused light field. In this embodiment, the method is applied to real-time laser parallel direct writing or fluorescent dynamic display.

[0089] like Figure 7 As shown in the figure, the laser parallel direct writing device includes a femtosecond laser, a zero-order half-wave plate, an achromatic zoom beam expander, an aperture stop, an SLM, a 4f system, a beam splitter, an objective lens, a sample, a 3D nano-displacement platform and two CMOS cameras arranged from front to back along the optical axis. Among them, the femtosecond laser (Light Conversion, PHAROS-PH2, 1030 nm) is used as the light source, and its polarization direction is manipulated by rotating the zero-order half-wave plate (Thorlabs, WPH10M-1030) to improve the SLM (Hamamatsu, X15213-03B, pixel size l ps=12.5μm). An achromatic zoom beam expander (Thorlabs, ZBE3B, 2-8×) and an aperture stop were used to amplify and adjust the incident beam diameter to improve the utilization of the SLM. The output beam of the SLM was scaled and propagated from the SLM to the back pupil of the objective lens using a 4f system, which included two plano-convex lenses, where the convex surface of lens 1 faced forward and had a focal length of f1=750mm, and the convex surface of lens 2 faced backward and had a focal length of f2=500mm. In the device, a reverse pinhole was placed between the two plano-convex lenses of the 4f system to mitigate the influence of zero-order light, which was achieved by adding a very small light block in the molten silicon window. The laser was divided into two beams by a beam splitter, one of which was input into the objective lens. The objective lens (Nikon, 50×, 0.8NA) focused the laser into the sample, which was placed on a 3D nano-displacement platform, which could precisely control the position of laser direct writing. The other laser beam passes through lens 3 and enters camera 2. Lens 3 is a plano-convex lens with the convex surface facing forward. Camera 2 is mounted on an axial displacement stage to obtain the intensity distribution of the focal plane at different depths. Camera 1 is used to observe the sample and the laser direct writing process in real time.

[0090] The sample used in this example is Sm 3+ Doped sodium aluminate glass (85B2O3-5Al2O3-10Na2O-0.1Sm2O3) was prepared by conventional melt-quenching technology, in which the powder reagents were mixed together and melted at 1250°C for 30 minutes in an atmospheric environment; the molten glass was then cast on a stainless steel plate to form a glass block with a thickness of about 1 mm, which was precision polished for laser direct writing.

[0091] The method of the present invention is used to carry out three-dimensional laser parallel direct writing, which greatly improves its speed and accuracy.

[0092] like Figure 8 As shown in the figure, the effect of processing a hologram in samarium-doped sodium aluminate glass (i.e., laser parallel direct writing) by using 3DCFH-Net in conjunction with a laser parallel direct writing device is demonstrated. Each row in the figure represents an example. It can be seen that the experimental results are well consistent with the simulation results. Using femtosecond lasers, multiple micro-nano structures can be induced simultaneously, and the laser focusing processing depth is about 30μm.

[0093] like Fig. 9 As shown in the figure, in order to more effectively observe the focal field modulation capability of 3DCFH-Net, a CMOS camera was used to record the intensity distribution of the focal plane at different axial distances (i.e., light field depth). Considering the inherent limitations of traditional discrete Fourier transform in terms of resolution and fidelity, a chirp Z-transform (CZT) was used to achieve lossless amplification of the reconstructed tight focus field. Fig. 9As can be seen from (a) and (b), no matter how the target focus is arranged, 3DCFH-Net always shows excellent ability to accurately reconstruct the focal field. In addition, Fig. 9 (c) shows the evolution of the holographic focal field generated by 3DCFH-Net in binary mode based on time series, and (d) shows the evolution of the holographic focal field generated by 3DCFH-Net in variable intensity distribution mode based on time series. The experimental results strongly prove that the proposed 3DCFH-Net can generate various complex light field structures in real time and accurately, which is of great significance for improving the speed and accuracy of three-dimensional parallel direct writing technology. The method of the present invention shows considerable application potential in various fields including binary / multi-level optical data storage, micro / nanostructure manufacturing and waveguides.

