A dynamic data processing method, device and program product based on eye refraction state

Abstract

The application relates to the field of intelligent medical treatment, in particular to a dynamic data processing method, device and program product based on an eye refraction state. The method comprises the following steps: acquiring initial refraction data and environment data of an eye of a person to be measured; preprocessing the initial refraction data and the environment data to obtain preprocessing data; calculating the eye refraction data by using a refraction state influence function; and wherein the refraction state influence function is obtained by calculating the dynamic change relationship between the initial eye refraction data and the environment data. The application can solve the personalized expression of the difference in light perception of different myopia degrees, dynamically predict the brightness and color temperature parameters, drive the actual lighting equipment, realize the dynamic balance of visual comfort and environmental adaptability, and has good practical application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent medical treatment, in particular to a dynamic data processing method and device based on eye refraction state, a program product and a computer readable storage medium. BACKGROUND

[0002] With the increasing dependence on electronic devices and artificial lighting environment in modern people's daily life, the number of people with visual health problems, especially refractive errors, is rapidly growing, and visual fatigue and light discomfort have become a common problem. For highly myopic users, their sensitivity to light changes is significantly higher than that of the general population. Strong brightness or inappropriate color temperature can easily cause glare, dry eyes and even visual fatigue. However, existing intelligent lighting systems mostly adjust light based on environmental sensors or fixed scene settings, and do not fully consider the differences in individual refraction states. SUMMARY

[0003] To solve the above problems, the present application provides a dynamic data processing method based on eye refraction state, which specifically comprises:

[0004] Obtaining initial refraction data of the eyes of the subject to be tested and environmental data, including environmental brightness and environmental color temperature;

[0005] Preprocessing the initial refraction data of the eyes and the environmental data to obtain preprocessed data;

[0006] The preprocessed data is used to calculate the eye refraction data through a refraction state influence function, wherein the refraction state influence function is obtained by calculating the dynamic change relationship between the initial refraction data of the eyes and the environmental data.

[0007] The dynamic change relationship of the refraction state influence function is obtained based on the nonlinear response of the individual refraction state to the light sensitivity;

[0008] The calculation formula of the refraction state influence function is:

[0009]

[0010] wherein, is a first refraction state adjustment hyperparameter, which controls the amplitude of the scaling factor,

[0011] is a second refraction state adjustment hyperparameter, which controls the curvature of the scaling factor, represents the average spherical power, and is the calculation dimension of a single sample.

[0012] The present application also provides a dynamic lighting prediction method based on refraction state, which comprises:

[0013] Obtaining the refraction data of the eyes of the subject to be tested;

[0014] The dioptric data is clustered and feature fused to obtain fused features;

[0015] The fused features are input to a trained prediction model to obtain luminance prediction values and color temperature prediction values;

[0016] The prediction model is initialized by compressing abnormal values of the dioptric data and compensating for distribution deviation of the dioptric clusters.

[0017] Optionally, the eye dioptric data is obtained by the above-mentioned dynamic data processing method based on the eye dioptric state.

[0018] The compressed abnormal values are compressed by a skewness perception scaling matrix; and the distribution deviation of the dioptric clusters is compensated by a bias correction term calculated based on dioptric cluster information.

[0019] The prediction model learns the feature relationship between dioptric sensitivity and luminance by using a dioptric perception adaptive activation function; and the dioptric perception adaptive activation function is obtained by dynamically adjusting the steepness and threshold position of the activation curve based on the average spherical power.

[0020] The clustering is based on the average spherical power in the dioptric data to obtain grouped sample clusters, a projection matrix is constructed to remove redundant features from the grouped sample clusters to obtain feature clusters; and the feature fusion is based on the density of the feature clusters to calculate fusion weights, and the feature clusters with different fusion weights are fused to obtain fused features.

[0021] The purpose of the present application is also to provide a dynamic lighting adjustment method based on the dioptric state, comprising: a light source controller and a lighting source.

[0022] The luminance prediction values and color temperature prediction values are obtained according to the above-mentioned dynamic lighting prediction method based on the dioptric state.

[0023] The luminance prediction values and color temperature prediction values are input to the light source controller.

[0024] The luminance and color temperature of the lighting source are dynamically adjusted by the light source controller.

