Preparation method of in-vitro color vision correction wearable device

Abstract

The invention relates to a preparation method of in-vitro color vision correction wearable equipment, which comprises the following steps of: accurately determining a color weakness / color blindness type by adopting triple verification of image measurement, a color vision correction simulation model and a naturalness evaluation index, and outputting a spectral line of an optical filter according with a color vision disorder condition by utilizing a naturalness and contrast evaluation system; constructing a PDK file between the spectral parameters of the nanoparticles and the structure and process parameters; the method comprises the following steps: establishing a reverse design scheme for synthesizing multi-component plasmon nanoparticles by adopting a gradient descent algorithm and a neural network algorithm, and providing spectral parameters of each component nanoparticle; and constructing a PDK file self-assembled by the nanoparticles, and providing customized full-process design for visual correction by combining the structure parameters and the process parameters of the target nanoparticles obtained by the process. According to the method, through the wearable filter, the image and color recognition capability of the individual patient watching the public display screen is directly changed, and patient-oriented differentiation and customization of low-cost color vision correction wearable equipment are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to a preparation method of an in-vitro color vision correction wearable device, belonging to the field of algorithm design / visual perception / nano-optics. BACKGROUND

[0002] With the in-depth research of plasmonic nano-filtering structure, the multi-dimensional adjustable feature enables it to be widely applied in visual regulation in the imaging field, especially in color vision correction. Display devices have been integrated into all aspects of life, and are one of the most important carriers of information transmission, and also the main shackles of the convenient life of color vision impaired groups. The low practicability of the visual correction filter of traditional dyeing materials requires multiple tests and feedbacks of color vision after the use of patients to verify the effectiveness, which is not only low in efficiency and high in cost, but also seriously dependent on the subjective expression of patients in the invasive diagnosis and treatment, so that the correction effect of the filter device is difficult to guarantee. The plasmonic nano-filtering structure provides a new possibility for the personalized, practical and low-cost customized color vision correction device. Today, the research on visual simulation has been in a stage with a relatively complete theoretical basis, and various new theoretical models have emerged in an endless stream. Especially, the establishment of the visual imaging simulation model based on the opponent color space has perfectly realized the simulation of various visual imaging effects. On this basis, a mature color vision correction simulation model needs to be established to provide help for the diagnosis and treatment of color vision impaired patients and the research and development of color vision correction equipment.

[0003] The spectral filtering link is introduced into the mature visual simulation model, which can realize the output of the sensory image after color vision correction. Before the color information of the display is perceived by the LMS cone cells of the human eye, the spectral filter corrects the color information, and then the simulated information of the cone cells is transmitted to the color opponent space for processing, and finally the picture is presented in the human brain. The color vision correction simulation model established by the present application simulates the above-mentioned visual imaging process, directly outputs the color vision correction perception image by adjusting the patient type, severity and digital filtering information, and combines with the accuracy evaluation system, so as to accurately obtain the color vision impaired patient information and appropriate filtering information of the patient.

[0004] The preparation of existing nano-structured optical filters relies on experimental experience, resulting in unstable optical spectra and high preparation costs. To improve the efficiency of nano-structured optical filter preparation, a PDK file can be established by drawing on the standardized process library of semiconductor devices, establishing a bridge between simulation models and application research and development. By fitting the relationship between the spectral parameters of nanoparticles and the structural parameters and process parameters, the required filter spectrum information of the patient is known, and the required nano-structure is accurately prepared. The reverse design idea is used to realize the whole process of diagnosis and treatment for color vision impaired patients. However, the existing plasmonic nano-structured optical filter is a single nanoparticle, which mostly has only one resonance peak, and the spectral change range is limited. In order to realize the multi-resonance peak of nano-structure, different nanoparticles need to be mixed to form a combined spectrum. This means that the fitting between the spectral parameters and the structural parameters and process parameters is a multi-variable optimization. Therefore, establishing an intelligent algorithm model to assist in diagnosis and treatment is also a new idea for color vision correction research to move towards clinical diagnosis and treatment. SUMMARY

[0005] TECHNICAL PROBLEM: The purpose of the present application is to provide a preparation method of an in-vitro color vision correction wearable device, which determines the type and degree of color weakness / color blindness of the patient by using a color vision test chart, a visual correction simulation model and a naturalness evaluation index, and outputs filter information that meets the visual correction requirements of the patient according to the contrast and naturalness evaluation system; establishes a standardized preparation and self-assembly process library of nanoparticles; uses an intelligent algorithm to realize the reverse design of obtaining the preparation information of the corresponding multi-component nanoparticles from the target filter information; and finally self-assembles the prepared multi-component nanoparticles to obtain a customized color vision correction wearable device for the patient.

[0006] TECHNICAL SCHEME: The preparation method of an in-vitro color vision correction wearable device of the present application comprises the following steps:

[0007] Step 1: Extract the three primary color light spectrum data of the display light source using a fiber optic spectrometer;

[0008] Step 2: Use color discrimination method and color vision correction simulation model, combined with naturalness evaluation index, to accurately determine the color discrimination type and degree;

[0009] Step 3: Input the corresponding color discrimination type and degree into the color vision correction simulation model, and extract the customized color vision correction filter spectrum data, i.e. the target spectrum, from the spectral filtering link of the color vision correction simulation model, combined with the contrast and naturalness evaluation system;

[0010] Step 4: Prepare metal plasmonic nanostructures using chemical synthesis processes, and obtain the corresponding spectral parameters by adjusting the structural parameters; based on the spectral information of the experimentally prepared nanostructures, construct PDK files corresponding to the spectral parameters, structural parameters, and process parameters of a single group of nanoparticles; realize the reverse design of spectral parameters and preparation parameters of single-component nanoparticles based on the PDK files;

[0011] Step 5: Using artificial intelligence algorithms, the design spectrum of multi-component nanoparticles that closely approximate the target spectrum is selected by gradient descent method, and the spectral parameters of each component nanoparticle are given. Then, the structural parameters and process parameters of the multi-component nanoparticles corresponding to the PDK file in Step 4 are used to realize the inverse design model for the synthesis of multi-component plasmonic nanoparticles.

