Monolithic full-visible light imaging diffractive lens and design method

Abstract

The application belongs to the field of optics, in order to solve the problems of high demand for computer resources in the ADL optimization design process, low energy utilization rate of the optimization result, mapping eccentricity in the design and the like, a monolithic full-visible light imaging diffractive lens and a design method are provided. By setting an input plane wave matrix, a plane wave neural network imaging model passing through the ADL is constructed, a loss function is used as an optimization standard of the ADL surface microstructure, and a monolithic full-visible light imaging diffractive lens is designed by taking the PSF and the OTF as imaging constraint conditions of the ADL. The lens can image full-visible light, and the algorithm complexity of the current ADL design is reduced, and the defects existing in the design are solved.

Description

Technical Field

[0001] This invention relates to the field of diffractive optical element imaging, specifically to a design method for a monolithic all-visible light imaging diffractive lens based on a gradient backpropagation mechanism. Background Technology

[0002] Diffractive optical elements (DOEs) are devices created through computer-aided design and microfabrication processes. These processes involve etching stepped or even continuous relief structures onto the surface of a substrate or traditional optical device to achieve phase modulation and wavefront transformation. DOEs are characterized by their small size, light weight, high diffraction efficiency, and high degree of design freedom.

[0003] Traditional DOE design requires knowledge of the incident light field and the light field distribution on the output plane of the optical system to calculate the phase distribution of the modulation element on the output plane, ensuring it correctly modulates the incident light field and achieves the desired function. DOE design based on scalar diffraction theory can be viewed as an optimization design problem. Currently, optimization algorithms based on this approach mainly include the Gerchberg-Saxton Algorithm (GS), Simulated Annealing Algorithm (SA), Genetic Algorithm (GA), Yang-Gu Algorithm (YG), and various hybrid algorithms.

[0004] With the proposal of an end-to-end optimization method by Sitzmann Vincent et al. of Stanford University, which combines the microstructure distribution of diffractive elements with the parameters of image restoration algorithms, an optical imaging model of a monolithic achromatic diffractive lens (ADL) was established. The neural network back propagation (BP) mechanism and hardware acceleration scheme were introduced into the ADL design. However, problems such as low optimization efficiency, computational complexity requiring high computing power, and information loss caused by ignoring the cutoff of the modulation transfer function have emerged. Summary of the Invention

[0005] This invention proposes a neural network-based all-visible light diffractive lens and its optimization design method. By establishing an efficient optimization design model and a reasonable loss function, the demand for computer resources in the ADL optimization design process is reduced, and the optimization design efficiency is improved.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] A monolithic all-visible light imaging diffractive lens and its design method are disclosed, including a design algorithm for ADL microstructure distribution optimization based on neural networks and an ADL imaging restoration algorithm.

[0008] In this invention, the design algorithm for optimizing the microstructure distribution of an ADL (Advanced Dynamic Linear Array) based on a neural network is derived by deriving the imaging process of a plane wave through an ADL, obtaining the optical functions PSF and OTF of the ADL as constraints, and establishing a loss function to optimize the surface microstructure of the ADL. Then, an ADL imaging restoration algorithm is used to simulate and output the results. In the optimization process of the ADL microstructure distribution based on a neural network, the two-dimensional ADL microstructure distribution is introduced into a concentric ring model to decompose a one-dimensional rotationally symmetric microstructure distribution model. Based on the one-dimensional rotationally symmetric microstructure distribution model of the ADL, the PSF and OTF at different wavelengths are derived as imaging evaluation results.

[0009] In this invention, there are two types of loss functions: the first is the loss function between the ideal image and the ADL simulation imaging; the second is the loss function generated by constrained ADL imaging. The loss function generated by constrained ADL imaging is further divided into: the loss function obtained by constraining the PSF at different wavelengths due to energy inhomogeneity at different wavelengths; the loss function of the frequency cutoff in the transfer function that suppresses the existence of MTF; and the loss function generated by constraining the continuity of the microstructure distribution in ADL. When using the loss function generated by constrained ADL imaging, this loss function is directly evaluated without performing simulation imaging to calculate the loss function between the ideal image and the ADL simulation imaging.

[0010] In this invention, after the design algorithm for ADL microstructure distribution optimization based on neural networks is constructed, the wavelength of the input plane wave can be arbitrarily selected.

[0011] This invention discloses the application of the above-mentioned monolithic full-visible light imaging diffraction lens in its fabrication or as an optical element.