[0025] The purpose of the present application is to provide a computer program product comprising a computer program or instructions thereon, which are executed by a processor to implement the above-mentioned dynamic data processing method based on the eye dioptric state, or to implement the above-mentioned dynamic lighting prediction method based on the dioptric state, or to implement the above-mentioned dynamic lighting adjustment method based on the dioptric state.

[0026] The computer device of the present application comprises a memory, a processor and a computer program or instructions stored on the memory, and the computer program or instructions are executed by the processor to implement the dynamic data processing method based on the eye refractive state, or to implement the dynamic lighting prediction method based on the refractive state, or to implement the dynamic lighting adjustment method based on the refractive state.

[0027] The computer readable storage medium of the present application stores a computer program or instructions, and the computer program or instructions are executed by the processor to implement the dynamic data processing method based on the eye refractive state, or to implement the dynamic lighting prediction method based on the refractive state, or to implement the dynamic lighting adjustment method based on the refractive state.

[0028] Advantages of the present application:

[0029] 1. For the data processing method, the traditional normalization method assumes that the features are processed in the same scale, and does not consider the nonlinear light sensitivity modulation of the refractive state, resulting in weak modeling ability of the model for individual visual differences. Therefore, the present application adopts a refractive state influence function, dynamically calculates a normalization scaling factor based on the average spherical power of the user, adjusts the feature scale, and solves the personalized expression of the myopia degree on the light perception difference.

[0030] 2. For the feature extraction method, the existing method mostly uses PCA or static dimension reduction strategy, ignores the refractive regulation dependence between features, and loses key interaction features such as "myopic users are more sensitive to brightness". Therefore, the present application groups based on the refractive state, constructs a clustering projection matrix and a fusion weight, reduces the dimension while retaining key discriminant features, and captures nonlinear interactions related to the refractive state.

[0031] 3. For the prediction model, the conventional activation function, initialization method and loss function are unstable for extreme refractive samples such as high myopia, difficult to converge during training, have poor generalization ability, and cannot accurately serve severe users. Therefore, the present application adopts a prediction model structure including an adaptive activation function, an adaptive initialization strategy, a refractive weighted loss function and a modulation path, embeds the refractive state modulation ability from input to output, and improves the prediction accuracy and generalization ability of high myopia users.

[0032] 4. The traditional lighting adjustment is mainly based on scene preset, lacks real-time perception and feedback adjustment of user visual physiological parameters, and cannot realize real-time optimization of lighting conditions according to the refractive state. The model of the present application adopts an end-to-end training method, can realize end-to-end closed-loop lighting adjustment after training, real-time collects user refractive and environmental information, dynamically outputs brightness and color temperature parameters, drives actual lighting devices, and realizes dynamic balance of visual comfort and environmental adaptability. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description only constitute some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0034] Figure 1 A dynamic data processing method flowchart based on eye refraction state provided by the embodiments of the present application;

[0035] Figure 2 A dynamic data processing system diagram based on eye refraction state provided by the embodiments of the present application;

[0036] Figure 3 A dynamic data processing device diagram based on eye refraction state provided by the embodiments of the present application;

[0037] Figure 4 A normalization effect comparison diagram under different refraction states provided by the embodiments of the present application;

[0038] Figure 5 A brightness prediction error distribution under different refraction states provided by the embodiments of the present application;

[0039] Figure 6 A color temperature prediction error distribution under different refraction states provided by the embodiments of the present application;

[0040] Figure 7 An influence of refraction state on light sensitivity provided by the embodiments of the present application;

[0041] Figure 8 An influence of refraction state on preferred brightness provided by the embodiments of the present application;

[0042] Figure 9 An influence of refraction state on light change response time under different technologies provided by the embodiments of the present application. DETAILED DESCRIPTION

[0043] In order to make the person skilled in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application.

[0044] In some of the flowcharts described in the specification and claims of the present application and in the above-mentioned drawings, a plurality of operations are included which occur in a particular order, but it should be clearly understood that these operations can be performed in the order in which they appear herein or in parallel, and the serial numbers of the operations, such as S101, S102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these flowcharts can include more or fewer operations, and the operations can be performed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. herein are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, nor do "first" and "second" represent different types.

[0045] Figure 1 The dynamic data processing method based on the eye refraction state provided by the embodiment of the application includes the following steps:

[0046] S1: obtaining initial eye refraction data and environmental data of a subject;

[0047] In one embodiment, the initial eye refraction data includes one or more of the following: spherical power, cylindrical power, axis, and pupil diameter.