[0012] Step 6: Within physical constraints, use a random function to generate a dataset of spectral parameters for each component nanoparticle, and use a neural network model to train and test the dataset. Input the target spectrum into the trained neural network model to obtain the spectral parameters of each component nanoparticle. Then, use the PDK file from Step 4 to correspond to the structural and process parameters of the multi-component nanoparticles to realize the second reverse design model for the synthesis of multi-component plasmonic nanoparticles.

[0013] Step 7: Using the anisotropic metal plasmon nanostructures prepared in Step 4 as the basic unit for constructing the filter, multiple units are combined to form an arbitrary filter spectrum, and a PDK file for the nanoparticle spectral filtering structure assembled by the ascorbic acid method is constructed; using the structural parameters and process parameters output by the reverse design model in Step 5 or Step 6, a multi-component nanoparticle solution is prepared, and a vision correction wearable device is prepared by using the PDK file for the nanoparticle spectral filtering structure assembled by the ascorbic acid method.

[0014] in,

[0015] Step 1 involves using a fiber optic spectrometer to extract the three primary color emission spectrum data of the display light source. The extraction steps are as follows:

[0016] a. Correct the dark current of the fiber optic spectrometer to remove noise from the detector itself;

[0017] b. The monitor displays red (R), green (G), and blue (B) sequentially across the entire screen;

[0018] c. The fiber optic probe is positioned approximately 1 cm away from the emitting area of ​​the screen to collect the spectrum of the display.

[0019] d. After the spectrum stabilizes, save and output the spectral data CSV file, which includes wavelength and light intensity information;

[0020] e. Repeat the above steps to save the spectral data of R, G, and B respectively.

[0021] Step 2 utilizes a triple approach—color discrimination method, color vision correction simulation model, and naturalness evaluation index—to accurately determine color discrimination information. The process is as follows:

[0022] a. Color vision test charts are used to determine the type of color vision deficiency and to make a preliminary assessment of its severity;

[0023] b. Use the naturalness index to determine the simulation effect of the color vision correction simulation model;

[0024] c. Input the type of color vision deficiency and different degrees of severity into the color vision correction simulation model, and output the visual simulation image after color vision correction for identification.

[0025] The color vision correction simulation model in step 2 simulates the following process: After the color information of the display is selectively filtered by the spectral filter, it is perceived by the LMS cone cells. The simulated signal enters the color space for further processing. The adversarial color space transforms the response of the spectral information filtered by the cone cells into three channels: black and white (WS), blue and yellow (YB), and red and green (RG). Finally, the brain generates a color perception image of the external environment. The adjustable variables in the color vision correction simulation model are: a. the type and severity of color vision deficiency; b. the type of spectral filtering; c. whether to introduce the adversarial color space.

[0026] To ensure the accuracy of visual image simulation, a comprehensive evaluation system of contrast / naturalness is introduced (other subjective or objective evaluation physical quantities can also be introduced). The formula for calculating contrast is:

[0027]

[0028] The following formulas are used for evaluation:

[0029] E symbol =|SIM(T(C) i ))-SIM(T(C j ))| 2 -|C i -C j | 2

[0030] Among them, C i C j SIM(·) represents the color in the color vision test image, T(·) represents the image after spectral filtering correction, and SIM(·) represents the image from the perspective of the color vision disorder.

[0031] The formula for calculating naturalness is:

[0032]

[0033] Naturalness characterizes the degree of color distortion perceived by the patient after using a spectral filter; the better the naturalness, the easier it is for colorblind patients to adapt to the viewing angle under the filter; E naturalness The larger the value, the worse the naturalness; convert to 1-Norm(E) naturalness ) to conduct the evaluation; 1-Norm(E naturalness The larger the value, the better the naturalness.

[0034] In step 3, customized color vision correction filter spectral line data are extracted from the spectral filtering stage of the color vision correction simulation model. The extraction process is as follows: based on the aforementioned triple judgment process, the visual simulation process for determining color vision impairment information and optimal correction effect is determined, and the spectral filtering response curve in this process is extracted as the target spectral line for the preparation of multi-component nanoparticles.

[0035] The matrix form for spectral filtering is:

[0036]

[0037] Where R, G, and B are the three primary color digital signals, and Ψ is the transmission coefficient of the spectral filter.

[0038] In step 4, a PDK file is constructed to correspond the spectral parameters, structural parameters, and process parameters of a single component. Isotropic seeds are used to prepare anisotropic nanoparticles. The correspondence between the spectral parameters, structural parameters, and process parameters is obtained through spectral testing. The spectral parameters include resonance peak intensity, center wavelength, full width at half maximum (FWHM), and resonance depth. The process parameters include the amount of seed crystals, particle concentration, and reaction time. The structural parameters include particle morphology, particle size, particle spacing, and particle type.

[0039] Based on the established PDK file, reverse design of "spectral parameters - preparation parameters" for single-component nanoparticles can be achieved. Since it is a single variable, the spectral parameters and preparation parameters can be easily fitted to obtain a corresponding relationship, thus obtaining a designed spectrum that is close to the target spectrum. More accurate design requires the introduction of reverse design for the synthesis of multi-component nanoparticles.

[0040] Step 5 establishes the first inverse design model for the synthesis of multi-component plasmonic nanoparticles. The establishment steps are as follows:

[0041] a. Define a function L to determine the difference between the design spectrum and the target spectrum:

[0042]

[0043] Where f t (λ) is the target spectral function, f d(x, λ) represents the filtered spectrum of the multi-component nanoparticles, where the wavelength λ exists in the range λ1 < λ < λ2. The target spectrum is a Gaussian function. The peak value, intensity, and full width at half maximum (FWHM) parameters are input into the spectral function to fit the spectral line of the target function. The corresponding parameters used in this invention are: wavelength range [370 nm, 760 nm], peak value at 440 nm, peak intensity of 0.9, and FWHM of 100 nm.