[0012] In this invention, the design algorithm for ADL microstructure distribution optimization based on neural networks is as follows:

[0013] Define the variable matrix D(m,n) of the microstructure of the ADL surface to be optimized, and define the input plane wave matrix. After the plane wave passes through the ADL surface, the phase delay generated by the wavefront complex amplitude matrix E(m,n,c) is U0(m,n,c). Here, m, n, and c represent the row sampling index m, column sampling index n, and wavelength sampling index c of the incident wavefront matrix.

[0014] Based on the functional expression of scalar diffraction theory, the matrix expression for the diffraction propagation of the wavefront complex amplitude U0(m,n,c) in air is derived, and the Fresnel diffraction formula is written in convolution form: U(x,y)=U0(x,y)*h(x,y); h(x,y) is subjected to Fourier transform and frequency domain transformation to obtain the matrix transfer function matrix H(m,n,c); and the point spread function PSF(m,n,c) received by the image sensor is obtained.

[0015] The wavefront complex amplitude matrix E(m,n,c) of the front surface of the ADL, the microstructure distribution D(m,n) of the ADL, the microstructure error distribution t(m,n), the refractive index n(c) corresponding to the wavelength λ(c) with sampling index c, the diffraction propagation distance z, and the wavefront sampling intervals δx and δy are used to determine the point spread function PSF(m,n,c) on the image plane.

[0016] The Cartesian coordinate system in D(m,n) is converted to the polar coordinate system, and a one-dimensional rotationally symmetric microstructure distribution model is introduced. That is, the two-dimensional distribution matrix D(m,n) is converted to D(r), and PSF(ρ,θ) is obtained after concentric circle decomposition.

[0017] From the complex amplitude distribution matrix E(r) of different wavefronts reaching the front surface of the ADL, the microstructure distribution D(r) of the ADL, the microstructure error distribution t(r), the refractive index n(c) corresponding to the sampling wavelength λ(c), the diffraction propagation distance z, and the concentric ring sampling interval d, the radial distribution of the one-dimensional rotationally symmetric ADL point spread function PSF(r,c) on the image plane is derived. Based on PSF(x,y,c) and OTF(f x ,f y Given the relationship between ρ and c, we obtain OTF(ρ,c):

[0018] .

[0019] The optical transfer function OTF(ρ,c) is calculated based on the above formula to evaluate the mapping performance of ADL. Then, the microstructure distribution D(r) of ADL is optimized based on the evaluation results.

[0020] Preferably, the optical functions PSF (point spread function) and OTF (optical transfer function) are used as constraints to optimize the one-dimensional array D(r) of the radial microstructure distribution of ADL.

[0021] Preferably, the radial distribution functions OTF(ρ,c) and MTF(ρ,c) of OTF and MTF (modulation transfer function) are calculated by the radial distribution of the one-dimensional rotationally symmetric point spread function PSF(r,c).

[0022] Preferably, PSF(r,c) and OTF(ρ,c) are combined, and a differentiable loss function is used to calculate the loss value to evaluate the performance of ADL. The gradient is then backpropagated to optimize the one-dimensional array D(r) of the radial microstructure distribution of ADL.

[0023] Depending on the choice of loss function, methods are divided into Method 1 and Method 2. Method 1's loss function is the loss function obtained by directly simulating the image after plane wave imaging via ADL. Method 2's loss function includes: L1 (due to energy inhomogeneity at different wavelengths), L2 (the frequency cutoff in the transfer function that suppresses MTF by constraining PSF(r,c) at different wavelengths), and L3 (resulting from constraining the continuity of the microstructure distribution in ADL). The total loss function of the front-end optical system can be written as:

[0024] .

[0025] k is a hyperparameter representing the concentration constraint weight of PSF(r,c); mean(·) represents the mean operation, variance(·) represents the variance operation, R indicates that the outermost ring of PSF(r,c) is the R-th ring, R(c), G(c), and B(c) represent the responses of the camera sensor to different wavelengths λ(c) at different color channels (red, green, and blue), respectively, and r1 represents the variance weight. zpt represents the index of the first frequency cutoff position of MTF. D(r) represents the radial microstructure distribution of ADL.