[0048] Optionally, the initial eye refraction data further includes axial length, corneal curvature, and tear film quality index.

[0049] In one embodiment, the environmental data includes one or more of the following: environmental brightness, environmental color temperature, and sensor X / Y axis readings.

[0050] In a specific embodiment, the user's refraction state parameters and environmental light characteristics are collected by a multi-source heterogeneous device:

[0051] The user's spherical power, cylindrical power, axis, and other core refraction parameters are obtained using an optometry instrument;

[0052] The pupil diameter is recorded by a pupilometer;

[0053] The environmental brightness, environmental color temperature, and sensor X / Y axis readings are captured in real time by an environmental light sensor;

[0054] The axial length, corneal curvature, and tear film quality index are obtained using a biometric device.

[0055] S2: preprocessing the initial eye refraction data and environmental data to obtain preprocessed data;

[0056] In one embodiment, the preprocessing includes one or more of the following: normalization processing, data cleaning, and data desensitization;

[0057] Optionally, the eye initial refraction data and the environment data are normalized to obtain normalized data, and the normalized data is used to calculate the eye refraction data by a refraction state influence function.

[0058] S3: the preprocessed data is used to calculate the eye refraction data by a refraction state influence function, wherein the refraction state influence function is obtained by calculating a dynamic change relationship between the eye initial refraction data and the environment data.

[0059] In one embodiment, the dynamic change relationship of the refraction state influence function is obtained based on a nonlinear response of an individual refraction state to light sensitivity.

[0060] In one embodiment, the generation process of the refraction state influence function includes:

[0061] In one embodiment, the generation process of the refraction state influence function includes:

[0062] In one embodiment, the generation process of the refraction state influence function includes:

[0063] In one embodiment, the different refraction state individuals include mild myopia individuals, severe myopia individuals, moderate myopia individuals, and normal individuals.

[0064] In one embodiment, the calculation formula of the refraction state influence function is:

[0065]

[0066] wherein, is a first refraction state adjustment hyperparameter, which controls the amplitude of the scaling factor,

[0067] is a second refraction state adjustment hyperparameter, which controls the curvature of the scaling factor, represents the average spherical power, and is the calculation dimension of a single sample.

[0068] In one specific embodiment, the spherical power, cylindrical power, and other refraction characteristics and the environment brightness and other environmental characteristics have the problems of non-uniform units and non-matching numerical ranges. High myopia users are more sensitive to changes in environmental light. Global normalization methods such as Min-Max normalization cannot effectively handle the multiscale characteristics and dynamic range differences of refraction state data and environment data, which easily leads to the normalization process ignoring the nonlinear influence of refraction state on light sensitivity and being difficult to adapt to individual differences.

[0069] The present application calculates the mean and standard deviation of each feature dimension based on the entire training data set, performs basic normalization on the original feature values, then adopts a refractive state influence function, which takes the average spherical power as input to calculate a scaling factor, and finally multiplies the basic normalized result by the scaling factor to obtain the normalized feature value after adapting to the refractive state, so that the normalization process can dynamically adjust the scaling ratio according to the refractive state of the user, expressed as:

[0070] In the formula, the normalized value of the i-th feature is the normalized value of the i-th feature, and the normalized value of the i-th feature is the normalized value of the i-th feature.

[0071] The average spherical power is expressed as , wherein is the right eye spherical power, is the left eye spherical power, and represents the adjustment effect of the refractive state of the user on the light sensitivity.

[0072] is the feature dimension index, which identifies the position of the feature in the sample, such as corresponding to the age feature;

[0073] The original value of the i-th feature is the original value of the i-th feature, which reflects the original attributes of the sample, such as spherical power, environmental brightness, etc.

[0074] The mean of the i-th feature of all samples is expressed as The standard deviation of the i-th feature of all samples is expressed as

[0075] The refractive state influence function is expressed as

[0076] The hyperbolic tangent function is expressed as

[0077] The first refractive state adjustment hyperparameter is expressed as

[0078] The second refractive state adjustment hyperparameter is expressed as

[0079] ;​​​​​​

[0080] represents a small offset, used to prevent division by zero or numerical instability when approaching zero, ensuring robustness of the normalization calculation, default value is .