[0044] b. Set initial values, i.e., the spectral parameters of each component nanoparticle and their constraints. The n spectral parameters are represented by an n×1 array x = [x1, x2, ..., x...]. n ] T ;

[0045] Normalization and denormalization functions are defined to map the spectral parameters of each component nanoparticle to the [0,1] space, and after optimization, they are mapped back to the actual physical space. The multi-component particles consist of one type of pentagonal nanorod and two types of triangular nanoplates, involving seven spectral parameters, as shown in the parameter list below:

[0046]

[0047]

[0048] c. Use the fmincon function to find the minimum value of the function in step a under the constraints, and save the minimum value and the corresponding independent variable; multiply the error calculated by the ErrorCal_rand function by an amplification factor of 100 to improve the optimization sensitivity;

[0049] d. Repeat step c by changing the initial value, and output the combination with the smallest gap function as the spectral parameters of each component nanoparticle;

[0050] c. Using the PDK file for the synthesis of single-component nanoparticles established according to claim 6, and comparing it with the spectral parameters output in step d, the structural and process parameters for the synthesis of single-component nanoparticles are obtained, and the target nanoparticles are then prepared.

[0051] Step 6 establishes the second inverse design model for the synthesis of multi-component plasmonic nanoparticles. The establishment steps are as follows:

[0052] a. Define 7 parameters and 7 boundary conditions for the spectrum of multi-component nanoparticles consistent with the parameter list, generate 1000 sets of spectral parameter datasets for each component nanoparticle using a random function, and call the spectrumCal function to calculate the corresponding spectra;

[0053] c. Import the spectral parameter datasets and spectral lines of each component, using 80% of the data as the training set and 20% as the test set;

[0054] d. Construct a neural network model and use tf.keras to build a sequential model:

[0055] The first convolutional layer contains 30 5-point convolutional kernels with a stride of 3; the second convolutional layer contains 15 5-point convolutional kernels with a stride of 3; the Dropout layer has a probability of 0.5 to suppress overfitting; after Flattening, a 990-dimensional vector is obtained, which is successively compressed to 250-100-7 dimensions, corresponding to 7 nanostructure parameters to be regressed.

[0056] e. Create the loss function:

[0057]

[0058] Wherein, the loss function is the spectral parameter x predicted by the neural network. true Compared with the actual normalized spectral parameters x in the training set pred The sum of squares of the differences, M represents the different weights added to each spectral parameter, and n represents the number of training iterations;

[0059] f. Train the neural network, repeat the iteration, observe the changes in the mean of the loss function on the dataset and the mean of the loss function on the cross-validation set, determine the appropriate number of iterations, and save the model;

[0060] g. Input the target spectrum into the trained model to output the spectral parameters of each component nanoparticle;

[0061] h. Based on the PDK file for the synthesis of single-component nanoparticles established in step 4, and by comparing the spectral parameters output in the steps, the structural and process parameters for the synthesis of single-component nanoparticles are obtained, and the target nanoparticles are then prepared.

[0062] In step 7, a PDK file for the self-assembly of ascorbic acid nanoparticles is constructed. The correspondence between the two is determined by fitting the spectral parameters and the assembly parameter spectral lines. The spectral parameters include resonance peak intensity, center wavelength, full width at half maximum (FWHM), and resonance depth. The assembly parameters include ascorbic acid content, nanoparticle type, multi-component ratio, and assembly time.

[0063] The nanoparticle spectral filtering structure assembly method in step 7 uses a high-transmittance colorless optical medium sheet as the assembly carrier, with the plasmonic nanoparticles attached to its surface. It is applied to near-eye protective glasses, eyeglasses and their clip-on lenses, Vision Pro glass panels, and helmet visors. After assembly, it becomes an external color vision correction wearable device.

[0064] Beneficial effects:

[0065] 1. This invention designs a complete diagnosis and treatment plan for people with color vision disorders. Based on color weakness / blindness test charts, the type and severity of the disorder are roughly determined. Then, using a color vision correction simulation model with naturalness evaluation, the visual images of patients with color vision disorders and their corrected vision are accurately simulated. After confirmation by the patient, a personalized target spectrum is output. The target spectrum is input into an established intelligent algorithm reverse design model, which outputs the structural and process parameters for the preparation of each component plasmon nanoparticle. Based on the PDK file for the synthesis of single-component nanoparticles, target nanoparticles of each component are prepared. Based on the PDK file for the self-assembly of ascorbic acid nanoparticles, the target nanoparticles of each component are assembled on a wearable carrier, ultimately producing a customized color vision correction wearable device. This end-to-end color vision correction diagnosis and treatment process is highly efficient. The patient's color vision information undergoes triple verification (color vision test, naturalness evaluation index, and patient confirmation). It establishes a bridge between simulation models and actual research and development, filling the gap in the preparation of wearable color vision correction devices for individual patients. Specifically, this method uses wearable physical filters to directly improve the image quality enhancement and color recognition capabilities of individual terminals when viewing public display screens.

[0066] 2. This invention constructs a shareable PDK standardized process document and establishes a reverse design model for the synthesis of plasmonic nanoparticles, which simplifies the process flow of nanoparticle preparation and assembly. It is not limited to the reverse synthesis of single-component nanoparticles. By combining gradient descent and neural network intelligent algorithms, a reverse design model for the synthesis of multi-component nanoparticles is formed. This model can obtain structural and process parameters of multi-component nanoparticles that are accurately adapted to the target spectrum, avoiding trial and error optimization in the preparation stage and reducing R&D costs.

[0067] 3. This invention digitizes the information expression of the visual synthesis process, enabling digital control of various aspects such as display light source, color vision correction spectral filtering, and cone cell response, thus meeting the customized requirements of various patients. Compared to commercial color-blind / color-weakness assistive glasses with limited wavelength selection and low applicability due to their single filter, the color vision correction wearable device prepared by this invention has the advantage of using a self-assembled filter to filter specific wavelengths, thus better adapting to patients with color vision disorders. Furthermore, the color vision filtering simulation model upon which this discovery is based achieves low color deviation, high comfort, and high output efficiency, advancing the clinical diagnosis and treatment of patients with color vision disorders and providing a new solution for the design and promotion of color vision correction products. Attached Figure Description

[0068] Figure 1 This is a flowchart for identifying patients with color vision deficiency.