[0026] The ADL imaging restoration algorithm is as follows: PODMPHN is an image deblurring algorithm based on DMPHN of segmented images, which adopts an adaptive weighted superposition method to improve its applicability. Beneficial effects

[0027] Compared with existing technologies, the design method of the monolithic full-visible light imaging diffraction lens proposed in this application reduces the demand for computer resources in the ADL optimization design process. The single-wavelength, single-step, and single-run time is basically unchanged with the ADL design aperture, thus improving the optimization design efficiency. It also solves the problems that inevitably occur in the design process, such as mapping eccentricity, energy non-uniformity, mid-frequency cutoff of transfer function, and discrete microstructure distribution. Attached Figure Description

[0028] Figure 1 shows the ADL imaging model of a single-piece full visible light imaging diffraction lens.

[0029] Figure 2 is a flowchart of the ADL optimization design process.

[0030] Figure 3 shows the surface microstructure distribution contour of ADL.

[0031] Figure 4 shows an ADL sample.

[0032] Figure 5 shows the complete shape and radial cutoff of the PSF of ADL at different wavelengths. Detailed Implementation

[0033] This invention discloses a design method for a monolithic all-visible-light imaging diffractive lens, including a design algorithm for ADL microstructure distribution optimization based on neural networks and an ADL imaging restoration algorithm. The design algorithm for ADL microstructure distribution optimization optimizes the surface microstructure of the ADL by deriving the imaging process of a plane wave passing through the ADL, obtaining the ADL's optical functions PSF and OTF as constraints, and establishing a loss function. Once the design requirements are met, the ADL imaging restoration algorithm is used to simulate and output the results. During the optimization process, the original two-dimensional ADL microstructure distribution is introduced into a concentric ring model to decompose a one-dimensional rotationally symmetric microstructure distribution model. Based on the one-dimensional rotationally symmetric microstructure distribution model of the ADL, the PSF and OTF at different wavelengths are derived as imaging evaluation results.

[0034] Figure 1 illustrates the basic process of this method.

[0035] There are two types of loss functions for optimizing ADL (Advanced Dynamics Image): the first is the loss function between the ideal image and the ADL simulation imaging; the second is the loss function generated by constraining ADL imaging. Due to energy unevenness at different wavelengths, there are loss functions obtained by constraining the PSF (Power Seepage Function) at different wavelengths, the loss function of the frequency cutoff in the transfer function that suppresses the MTF (Mean Transmission Function), and the loss function generated by constraining the continuity of the microstructure distribution in the ADL. When using the loss function generated by constrained ADL imaging for optimization, we directly judge whether this loss function meets the requirements without performing ADL simulation imaging to calculate the loss function between the ideal image and the ADL simulation imaging. After optimization is complete, we then perform ADL simulation imaging to output the results.

[0036] Figure 2 illustrates two different implementation approaches for this method.

[0037] In this invention, after the design algorithm for ADL microstructure distribution optimization based on neural networks is constructed, the input plane wave wavelength can be arbitrarily selected. A single wavelength or a wavelength range can be chosen.

[0038] In this invention, the design algorithm for ADL microstructure distribution optimization based on neural networks is as follows:

[0039] Define the variable matrix D(m,n) of the microstructure of the ADL surface to be optimized, and define the input plane wave matrix. After the plane wave passes through the ADL surface, the phase delay generated by the wavefront complex amplitude matrix E(m,n,c) is U0(m,n,c). Here, m, n, and c represent the row sampling index m, column sampling index n, and wavelength sampling index c of the incident wavefront matrix.

[0040] Based on the functional expression of scalar diffraction theory, the matrix expression for the diffraction propagation of the wavefront complex amplitude U0(m,n,c) in air is derived, and the Fresnel diffraction formula is written in convolution form: U(x,y)=U0(x,y)*h(x,y); h(x,y) is subjected to Fourier transform and frequency domain transformation to obtain the matrix transfer function matrix H(m,n,c); and the point spread function PSF(m,n,c) received by the image sensor is obtained.

[0041] The wavefront complex amplitude matrix E(m,n,c) of the front surface of the ADL, the microstructure distribution D(m,n) of the ADL, the microstructure error distribution t(m,n), the refractive index n(c) corresponding to the wavelength λ(c) with sampling index c, the diffraction propagation distance z, and the wavefront sampling intervals δx and δy are used to determine the point spread function PSF(m,n,c) on the image plane.

[0042] The Cartesian coordinate system in D(m,n) is converted to the polar coordinate system, and a one-dimensional rotationally symmetric microstructure distribution model is introduced. That is, the two-dimensional distribution matrix D(m,n) is converted to D(r), and PSF(ρ,θ) is obtained after concentric circle decomposition.