[0094] like Fig.10 As shown, the fluorescence dynamic display device includes a continuous laser, a fiber collimator, a polarizer, a half-wave plate, an aperture stop, a right-angle reflection prism, an SLM, a 4f system, a long-wave dichroic mirror, an objective lens, a sample, a plane mirror, a 3D translation stage and a camera, which are arranged in sequence from front to back along the optical axis. The continuous ultraviolet laser (wavelength of 405nm) emitted by the continuous laser is used as the excitation beam. The excitation beam is collimated by a fiber collimator (Thorlabs, F810APC-405), and then passes through a polarizer and a half-wave plate in sequence. The direction of the light path is adjusted by two plane mirrors, so that the laser is reflected by the right-angle reflection prism and then enters the SLM (Holoeye, PLUTO-2.1 UV-099, pixel size l) at a small angle. ps =8μm). The laser carries the phase information loaded in the SLM. After adjusting the direction of the optical path through two plane mirrors, the 4f system is used to scale the output beam from the SLM and propagate it to the rear pupil plane of the objective lens. The 4f system includes two plano-convex lenses with opposite planes. The focal length of the front lens is f1=500mm, and the focal length of the rear lens is f2=250mm. A reverse pinhole is placed between the two plano-convex lenses of the 4f system to reduce the influence of zero-order light, which is achieved by adding a very small light block in the molten silicon window. The light beam output by the 4f system is divided into two beams by a long-wave dichroic mirror. One beam is input into the objective lens. The objective lens (Olympus, 50×, 0.8 NA) focuses the laser onto the sample surface. The sample is placed on a 3D translation stage, and the position of the dynamic display of fluorescence can be precisely controlled. The other laser beam passes through the filter and lens and enters the camera. The lens is a plano-convex lens with the convex surface facing forward. The camera is used to observe the sample and the dynamic display of fluorescence in real time.

[0095] The sample used in this embodiment is perovskite nanocrystals (hereinafter referred to as PNCs) glass, and the preparation method is as follows: the powder reagent (27SiO2-42B2O3-5CaO-10ZnO-5Na2O-2Cs2O-1PbBr2-4NaBr-4NaI) is enclosed in a corundum crucible and placed in a muffle furnace in an atmospheric environment for sintering at 1000°C for 20 minutes. The hot molten liquid is then poured onto a preheated stainless steel plate for quenching, and pressed with a brass plate to form a precursor glass sample. The quenched precursor glass is then annealed at 300°C in a muffle furnace for 1 hour to release residual thermal stress. These glass samples undergo different heat treatments (temperature and time) to adjust the PL wavelength and are precision polished for holographic fluorescence display.

[0096] The method of the present invention is used to perform dynamic fluorescence display, thereby realizing fluorescence excitation with arbitrary multiple focal points to replace the traditional single-point excitation mode and improving work efficiency.

[0097] like Fig.11 As shown in (a), the 520nm photoluminescence (PL) image of the QR code focus array of the Zhijiang Laboratory homepage modulated by 3DCFH-Net is displayed on the perovskite nanocrystal glass, and the fluorescence can be successfully identified, proving the accuracy of holographic fluorescence display. Fig.11 As shown in (b), we also achieved colorful dynamic display in PNCs (perovskite nanocrystals) glass with tunable composition, demonstrating the feasibility of holographic fluorescent dynamic display.