[0069] Figure 2 This is the calculation process for the visual correction simulation model.

[0070] Figure 3 The gradient descent method is used to design the fitting effect between the spectrum and the target spectrum.

[0071] Figure 4 It is the change of the mean of the loss function with the number of training iterations. Detailed Implementation

[0072] This invention establishes a wearable color vision correction device. It employs a triple verification process—conventional color weakness / blindness mapping, a color vision correction simulation model, and a naturalness evaluation index—to accurately determine the type and severity of color weakness / blindness. Then, using a naturalness and contrast evaluation system, it outputs the spectral lines of filters that match the color vision impairment. A PDK file is constructed to connect the spectral, structural, and process parameters of the nanoparticles. Two intelligent algorithms—gradient descent and neural networks—are used to establish two reverse design schemes for the synthesis of multi-component plasmon nanoparticles, accurately providing the spectral parameters of each component nanoparticle. A PDK file for nanoparticle self-assembly is constructed. Combined with the target nanoparticle structural and process parameters obtained from the above process, the development of a customized wearable color vision correction device for individuals with color vision impairment is achieved.

[0073] Figure 1 A flowchart for identifying individuals with color vision deficiencies is presented. First, a spectrometer is used to extract the three primary colors from the display. Then, using color discrimination methods and a color vision correction simulation model, combined with evaluation indicators, the type and degree of color weakness are accurately determined. The corresponding color weakness type and degree are input into the color vision correction simulation model, and customized color vision correction filter lines, i.e., the target spectrum, are obtained from the model's spectral filtering stage. A PDK file is constructed corresponding to the spectral parameters, structural parameters, and process parameters of single-component nanoparticles. Gradient descent is used to screen for design spectra that closely approximate the target spectrum of multi-component nanoparticles, and the spectral parameters of each component nanoparticle are given. Subsequently, the PDK file corresponding to the spectral parameters, structural parameters, and process parameters of multi-component nanoparticles is used to establish the inverse design model for the synthesis of multi-component plasmon nanoparticles. Within physical constraints, a random function is used to generate a dataset of spectral parameters for each component nanoparticle, and a neural network model is used to train and test the dataset. By inputting the target spectrum into the trained neural network model, the spectral parameters of each component nanoparticle can be obtained. Then, by using the PDK files corresponding to the spectral parameters, structural parameters, and process parameters of the multi-component nanoparticles, a second inverse design model for the synthesis of multi-component plasmonic nanoparticles is established; a PDK file for assembling the nanoparticle spectral filtering structure is constructed. Using the structural and process parameters output from one of the inverse design models, a multi-component nanoparticle solution is prepared. Using the PDK file assembled from the nanoparticle spectral filtering structure, a wearable medical device for visual correction is developed.

[0074] The present invention provides a method for preparing an external color vision correction wearable device, comprising the following steps:

[0075] Step 1: Use a fiber optic spectrometer to extract the three primary color emission spectrum data of the display light source;

[0076] Step 2: Using color discrimination methods and color vision correction simulation models, combined with naturalness evaluation indicators, accurately determine the type and degree of color discrimination;

[0077] Step 3: Input the corresponding color discrimination type and degree into the color vision correction simulation model, and extract customized color vision correction filter spectral line data, i.e., the target spectrum, from the spectral filtering stage of the color vision correction simulation model in combination with the contrast and naturalness evaluation system;

[0078] Step 4: Prepare metal plasmonic nanostructures using chemical synthesis processes, and obtain the corresponding spectral parameters by adjusting the structural parameters; based on the spectral information of the experimentally prepared nanostructures, construct PDK files corresponding to the spectral parameters, structural parameters, and process parameters of a single group of nanoparticles; realize the reverse design of spectral parameters and preparation parameters of single-component nanoparticles based on the PDK files;

[0079] Step 5: Using artificial intelligence algorithms, the design spectrum of multi-component nanoparticles that closely approximate the target spectrum is selected by gradient descent method, and the spectral parameters of each component nanoparticle are given. Then, the structural parameters and process parameters of the multi-component nanoparticles corresponding to the PDK file in Step 4 are used to realize the inverse design model for the synthesis of multi-component plasmonic nanoparticles.

[0080] Step 6: Within physical constraints, use a random function to generate a dataset of spectral parameters for each component nanoparticle, and use a neural network model to train and test the dataset. Input the target spectrum into the trained neural network model to obtain the spectral parameters of each component nanoparticle. Then, use the PDK file from Step 4 to correspond to the structural and process parameters of the multi-component nanoparticles to realize the second reverse design model for the synthesis of multi-component plasmonic nanoparticles.

[0081] Step 7: Using the anisotropic metal plasmon nanostructures prepared in Step 4 as the basic unit for constructing the filter, multiple units are combined to form an arbitrary filter spectrum, and a PDK file for the nanoparticle spectral filtering structure assembled by the ascorbic acid method is constructed; using the structural parameters and process parameters output by the reverse design model in Step 5 or Step 6, a multi-component nanoparticle solution is prepared, and a vision correction wearable device is prepared by using the PDK file for the nanoparticle spectral filtering structure assembled by the ascorbic acid method.

[0082] in,

[0083] Step 1 involves using a fiber optic spectrometer to extract the three primary color emission spectrum data of the display light source. The extraction steps are as follows:

[0084] a. Correct the dark current of the fiber optic spectrometer to remove noise from the detector itself;

[0085] b. The monitor displays red (R), green (G), and blue (B) sequentially across the entire screen;

[0086] c. The fiber optic probe is positioned approximately 1 cm away from the emitting area of ​​the screen to collect the spectrum of the display.