[0081] In one embodiment, the refractive data and environmental data attributes of a sample and the original value examples are as follows:

[0082] User age (35, unit: years), gender (1, male, unitless, One-hot encoding format), right eye spherical power (-4.5, unit: diopter), cylinder power right eye (-1.2, unit: diopter), axis right eye (90, unit: degrees), left eye spherical power (-4.0, unit: diopter), cylinder power left eye (-1.0, unit: diopter), axis left eye (85, unit: degrees), pupil diameter (3.8, unit: millimeters), ambient brightness (300, unit: lux), ambient color temperature (4000, unit: Kelvin), axial length (24.2, unit: millimeters), corneal curvature (43.5, unit: diopter), tear film quality index (85, percentage value, unit: %).

[0083] And, based on the attribute values of the sample, the average spherical power is calculated according to the formula , where is the right eye spherical power, is the left eye spherical power, and the right eye spherical power is diopter, and the left eye spherical power is diopter, then (diopter), that is, the average spherical power is diopter.

[0084] In one embodiment, the conventional normalization ignores the refractive state modulation, and the refractive state influence function dynamically adapts to individual differences, which is designed based on the nonlinear characteristics of the refractive state on light sensitivity, the term provides a smooth S-shaped curve through the hyperbolic tangent function, controls the scaling amplitude, controls the curvature, and when represents high myopia, amplifies the feature value, improves the discrimination in low light areas, and when , reduces interference.

[0085] In addition, the output value represents the scaling factor, which dynamically adjusts The scaling ratio is adjusted to magnify or compress the range of characteristic values ​​to accommodate the nonlinear response of users with different refractive states to light sensitivity. For example, users with high myopia require greater scaling to enhance low-light discrimination.

[0086] In one specific embodiment, to verify the feature discrimination improvement effect of the proposed adaptive normalization method on user groups with different refractive states, the feature discrimination index, a key indicator for measuring the quality of feature spatial distribution, is used. This index directly affects the learning ability and prediction accuracy of the subsequent model. Experiments compared the performance of the conventional normalization method (global normalization) and the adaptive normalization method of this invention on three user groups: high myopia, moderate myopia, and mild myopia. A higher feature discrimination index indicates that the feature distribution in space is more conducive to the model distinguishing different patterns, such as... Figure 4 As shown, in the high myopia user group, the feature discrimination of the method of the present invention is significantly higher than that of the conventional method, and the data distribution is more concentrated (as reflected by the width of the box plot and violin plot). In the moderate myopia group, the method of the present invention also maintains a significant advantage, with a smaller range of data fluctuation. In the mild myopia group, the gap between the two methods narrows, but the method of the present invention still maintains a slight advantage. The experimental results show that the present invention, by adopting the refractive state influence function, can dynamically adjust the feature scaling ratio according to the user's refractive state, improve the feature representation of high myopia users, and solve the problem that conventional normalization ignores the nonlinear influence of refractive state on light sensitivity. The vertical axis "feature discrimination index" in the figure is a dimensionless index with no unit, ranging from 0 to 1. The closer it is to 1, the better the discrimination.

[0087] This invention provides a dynamic illumination prediction method based on refractive state, comprising:

[0088] Obtain the refractive data of the subject's eye;

[0089] The refractive data are clustered and feature fusion is performed to obtain fused features;

[0090] The fused features are input into the trained prediction model to obtain the brightness prediction value and the color temperature prediction value;

[0091] The prediction model initializes its weight parameters by compressing outliers in the refractive data and compensating for distribution biases in refractive clustering.

[0092] In one embodiment, the ocular refractive data is obtained based on the dynamic data processing method based on the ocular refractive state described above.

[0093] In one embodiment, the compressed outliers are compressed using a skewness-aware scaling matrix; the compensation for the distribution bias of refractive clustering is achieved by calculating a bias correction term using refractive clustering information, and then compensating using the bias correction term.

[0094] In an embodiment, the skewness-aware scaling matrix is obtained by calculating skewness values of each dimension of the fusion feature vector, calculating a scaling coefficient of each dimension based on the skewness value and a scaling adjustment hyperparameter, and then constructing the skewness-aware scaling matrix from the scaling coefficients.

[0095] In an embodiment, the bias correction term is calculated by:

[0096] calculating a global mean vector of the fusion feature vectors of all training samples;

[0097] calculating a difference between the centroid of each cluster and the global mean, and performing a weighted summation of the difference by the fusion weights of the clusters;

[0098] multiplying the weighted summation result by a bias learning rate hyperparameter to obtain the bias correction term.