[0098] The 3DCFH-Net constructed by the method of the present invention integrates vector diffraction theory and convolutional neural network, fully considers the polarization characteristics and spherical aberration effect of the tightly focused electric field under a high numerical aperture objective, and accurately describes the electric field distribution focused on different z planes under two media; the global constraint of reconstructing the three-dimensional focused field is strengthened by adopting a layer-based learning strategy; according to the forward diffraction propagation model, the overall loss function between the reconstructed light field and the target light field is directly calculated, so that the network works in an unsupervised manner. 3DCFH-Net consists of two cascaded U-Net modules, the first U-Net module focuses on the initial target complex amplitude prediction on the main horizontal polarization component, and the second U-Net module further optimizes the holographic phase under the vector tight focusing mechanism; through this framework, the modulation of the 3D tightly focused vector light field can be simply and efficiently realized. 3DCFH-Net is embedded with a batch-based depth generator (i.e., a deep adaptive module), which can dynamically and randomly assign different layer spacings and focal depths to all training samples and test samples (for the calculation of the reverse diffraction propagation model and the forward diffraction propagation model), thereby describing the convergence and divergence of a changing tightly focused electric field, enhancing the flexibility and versatility of its modulated focal field, and increasing the data diversity and generalization performance of the network. By creating a large number of light field intensity distribution samples with regions of interest, the network parameters are fully trained. After training, 3DCFH-Net can quickly generate batches of holograms, realize rapid modulation of complex three-dimensional light fields, and realize large-scale generation of high-speed, high-precision holographic phases. Its deployment in laser direct writing and holographic fluorescence display applications has verified the practicality, versatility, and potential influence of 3DCFH-Net on future technological development.

[0099] On the basis of the above-mentioned micro-nano light field real-time control method based on three-dimensional tight focusing holographic network, the present invention also provides a micro-nano light field real-time control device based on three-dimensional tight focusing holographic network, including: a training data set acquisition module, a three-dimensional tight focusing holographic network training module, a target light field phase calculation module, a phase loading module, and a three-dimensional tight focusing light field modulation system. The three-dimensional tight focusing light field modulation system is an optical system including a 4f system and a high NA objective lens.

[0100] The training data set acquisition module is used to obtain the corresponding target intensity distribution tensor according to the randomly generated target light field intensity distribution, and integrate it to obtain the training data set. The three-dimensional tightly focused holographic network training module is used to optimize the parameters of 3DCFH-Net through the gradient descent algorithm. The target light field phase calculation module is used to input the target light field into the trained 3DCFH-Net. The final holographic phase is loaded into the SLM through the phase loading module, and then modulated by the three-dimensional tightly focused light field modulation system to obtain the micro-nano light field.

[0101] like Fig.12As shown, an electronic device provided by an embodiment of the present invention includes a memory and one or more processors, wherein executable codes are stored in the memory, and when the one or more processors execute the executable codes, they are used to implement the real-time control method of micro-nano light field based on three-dimensional tightly focused holographic network in the above embodiment.

[0102] An electronic device of the present invention can be applied to any device with data processing capability, and the device with data processing capability can be a device or apparatus such as a computer. The electronic device can be implemented by software, or by hardware, or by a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, the processor of any device with data processing capability reads the corresponding computer program instructions in the non-volatile memory into the memory and runs them. From the hardware level, if Fig.12 As shown, it is a hardware structure diagram of an electronic device of the present invention, in which any device with data processing capability is located, except Fig.12 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the electronic device of the present invention is located in the embodiment may also include other hardware, which will not be described in detail.

[0103] The implementation process of the functions and effects of each unit in the above electronic device is specifically described in the implementation process of the corresponding steps in the above method, which will not be repeated here.

[0104] For the electronic device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The electronic device embodiment described above is only schematic, and the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0105] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the method for real-time control of a micro-nano light field based on a three-dimensional tightly focused holographic network in the above embodiment is implemented.