[0087] d. After the spectrum stabilizes, save and output the spectral data CSV file, which includes wavelength and light intensity information;

[0088] e. Repeat the above steps to save the spectral data of R, G, and B respectively.

[0089] Step 2 utilizes a triple approach—color discrimination method, color vision correction simulation model, and naturalness evaluation index—to accurately determine color discrimination information. The process is as follows:

[0090] a. Color vision test charts are used to determine the type of color vision deficiency and to make a preliminary assessment of its severity;

[0091] b. Use the naturalness index to determine the simulation effect of the color vision correction simulation model;

[0092] c. Input the type of color vision deficiency and different degrees of severity into the color vision correction simulation model, and output the visual simulation image after color vision correction for identification.

[0093] The color vision correction simulation model in step 2 works as follows: After selective filtering by a spectral filter, the color information from the display is perceived by the LMS cone cells. The simulated signal enters the visual color space for further processing. The adversarial color space transforms the response of the filtered spectral information from the cone cells into three channels: black and white (WS), blue and yellow (YB), and red and green (RG). Ultimately, this generates a color perception image of the external environment in the brain. The tunable variables in the color vision correction simulation model include: a. the type and severity of color vision impairment; b. the type of spectral filtering; and c. whether to introduce an adversarial color space. The calculation process of the visual correction simulation model after introducing spectral filtering is as follows: Figure 2 L(λ), M(λ), and S(λ) are the normalized relative response intensity coefficients of the three types of cone cells in the visible light range, respectively, and Filter(λ) is the transmission coefficient of the spectral filter film in the visible light range obtained by measuring with a fiber optic spectrometer. In the Machado model, the LMS matrix is ​​left-multiplied by the transition matrix T. LMS2Opp The signal is transferred from the LMS space corresponding to the cone cells to the three adversarial color space channels: WS, YB, and RG.

[0094] The specific principle of the color vision correction simulation model is as follows:

[0095] Input the type and severity of color vision deficiency, along with the light efficiency curve of cone cells for normal color vision, into the color vision correction simulation model. Calculate the abnormal light efficiency curve of cone cells for color blindness / color weakness based on the type and severity of the deficiency. Taking a patient with green color weakness as an example, if the patient's color weakness is α, then their L and S cell light efficiency curves are normal. The M cell light efficiency curve needs to be calculated using the following formula:

[0096] Area L =∫L(λ)dλ

[0097] Area M =∫M(λ)dλ

[0098]

[0099] Multiplying the transmission spectrum of the filter by the light efficiency curve yields the cone cell light efficiency curve corrected by spectral filtering:

[0100]

[0101] The obtained light efficiency curves are converted to the opposite color space to obtain the response curves for the black and white, red and green, and blue and yellow channels:

[0102]

[0103] The transformation matrix for converting from the adversarial color space to the RGB color space is obtained through integration and normalization steps, as follows:

[0104]

[0105] WS R +WS G +WS B =1

[0106] YB R +YB G +YB B =1

[0107] RG R +RG G +RG B =1

[0108] in, This involves pre-inputting the energy distribution of the display's three primary colors into the model, taking the light intensity of each color at 255 in 8-bit sRGB. To enable the model to perform normal matrix transformations in the RGB color gamut, the components of the three channels after integration are normalized. ρ can be determined using these formulas. wsρ RB ρ RG Three normalization coefficients.

[0109] To ensure the accuracy and rationality of the color vision correction simulation model, a contrast and naturalness evaluation system is introduced to optimize the spectrum (other subjective or objective evaluation physical quantities can also be introduced), and a naturalness evaluation index is used to determine the realism of the simulated image. The contrast calculation formula is as follows:

[0110]

[0111] A comprehensive evaluation needs to be conducted using the following formulas:

[0112] E symbol =|SIM(T(C) i ))-SIM(T(C j ))| 2 -|C i -C j | 2

[0113] Among them, C i C j Let E represent the colors in the color vision test image, T(·) be the image after spectral filtering, and SIM(·) be the image from the perspective of the person with color vision deficiency; if E symbol If the value is greater than 0, the contrast is improved. contrast The larger the value of E, the more significant the improvement in color discrimination for the patient through spectral filtering correction. symbol If the value is less than 0, the contrast will decrease instead of increase, and this filter curve should not be used.

[0114] The formula for calculating naturalness is:

[0115]

[0116] Naturalness characterizes the degree of color distortion seen after using a spectral filter; the better the naturalness, the easier it is for people with color vision deficiencies to adapt to the viewing angle under the filter; E naturalness The larger the value, the worse the naturalness; convert to 1-Norm(E) naturalness ) to conduct the evaluation; 1-Norm(E naturalness The larger the value, the better the naturalness.

[0117] In step 3, customized color vision correction filter spectral line data are extracted from the spectral filtering stage of the color vision correction simulation model. The extraction process is as follows: based on the aforementioned triple judgment process, the visual simulation process for determining color vision impairment information and optimal correction effect is determined, and the spectral filtering response curve in this process is extracted as the target spectral line for the preparation of multi-component nanoparticles.

[0118] The matrix form for spectral filtering is:

[0119]

[0120] Where R, G, and B are the three primary color digital signals, and Ψ is the transmission coefficient of the spectral filter.

[0121] In step 4, a PDK file is constructed to correspond the spectral parameters, structural parameters, and process parameters of a single component. Isotropic seeds are used to prepare anisotropic nanoparticles. The correspondence between the spectral parameters, structural parameters, and process parameters is obtained through spectral testing. The spectral parameters include resonance peak intensity, center wavelength, full width at half maximum (FWHM), and resonance depth. The process parameters include the amount of seed crystals, particle concentration, and reaction time. The structural parameters include particle morphology, particle size, particle spacing, and particle type.

[0122] Based on the established PDK file, reverse design of "spectral parameters - preparation parameters" for single-component nanoparticles can be achieved. Since it is a single variable, the spectral parameters and preparation parameters can be easily fitted to obtain a corresponding relationship, thus obtaining a designed spectrum that is close to the target spectrum. More accurate design requires the introduction of reverse design for the synthesis of multi-component nanoparticles.