[0043] From the complex amplitude distribution matrix E(r) of different wavefronts reaching the front surface of the ADL, the microstructure distribution D(r) of the ADL, the microstructure error distribution t(r), the refractive index n(c) corresponding to the sampling wavelength λ(c), the diffraction propagation distance z, and the concentric ring sampling interval d, the radial distribution of the one-dimensional rotationally symmetric ADL point spread function PSF(r,c) on the image plane is derived. Based on PSF(x,y,c) and OTF(f x ,f y Given the relationship between ρ and c, we obtain OTF(ρ,c):

[0044]

[0045] The optical transfer function OTF(ρ,c) is calculated based on the above formula to evaluate the mapping performance of ADL. Then, the microstructure distribution D(r) of ADL is optimized based on the evaluation results.

[0046] Preferably, the optical functions PSF (point spread function) and OTF (optical transfer function) are used as constraints to optimize the one-dimensional array D(r) of the radial microstructure distribution of ADL.

[0047] Preferably, the radial distribution functions OTF(ρ,c) and MTF(ρ,c) of OTF and MTF (modulation transfer function) are calculated by the radial distribution of the one-dimensional rotationally symmetric point spread function PSF(r,c).

[0048] Preferably, PSF(r,c) and OTF(ρ,c) are combined, and a differentiable loss function is used to calculate the loss value to evaluate the performance of ADL. The gradient is then backpropagated to optimize the one-dimensional array D(r) of the radial microstructure distribution of ADL.

[0049] Depending on the choice of loss function, methods are divided into Method 1 and Method 2. Method 1's loss function is the loss function obtained by directly simulating the image after plane wave imaging via ADL. Method 2's loss function includes: L1 (due to energy inhomogeneity at different wavelengths), L2 (the frequency cutoff in the transfer function that suppresses MTF by constraining PSF(r,c) at different wavelengths), and L3 (resulting from constraining the continuity of the microstructure distribution in ADL). The total loss function of the front-end optical system can be written as:

[0050]

[0051] k is a hyperparameter representing the concentration constraint weight of PSF(r,c); mean(·) represents the mean operation, variance(·) represents the variance operation, R indicates that the outermost ring of PSF(r,c) is the R-th ring, R(c), G(c), and B(c) represent the responses of the camera sensor to different wavelengths λ(c) at different color channels (red, green, and blue), respectively, and r1 represents the variance weight. zpt represents the index of the first frequency cutoff position of MTF. D(r) represents the radial microstructure distribution of ADL.

[0052] The ADL image restoration algorithm is as follows: There are many methods for image restoration, which can be classified into blind restoration and non-blind restoration methods based on whether the point spread function is known. Due to the diversity of algorithms, this paper is not limited to the application of a single method. This invention uses the PODMPHN algorithm, which is an image deblurring algorithm based on DMPHN of the segmented image, and adopts an adaptive weighted superposition method to improve its applicability.

[0053] This invention provides a design method for a monolithic all-visible light imaging diffraction lens. To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention. The specific algorithms, calculations, preparation operations, raw materials, and testing involved in this invention are all conventional techniques.

[0054] The design process of a single-piece all-visible light imaging diffraction lens is as follows:

[0055] Step 1: Confirm the input of the design component. The component aperture is 10mm, the focal length is 100mm, and the design wavelength is 400-700nm.

[0056] Step 2: Determine the polychromatic light plane wave matrix based on the design wavelength, and at the same time randomly generate the microstructure distribution matrix of the ADL.

[0057] Step 3: Obtain the complex plane wave matrix of the polychromatic light according to scalar diffraction theory, pass it through the complex amplitude matrix of ADL, and add noise to obtain... U 0 (m,n,c) The expression for the complex amplitude of the wavefront propagating through the air to the image plane. U(x,y) The PSF function is obtained by taking the modulus. PSF(m,n,c) Furthermore, the MTF and PTF are obtained by performing a Fourier transform on the PSF function.

[0058] Step 4: If the loss function of Method 1 is used, simulate imaging of the ADL obtained in the above steps and calculate the loss function between the restored image and the simulated ideal image. If the loss function meets the design requirements or no longer decreases, output the microstructure distribution matrix of the ADL and the imaging results. If the loss function of Method 1 does not meet the requirements, evaluate the performance of the ADL using the result of the loss function, backpropagate the gradient, and optimize the one-dimensional array of radial microstructure distribution of the ADL. D(r) The optimized ADL microstructure distribution matrix is ​​then added to the noise matrix, and step 3 is repeated for further evaluation until the requirements are met.