[0106] The computer-readable storage medium may be an internal storage unit of any device with data processing capability in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capability, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), an SD card, a flash card (Flash Card), etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capability. The computer-readable storage medium is used to store computer programs and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0107] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0108] The above description is only the preferred embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and variations, and the various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A real-time control method for micro-nano light field based on three-dimensional tight focusing holographic network, characterized in that: The following steps are involved: S1: According to the randomly generated three-dimensional tightly focused target light field intensity distribution, the corresponding target intensity distribution tensor I is obtained tar , the training data set is obtained; S2: Construct a three-dimensional tightly focused holographic network 3DCFH-Net, and input the training data set into the network for training; the 3DCFH-Net uses two cascaded U-Net modules, and the training process is: Will I tar Input the first U-Net module to establish the mapping relationship between the noise-free target light field intensity and its optimal complex amplitude in the X polarization direction, and obtain the initial amplitude A′ of each target z plane 2x and the initial phase φ′ 2x ; Then, the complex amplitude distribution of the rear pupil plane is derived through the optical reverse diffraction propagation model, and the initial phase φ′0 on the SLM is obtained by superposition; φ′0 and the incident laser amplitude A of the SLM are input into the second U-Net module, and the final phase φ0 is output; According to the optical forward diffraction propagation model, the intensity distribution of the tightly focused light field under the X-polarization component is reconstructed based on φ′0, and the intensity distribution of the tightly focused light field under full polarization is reconstructed based on φ0; The loss value between the reconstruction result and the actual target light field is calculated, and the parameters in the 3DCFH-Net are adjusted by using the back propagation and gradient descent algorithms, so that the 3DCFH-Net converges to the optimal state, and the trained 3DCFH-Net is obtained; S3: I tar Input into the trained 3DCFH-Net to generate the corresponding final phase φ0; S4: φ0 is synchronously loaded onto the SLM, and then the micro-nano light field is modulated by the optical system including the 4f system and the objective lens, thereby realizing the real-time modulation of large-scale three-dimensional tightly focused light field.

2. The real-time control method of micro-nano light field based on three-dimensional tight focusing holographic network according to claim 1 is characterized in that: In S2, the expression of the forward diffraction propagation model is as follows: H ψ =exp[ikd(n2cosθ2-n1cosθ1)]; H z =exp[ikn2cosθ2z]; In the formula, E 0x represents the distribution of X-polarization of the electric field E0 at the entrance pupil plane; H ψ represents the defocus aberration caused by refractive index mismatch, H z represents the phase accumulation along the z axis; k represents the wave number, d represents the detection depth of the objective lens, n1 represents the refractive index of the objective lens, n2 represents the refractive index of the sample, θ1 represents the convergence angle of the objective lens, and θ2 represents the convergence angle of the sample; p x The coefficient of the X-polarization component corresponding to E0, p y The coefficient of the Y polarization component corresponding to E0, p z The coefficient of the Z polarization component corresponding to E0; m 11 、m 21 、m 31 The transformation matrix M = [m wt ] 3×3 The elements in M ​​are related to the electric field at the entrance pupil plane and the vector electric field refraction at the interface, w is the row index of the matrix element, and t is the column index; The expression of the reverse diffraction propagation model is as follows: In the formula, Indicates p x The complex conjugate of 2x represents the distribution of X polarization of the target light field, E 2y represents the distribution of the Y polarization of the target light field, E 2z Represents the distribution of Z polarization of the target light field.

3. The real-time control method of micro-nano light field based on three-dimensional tight focusing holographic network according to claim 1 is characterized in that: In S2, 3DCFH-Net uses a 3×3 convolution block to transform the target amplitude Mapping into a high-dimensional feature space, the spatial resolution remains unchanged, allowing the network to capture more complex local and global patterns without reducing spatial details, enhancing its feature learning and generalization capabilities; the mapping result is expressed as C0 means I tar The number of channels, H represents F1 in Height, W means F1 in Width, C1 means F1 in The number of channels; the 3×3 convolution block includes a 3×3 convolution layer, a BN layer and a ReLU layer; then, the first U-Net module is used to train the feature F1 in Encode, decode, and output features C2 means F1 out The number of channels is , and it passes through a 3×3 convolutional layer to generate a tensor containing the target amplitude and phase of the C0 layer; then, it is channel-separated and divided into two parts, which are activated with ReLU and Hardtanh functions respectively to obtain the X-polarized complex amplitude of all z planes, including the initial phase φ′0 on the SLM; φ′0 and A are connected along the channel direction to become a new tensor, which passes through a 3×3 convolutional block, another U-Net submodule and the Hardtanh activation function to obtain the final phase distribution φ0.

4. The real-time control method of micro-nano light field based on three-dimensional tight focusing holographic network according to claim 1 is characterized in that: In S2, the loss function L for training 3DCFH-Net is expressed as follows: Where, L cs represents the cosine similarity loss function, L MSE represents the mean square error loss function; η is L cs The weight factor of L MSE The weight factor of represents the intensity distribution of the tightly focused light field under the reconstructed X-polarization component, represents the intensity distribution of the tightly focused light field under full polarization reconstructed; <·> represents the averaging operation, Ω is the set of all focal points in the target light field, represents the target strength in the set Ω, represents the reconstruction strength in the set Ω.