[0123] Based on the aforementioned positive fitting of spectral parameters with structural and process parameters of single-component nanoparticles, a reverse prediction mathematical model for multi-component nanoparticles is further established. The mathematical model used for spectral fitting is as follows:

[0124] An ideal spectrum can be viewed as a superposition of Gaussian functions. If there are n superimposed Gaussian functions, then:

[0125]

[0126] This can represent n resonant peaks, where a is the peak maximum value, b represents the wavelength corresponding to the peak value, and c is the full width at half maximum (FWHM). x represents wavelength (independent variable), and y represents extinction intensity (dependent variable). Since the actual spectrum is not ideal, a Gaussian term is needed for approximation in applications.

[0127] In step 5, an inverse design model for the synthesis of multi-component plasmon nanoparticles is established, which is constructed using the gradient descent method as follows:

[0128] 1. Establishing the target spectrum and preprocessing. A Gaussian function is used to generate a spectrum with fixed peak value, full width at half maximum (FWHM), and peak intensity, which serves as the target spectrum for simulation. The wavelength and intensity data are arranged according to a certain pattern to ensure the stability of subsequent interpolation.

[0129] 2. Constructing the parameter space and normalization. Optimize the spectral parameters and boundary conditions for each nanoparticle band, and construct bidirectional mapping functions normalizing_function and anti_normalizing_function to transform the physical values ​​between the [0,1] space;

[0130] 3. Encapsulation of the objective function. The spectral difference comparison is transformed into a differentiable scalar minimization problem. The ErrorCal_rand function is called, taking into account the parameters of each nanoparticle, the wavelength vector, and the target spectrum, and returning the normalized mean square error.

[0131] 4. Multi-starting point stochastic gradient descent optimization sets upper and lower bounds, calls the fmincon function, and performs multiple random initial values ​​to avoid local minima and improve the probability of global optimization.

[0132] 5. Result Screening and Data Presentation. The spectrum is reconstructed using the `spectrumCal` function, the matching degree is evaluated, and the physical parameters with the smallest error are selected from the 10 runs. The fitting effect between the designed spectrum and the target spectrum is shown below. Figure 3 .

[0133] The specific steps for building the model are as follows:

[0134] a. Define a function L to determine the difference between the design spectrum and the target spectrum:

[0135]

[0136] Where f t (λ) is the target spectral function, f d (x, λ) represents the filtered spectrum of the multi-component nanoparticles, where the wavelength λ exists in the range λ1 < λ < λ2. The target spectrum is a Gaussian function. The peak value, intensity, and full width at half maximum (FWHM) parameters are input into the spectral function to fit the spectral line of the target function. The corresponding parameters used in this invention are: wavelength range [370 nm, 760 nm], peak value at 440 nm, peak intensity of 0.9, and FWHM of 100 nm.

[0137] b. Set initial values, i.e., the spectral parameters of each component nanoparticle and their constraints. The n spectral parameters are represented by an n×1 array x = [x1, x2, ..., x...]. n ] T ;

[0138] Normalization and denormalization functions are defined to map the spectral parameters of each component nanoparticle to the [0,1] space, and after optimization, they are mapped back to the actual physical space. The multi-component particles consist of one type of pentagonal nanorod and two types of triangular nanoplates, involving seven spectral parameters, as shown in the parameter list below:

[0139] Symbol Name Meaning Unit Range x1 Ag5NP_lambda_of_peak_s Ag5NP_lambda_of_peak_l Ag5NP_height [385,450] x2 Ag3NP1_lambda_of_peak Ag3NP1_height Ag3NP2_lambda_of_peak [420,850] x3 Ag3NP2_height Figure 4 ​ [0,+∞) x4 ​ ​ ​ [500,950] x5 ​ ​ ​ [0,+∞) x6 ​ ​ ​ [500,950] x7 ​ ​ ​ [0,+∞)

[0140] c. Use the fmincon function to find the minimum value of the function in step a under the constraints, and save the minimum value and the corresponding independent variable; multiply the error calculated by the ErrorCal_rand function by an amplification factor of 100 to improve the optimization sensitivity;

[0141] d. Repeat step c by changing the initial value, and output the combination with the smallest gap function as the spectral parameters of each component nanoparticle;

[0142] c. Using the PDK file for the synthesis of single-component nanoparticles established according to claim 6, and comparing it with the spectral parameters output in step d, the structural and process parameters for the synthesis of single-component nanoparticles are obtained, and the target nanoparticles are then prepared.

[0143] In step 6, a second inverse design model for the synthesis of multi-component plasmonic nanoparticles is established. The construction process using a neural network is as follows:

[0144] 1. Dataset Generation. Set boundary conditions and call the spectrumCal function to generate 10,000 sets of spectral parameter samples for each nanoparticle;

[0145] 2. Dataset preparation. Read the data samples, using the first 80% as the training set and the last 20% as the test set. Input the spectra and output the spectral parameters of each nanoparticle.

[0146] 3. Normalization. The optical profile parameters of each output nanoparticle are normalized, and the input spectrum is already in the [0,1] space;

[0147] 4. Construct a deep mapping model. Convolutional layers automatically extract local peak position / peak width features, reducing the number of parameters;

[0148] 5. Define loss and training data. Design a weighted MSE loss, use the training dataset, automatically save the optimal loss value each round, and measure the change in the mean of the loss function with the number of training iterations. ​ This allows us to determine the optimal number of training sessions;

[0149] 6-parameter prediction. Load the trained model to obtain normalized predictions. Reconstruct the spectrum using calc_spectrum based on the predicted parameters and compare the difference between the target spectrum and the reconstructed design spectrum.