[0059] If the loss function of Method 2 is used, i.e., the loss function of the front-end optical system is directly calculated, it is determined whether the loss function meets the requirements. If it meets the design requirements, simulation imaging is performed on the ADL that meets the requirements, and the microstructure distribution matrix of the ADL and the imaging results are output. If the loss function of Method 2 does not meet the requirements, the performance of the ADL is evaluated using the result of the loss function, and the gradient is backpropagated to optimize the one-dimensional array of radial microstructure distribution of the ADL. D(r) The optimized ADL microstructure distribution matrix is ​​then added to the noise matrix, and step 3 is repeated for further evaluation until the requirements are met.

[0060] The loss function in a neural network is between 0 and 1. Generally speaking, the smaller the loss function, the better the optimization effect, but it cannot be reduced to 0. Since training requires a lot of time, the loss function is generally less than 0.4.

[0061] Example 1

[0062] The design process of a single-piece all-visible light imaging diffraction lens is as follows:

[0063] Step 1: Confirm the input of the design component. The component aperture is 10mm, the focal length is 100mm, the design wavelength is 400-700nm, and the sampling interval is 25.

[0064] Step 2: Determine the polychromatic light plane wave matrix based on the design wavelength, and at the same time randomly generate the microstructure distribution matrix of the ADL.

[0065] Step 3: Obtain the complex plane wave matrix of the polychromatic light according to scalar diffraction theory, pass it through the complex amplitude matrix of ADL, and add noise to obtain... U 0 (m,n,c) The expression for the complex amplitude of the wavefront propagating through the air to the image plane. U(x,y) The PSF function is obtained by taking the modulus. PSF(m,n,c) Furthermore, the MTF and PTF are obtained by performing a Fourier transform on the PSF function.

[0066] Step 4: Using the loss function of Method 1, simulate imaging of the ADL obtained in the previous steps, and calculate the loss function between the restored image and the simulated ideal image. If the loss function meets the design requirements (loss function less than 0.4), output the microstructure distribution matrix of the ADL and the imaging results. If the loss function of Method 1 does not meet the requirements, evaluate the performance of the ADL using the result of the loss function, and backpropagate the gradient to optimize the one-dimensional array of radial microstructure distribution of the ADL. D(r) The optimized ADL microstructure distribution matrix is ​​then added to the noise matrix, and step 3 is repeated for further evaluation until the requirements are met.

[0067] Once the requirements are met, the surface microstructure distribution profile of ADL can be exported, such as... Figure 3 .

[0068] Taking into account factors such as the designed ADL aperture, surface microstructure height, and minimum feature size, laser direct writing was adopted as the fabrication method. The specific fabrication process is as follows:

[0069] Will as Figure 3 The ADL surface microstructure distribution profile map is inverted to grayscale to generate an 8-bit grayscale exposure characterization map for processing, which is then saved as a ".bmp" format file.

[0070] Clean and dry the quartz substrate, apply an adhesive, and bake it in a 90℃ oven for about one hour.

[0071] A 2 μm thick layer of photoresist was spin-coated onto a quartz substrate using a spin coater at 4500 r / s for 30 s. The photoresist used was AZ P4620 MicroChemicals.

[0072] A quartz substrate coated with photoresist was placed in a grayscale lithography device, with the laser energy set to 380 mw and the single exposure time to 50 ms.

[0073] Prepare an 8‰ NaOH developer solution, place the exposed sample in it for 15 seconds, and obtain the prepared ADL. Microscopic images of the microstructure of the ADL sample are shown below. Figure 4 As shown.

[0074] To demonstrate the superiority and benefits of this method in design and manufacturing, such as Figure 5 This indicates that the PSF at the sampling wavelength has good focusing performance and energy consistency. Table 1 shows the comparison of single-step optimization time for ADL at a single wavelength. The relationship between the single-step optimization time and aperture index for a single wavelength (this is for ease of comparison; in actual optimization design, multiple wavelengths are used) shows that the optimization time remains basically unchanged because of the high optimization efficiency.

[0075]

[0076] Example 2

[0077] The design process of a single-piece all-visible light imaging diffraction lens is as follows:

[0078] Step 1: Confirm the input of the design component. The component aperture is 10mm, the focal length is 100mm, the design wavelength is 400-700nm, and the sampling interval is 25.

[0079] Step 2: Determine the polychromatic light plane wave matrix based on the design wavelength, and at the same time randomly generate the microstructure distribution matrix of the ADL.