5. The real-time control method of micro-nano light field based on three-dimensional tight focusing holographic network according to claim 1 is characterized in that: In the S4, the optical system including the 4f system and the objective lens is a laser parallel direct writing device, which includes: a 3D nano-displacement platform and two cameras, and a femtosecond laser, a half-wave plate, a beam expander, an aperture stop, an SLM, a 4f system, a beam splitter, an objective lens, and a sample arranged in sequence along the optical axis; the femtosecond laser is a light source, the half-wave plate is used to manipulate the polarization direction of the laser, and the beam expander and the aperture stop are used to amplify and adjust the diameter of the incident light beam; the 4f system is used to scale the output light beam from the SLM and propagate it to the back pupil of the objective lens; the beam splitter is used to divide the incident light beam into three paths, one of which is input into the objective lens, and the objective lens is used to focus the light beam onto the sample; another path is input into camera one, and camera one is used to observe the sample and the laser direct writing process in real time; the third path is input into camera two, and camera two is used to obtain the intensity distribution of the focal plane at different depths.

6. The real-time control method of micro-nano light field based on three-dimensional tight focusing holographic network according to claim 1 is characterized in that: In the S4, the optical system including the 4f system and the objective lens is a fluorescence dynamic display device, which includes: a 3D translation stage and a camera, as well as a continuous laser, a fiber collimator, a polarizer, a half-wave plate, an aperture stop, a right-angle reflection prism, an SLM, a 4f system, a long-wave pass dichroic mirror, an objective lens, and a sample arranged in sequence along the optical axis; the excitation light beam emitted by the continuous laser is collimated by the fiber collimator, and then passes through the polarizer and the half-wave plate in sequence, and the right-angle reflection prism is used to make the light beam enter the SLM at an angle less than 10°; the 4f system is used to scale the output light beam from the SLM and propagate it to the rear pupil surface of the objective lens; the long-wave pass dichroic mirror is used to divide the incident light beam into two paths, one path is input to the objective lens, and the objective lens is used to focus the light beam on the sample; the other path is input to the camera, and the camera is used to observe the sample and the process of fluorescence dynamic display in real time.

7. The real-time control method of micro-nano light field based on three-dimensional tight focusing holographic network according to claim 1 is characterized in that: Each sample tensor in the training data set of S1 contains several focuses, the intensity and position of the focuses are randomly generated, and the focuses are distributed in the region of interest of the entire tensor. The position and range of the region of interest are determined by the optical aperture and focusing range of the actual optical path system device.

8. A device for real-time control of micro-nano light fields based on a three-dimensional tightly focused holographic network, used to implement the method for real-time control of micro-nano light fields based on a three-dimensional tightly focused holographic network as described in any one of claims 1 to 7, characterized in that: include: Training data set acquisition module, three-dimensional tightly focused holographic network training module, target light field phase calculation module, phase loading module, three-dimensional tightly focused light field modulation system; The three-dimensional tightly focused light field modulation system is an optical system including a 4f system and an objective lens; The training data set acquisition module is used to obtain the corresponding target intensity distribution tensor according to the randomly generated target light field intensity distribution, and integrate it to obtain the training data set; the three-dimensional tightly focused holographic network training module is used to optimize the parameters of 3DCFH-Net through back propagation and gradient descent algorithms; the target light field phase calculation module is used to input the target light field into the trained 3DCFH-Net, and the final holographic phase is loaded into the SLM through the phase loading module, and then modulated by the three-dimensional tightly focused light field modulation system to obtain the micro-nano light field.

9. An electronic device, characterized in that: It comprises a memory and one or more processors, wherein the memory stores executable codes, and when the one or more processors execute the executable codes, they are used to implement the real-time control method of micro-nano light field based on three-dimensional tightly focused holographic network as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, the real-time control method of micro-nano light field based on a three-dimensional tightly focused holographic network as described in any one of claims 1 to 7 is implemented.

Images