[0150] The specific steps for building the model are as follows:

[0151] a. Define 7 parameters and 7 boundary conditions for the spectrum of multi-component nanoparticles consistent with the parameter list, generate 1000 sets of spectral parameter datasets for each component nanoparticle using a random function, and call the spectrumCal function to calculate the corresponding spectra;

[0152] c. Import the spectral parameter datasets and spectral lines of each component, using 80% of the data as the training set and 20% as the test set;

[0153] d. Construct a neural network model and use tf.keras to build a sequential model:

[0154] The first convolutional layer contains 30 5-point convolutional kernels with a stride of 3; the second convolutional layer contains 15 5-point convolutional kernels with a stride of 3; the Dropout layer has a probability of 0.5 to suppress overfitting; after Flattening, a 990-dimensional vector is obtained, which is successively compressed to 250-100-7 dimensions, corresponding to 7 nanostructure parameters to be regressed.

[0155] e. Create the loss function:

[0156]

[0157] Wherein, the loss function is the spectral parameter x predicted by the neural network. true Compared with the actual normalized spectral parameters x in the training set pred The sum of squares of the differences, M represents the different weights added to each spectral parameter, and n represents the number of training iterations;

[0158] f. Train the neural network, repeat the iteration, observe the changes in the mean of the loss function on the dataset and the mean of the loss function on the cross-validation set, determine the appropriate number of iterations, and save the model;

[0159] g. Input the target spectrum into the trained model to output the spectral parameters of each component nanoparticle;

[0160] h. Based on the PDK file for the synthesis of single-component nanoparticles established in step 4, and by comparing the spectral parameters output in the steps, the structural and process parameters for the synthesis of single-component nanoparticles are obtained, and the target nanoparticles are then prepared.

[0161] In step 7, a PDK file for the self-assembly of ascorbic acid nanoparticles is constructed. The correspondence between the two is determined by fitting the spectral parameters and the assembly parameter spectral lines. The spectral parameters include resonance peak intensity, center wavelength, full width at half maximum (FWHM), and resonance depth. The assembly parameters include ascorbic acid content, nanoparticle type, multi-component ratio, and assembly time.

[0162] The nanoparticle spectral filtering structure assembly method in step 7 uses a high-transmittance five-color optical medium sheet as the assembly carrier, with the plasmonic nanoparticles attached to its surface. It is applied to near-eye protective glasses, eyeglasses and their clip-on lenses, Vision Pro glass panels, and helmet visors. After assembly, it becomes an external color vision correction wearable device.

Claims

1. A method for preparing an external color vision correction wearable device, characterized in that, Includes the following steps: Step 1: Use a fiber optic spectrometer to extract the three primary color emission spectrum data of the display light source; Step 2: Using color discrimination methods and color vision correction simulation models, combined with naturalness evaluation indicators, accurately determine the type and degree of color discrimination; Step 3: Input the corresponding color discrimination type and degree into the color vision correction simulation model, and extract customized color vision correction filter spectral line data, i.e., the target spectrum, from the spectral filtering stage of the color vision correction simulation model in combination with the contrast and naturalness evaluation system; Step 4: Prepare metal plasmonic nanostructures using chemical synthesis processes, and obtain the corresponding spectral parameters by adjusting the structural parameters; based on the spectral information of the experimentally prepared nanostructures, construct PDK files corresponding to the spectral parameters, structural parameters, and process parameters of a single group of nanoparticles; realize the reverse design of spectral parameters and preparation parameters of single-component nanoparticles based on the PDK files; Step 5: Using artificial intelligence algorithms, the design spectrum of multi-component nanoparticles that closely approximate the target spectrum is selected by gradient descent method, and the spectral parameters of each component nanoparticle are given. Then, the structural parameters and process parameters of the multi-component nanoparticles corresponding to the PDK file in Step 4 are used to realize the inverse design model for the synthesis of multi-component plasmonic nanoparticles. Step 6: Within physical constraints, use a random function to generate a dataset of spectral parameters for each component nanoparticle, and use a neural network model to train and test the dataset. Input the target spectrum into the trained neural network model to obtain the spectral parameters of each component nanoparticle. Then, use the PDK file from Step 4 to correspond to the structural and process parameters of the multi-component nanoparticles to realize the second reverse design model for the synthesis of multi-component plasmonic nanoparticles. Step 7: Using the anisotropic metal plasmon nanostructures prepared in Step 4 as the basic unit for constructing the filter, multiple units are combined to form an arbitrary filter spectrum, and a PDK file for the nanoparticle spectral filtering structure assembled by the ascorbic acid method is constructed; using the structural parameters and process parameters output by the reverse design model in Step 5 or Step 6, a multi-component nanoparticle solution is prepared, and a vision correction wearable device is prepared by using the PDK file for the nanoparticle spectral filtering structure assembled by the ascorbic acid method.

2. The method for preparing an external color vision correction wearable device according to claim 1, characterized in that, Step 1 involves using a fiber optic spectrometer to extract the three primary color emission spectrum data of the display light source. The extraction steps are as follows: a. Correct the dark current of the fiber optic spectrometer to remove noise from the detector itself; b. The monitor displays red (R), green (G), and blue (B) sequentially across the entire screen; c. The fiber optic probe is positioned approximately 1 cm away from the emitting area of ​​the screen to collect the spectrum of the display. d. After the spectrum stabilizes, save and output the spectral data CSV file, which includes wavelength and light intensity information; e. Repeat the above steps to save the spectral data of R, G, and B respectively.

3. The method for preparing an external color vision correction wearable device according to claim 2, characterized in that... Step 2 utilizes a triple approach—color discrimination method, color vision correction simulation model, and naturalness evaluation index—to accurately determine color discrimination information. The process is as follows: a. Color vision test charts are used to determine the type of color vision deficiency and to make a preliminary assessment of its severity; b. Use the naturalness index to determine the simulation effect of the color vision correction simulation model; c. Input the type of color vision deficiency and different degrees of severity into the color vision correction simulation model, and output the visual simulation image after color vision correction for identification.