[0080] Step 3: Obtain the complex plane wave matrix of the polychromatic light according to scalar diffraction theory, pass it through the complex amplitude matrix of ADL, and add noise to obtain... U 0 (m,n,c) The expression for the complex amplitude of the wavefront propagating through the air to the image plane. U(x,y) The PSF function is obtained by taking the modulus. PSF(m,n,c) Furthermore, the MTF and PTF are obtained by performing a Fourier transform on the PSF function.

[0081] Step 4: Using the loss function of Method 2, i.e., directly calculating the loss function of the front-end optical system, determine whether the loss function meets the requirements (loss function less than 0.4). If it meets the design requirements, perform simulation imaging on the ADL that meets the requirements, and output the microstructure distribution matrix of the ADL and the imaging results. If the loss function of Method 2 does not meet the requirements, evaluate the performance of the ADL using the result of the loss function, and backpropagate the gradient to optimize the one-dimensional array of radial microstructure distribution of the ADL. D(r) The optimized ADL microstructure distribution matrix is ​​then added to the noise matrix, and step 3 is repeated for further evaluation until the requirements are met.

[0082] Taking into account factors such as the designed ADL aperture, surface microstructure height, and minimum feature size, laser direct writing was adopted as the processing method, as described in Example 1.

[0083] Advances in neural network computing technology can be seen in algorithm complexity and runtime. The method proposed in this invention first reduces algorithm complexity structurally (because the loss function of the front-end optical elements can be separated from the loss function of the back-end image); secondly, it reduces restoration time; and thirdly, compared with the OTF calculated by conventional algorithms, the algorithm of this invention more closely approximates the calculated value, i.e., it is more accurate. The above description is merely a preferred embodiment of this invention and is not intended to limit the scope of this invention. All equivalent changes or modifications made within the scope of the claims of this invention are covered by this invention.

Claims

1. A design method for a monolithic all-visible light imaging diffraction lens, characterized in that, This includes a design algorithm for ADL microstructure distribution optimization based on neural networks and an ADL imaging restoration algorithm. The design algorithm for ADL microstructure distribution optimization optimizes the surface microstructure of the ADL by deriving the imaging process of a plane wave passing through the ADL, obtaining the ADL's optical functions PSF and OTF as constraints, and establishing a loss function. Then, the ADL imaging restoration algorithm is used to simulate and output the microstructure. In the ADL microstructure distribution optimization process based on neural networks, the two-dimensional ADL microstructure distribution is introduced into a concentric ring model to decompose a one-dimensional rotationally symmetric microstructure distribution model. Based on the microstructure distribution model, PSF and OTF at different wavelengths are derived as imaging evaluation results; the loss function is the loss function generated by constraining ADL imaging; the loss function generated by constraining ADL imaging is divided into: the loss function obtained by constraining PSF at different wavelengths due to energy inhomogeneity at different wavelengths, the loss function of frequency cutoff in the transfer function that suppresses MTF, and the loss function generated by constraining the continuity of microstructure distribution in ADL; when using the loss function generated by constraining ADL imaging, the loss function is directly judged without performing simulation imaging calculation on ADL to determine the loss function between the ideal image and the ADL simulation imaging; In the loss function generated by constrained ADL imaging, due to energy non-uniformity at different wavelengths, L1 is obtained by constraining PSF(r,c) at different wavelengths; L2 is the frequency cutoff in the transfer function that suppresses the existence of MTF; L3 is generated by constraining the continuity of microstructure distribution in ADL; the total loss function of the front-end optical system can be written as: ； k is a hyperparameter representing the concentration constraint weight of PSF(r,c); mean(·) represents the mean operation, variance(·) represents the variance operation, R indicates that the outermost ring of PSF(r,c) is the Rth ring, R(c), G(c) and B(c) represent the responses of the camera sensor to different wavelengths λ(c) at different color channels of red, green and blue, respectively, r1 represents the variance weight; zpt represents the index of the first frequency cutoff position of MTF; D(r) represents the radial microstructure distribution of ADL.

2. The design method of the monolithic all-visible light imaging diffraction lens according to claim 1, characterized in that, Once the design algorithm for ADL microstructure distribution optimization based on neural networks is constructed, the wavelength of the input plane wave can be arbitrarily selected.

3. The design method of the monolithic all-visible light imaging diffractive lens according to claim 1, characterized in that, The input of the design element is confirmed, and then a monolithic full-visible light imaging diffraction lens is prepared based on the design algorithm of ADL microstructure distribution optimization and ADL imaging restoration algorithm using neural networks.

Images