4. The method for preparing an external color vision correction wearable device according to claim 3, characterized in that... The color vision correction simulation model in step 2 simulates the following process: After the color information of the display is selectively filtered by the spectral filter, it is perceived by the LMS cone cells. The simulated signal enters the color space for further processing. The adversarial color space transforms the response of the spectral information filtered by the cone cells into three channels: black and white (WS), blue and yellow (YB), and red and green (RG). Finally, a color perception image of the external environment is generated in the brain. The tunable variables in the color vision correction simulation model include: a. the type and severity of color vision impairment; b. the type of spectral filtering; c. Whether to introduce a contrasting color space.

5. The method for preparing an external color vision correction wearable device according to claim 4, characterized in that, In step 3, customized color vision correction filter spectral line data are extracted from the spectral filtering stage of the color vision correction simulation model. The extraction process is as follows: based on the aforementioned triple judgment process, the visual simulation process for determining color vision impairment information and optimal correction effect is determined, and the spectral filtering response curve in this process is extracted as the target spectral line for the preparation of multi-component nanoparticles. The matrix form for spectral filtering is: Where R, G, and B are the three primary color digital signals, and Ψ is the transmission coefficient of the spectral filter.

6. The method for preparing an external color vision correction wearable device according to claim 5, characterized in that, In step 4, a PDK file is constructed to correspond the spectral parameters, structural parameters, and process parameters of a single component. Isotropic seeds are used to prepare anisotropic nanoparticles. The correspondence between the spectral parameters, structural parameters, and process parameters is obtained through spectral testing. The spectral parameters include resonance peak intensity, center wavelength, full width at half maximum (FWHM), and resonance depth. The process parameters include the amount of seed crystals, particle concentration, and reaction time. The structural parameters include particle morphology, particle size, particle spacing, and particle type.

7. The method for preparing an external color vision correction wearable device according to claim 6, characterized in that, Step 5 establishes the first inverse design model for the synthesis of multi-component plasmonic nanoparticles. The establishment steps are as follows: a. Define a function L to determine the difference between the design spectrum and the target spectrum: Where f t (λ) is the target spectral function, f d (x,λ) represents the filtered spectrum of the multi-component nanoparticles, where the wavelength λ exists in the range λ1<λ<λ2; the target spectrum is a Gaussian function, and the peak value, intensity, and full width at half maximum (FWHM) parameters are input into the spectral function to fit the spectral line of the target function. b. Set initial values, i.e., the spectral parameters of each component nanoparticle and their constraints. The n spectral parameters are represented by an n×1 array x = [x1, x2, ..., x...]. n ] T ; Normalization and denormalization functions are defined to map the spectral parameters of each component nanoparticle to the [0,1] space, and after optimization, they are mapped back to the actual physical space. The multi-component particles consist of one type of pentagonal nanorod and two types of triangular nanoplates, involving seven spectral parameters, as shown in the parameter list below: c. Use the fmincon function to find the minimum value of the function in step a under the constraints, and save the minimum value and the corresponding independent variable; multiply the error calculated by the ErrorCal_rand function by an amplification factor of 100 to improve the optimization sensitivity; d. Repeat step c by changing the initial value, and output the combination with the smallest gap function as the spectral parameters of each component nanoparticle; c. Using the PDK file for the synthesis of single-component nanoparticles established according to claim 6, and comparing it with the spectral parameters output in step d, the structural and process parameters for the synthesis of single-component nanoparticles are obtained, and the target nanoparticles are then prepared.

8. The method for preparing an external color vision correction wearable device according to claim 7, characterized in that, Step 6 establishes the second inverse design model for the synthesis of multi-component plasmonic nanoparticles. The establishment steps are as follows: a. Define 7 parameters and 7 boundary conditions for the spectrum of multi-component nanoparticles consistent with the parameter list, generate 1000 sets of spectral parameter datasets for each component nanoparticle using a random function, and call the spectrumCal function to calculate the corresponding spectra; c. Import the spectral parameter datasets and spectral lines of each component, using 80% of the data as the training set and 20% as the test set; d. Construct a neural network model and use tf.keras to build a sequential model: The first convolutional layer contains 30 5-point convolutional kernels with a stride of 3; the second convolutional layer contains 15 5-point convolutional kernels with a stride of 3; the Dropout layer has a probability of 0.5 to suppress overfitting; after Flattening, a 990-dimensional vector is obtained, which is successively compressed to 250-100-7 dimensions, corresponding to 7 nanostructure parameters to be regressed. e. Create the loss function: Wherein, the loss function is the spectral parameter x predicted by the neural network. true Compared with the actual normalized spectral parameters x in the training set pred The sum of squares of the differences, M represents the different weights added to each spectral parameter, and n represents the number of training iterations; f. Train the neural network, repeat the iteration, observe the changes in the mean of the loss function on the dataset and the mean of the loss function on the cross-validation set, determine the appropriate number of iterations, and save the model; g. Input the target spectrum into the trained model to output the spectral parameters of each component nanoparticle; h. Based on the PDK file for the synthesis of single-component nanoparticles established in step 4, and by comparing the spectral parameters output in the steps, the structural and process parameters for the synthesis of single-component nanoparticles are obtained, and the target nanoparticles are then prepared.

9. The method for preparing an external color vision correction wearable device according to claim 8, characterized in that, In step 7, a PDK file for the self-assembly of ascorbic acid nanoparticles is constructed. The correspondence between the two is determined by fitting the spectral parameters and the assembly parameter spectral lines. The spectral parameters include resonance peak intensity, center wavelength, full width at half maximum (FWHM), and resonance depth. The assembly parameters include ascorbic acid content, nanoparticle type, multi-component ratio, and assembly time.

10. The method for preparing an external color vision correction wearable device according to claim 9, characterized in that, The nanoparticle spectral filtering structure assembly method in step 7 uses a high-transmittance colorless optical medium sheet as the assembly carrier, with the plasmonic nanoparticles attached to its surface. It is applied to near-eye protective glasses, eyeglasses and their clip-on lenses, Vision Pro glass panels, and helmet visors. After assembly, it becomes an external color vision correction wearable device.

Images