[0099] In an embodiment, the weight parameter initialization is performed by calculating a mean vector of the fusion feature vectors of all samples, multiplying the mean vector by the skewness-aware scaling matrix, and adding the bias correction term to obtain the initialized weight matrix of the first layer of the neural network.

[0100] In an embodiment, the weight initialization is performed only for the first layer of the prediction model to compress the outliers of the refractive data and compensate for the distribution deviation of the refractive clusters, and random weight initialization is performed for other layers of the prediction model.

[0101] In an embodiment, the prediction model learns the feature relationship between the refractive sensitivity and the brightness through a refractive-aware adaptive activation function. The refractive-aware adaptive activation function is obtained by dynamically adjusting the steepness and threshold position of the activation curve based on the average spherical power.

[0102] In an embodiment, the prediction model includes an input layer, a feature extraction module, and a prediction module, and the fusion features sequentially pass through the input layer, the feature extraction module, and the prediction module. The prediction module includes a double-branch structure. The first branch linearly maps a deep feature vector through a weight vector and a global bias term to obtain a basic prediction value. The second branch generates a refractive dynamic modulation factor for brightness prediction and a refractive dynamic modulation factor for color temperature prediction based on the deep feature vector and the average spherical power. The brightness prediction value and the color temperature prediction value are obtained based on the basic prediction value and the refractive dynamic modulation factors.

[0103] In an embodiment, the clustering is performed based on the average spherical power in the refractive data to obtain grouped sample clusters, and a projection matrix is constructed to remove redundant features from the grouped sample clusters to obtain feature clusters. The feature fusion is performed based on the density of the feature clusters to calculate fusion weights, and the feature clusters with different fusion weights are fused to obtain fusion features.

[0104] In one embodiment, the construction of the projection matrix is to perform singular value decomposition on the feature vectors of the samples in the cluster to obtain a left singular matrix and a singular value matrix, and the projection matrix of the current cluster is calculated based on the left singular matrix and the singular value matrix.

[0105] Optionally, the fusion weight calculation is to take the average square distance of the samples in the cluster to the cluster center as the inverse of the cluster density, and the fusion weight of each cluster is calculated based on the density inverses of all clusters.

[0106] In one embodiment, the method further comprises behavior data; obtaining behavior data; inputting the behavior data and the fusion features into the trained first prediction model to obtain the brightness prediction value and the color temperature prediction value.

[0107] In one embodiment, the training process of the first prediction model is:

[0108] Obtaining the eye refraction data set, the environmental data set, the behavior data set and the label of the subject;

[0109] Inputting the eye refraction data set, the environmental data set, the behavior data set and the label into the prediction model for training, comparing the prediction value with the true value, calculating the loss, and iterating the prediction model until the loss function is constant, to obtain the trained prediction model.

[0110] In one embodiment, the behavior data includes the preferred brightness level and the preferred color temperature level set by the user, and the historical average brightness and color temperature automatically counted by the system.

[0111] In one embodiment, the network architecture of the prediction model is the same as that of the first prediction model, including but not limited to the same parameter initialization and loss function processing method, and the difference lies in the different input data, forming the prediction model and the first prediction model.

[0112] In one specific embodiment, the user's refraction state parameters, environmental light features and behavior context data are collected by multiple source heterogeneous devices, and the specific collection methods include:

[0113] The core refraction parameters such as spherical power, cylindrical power and axis position of the user's eyes are obtained by using an optometry instrument; the pupil diameter is recorded by a pupil instrument; the environmental brightness, environmental color temperature and sensor X / Y axis readings are captured in real time by combining with an environmental light sensor; the user's age, gender, use time, activity type, screen brightness and screen color temperature are extracted by using device logs; the axial length, corneal curvature and tear film quality index are obtained by using a biometric device.

[0114] At the same time, the preferred brightness level and the preferred color temperature level set by the user, and the historical average brightness and color temperature automatically counted by the system are recorded.

[0115] Further, the data is labeled, and the data labeling focuses on the target parameters of the dynamic lighting mode. According to the subjective comfort feedback of the user and the environmental adaptation demand, a professional visual engineer labels the target brightness value and the target color temperature value corresponding to each sample to form a label pair for supervised learning, and the labeling categories strictly distinguish the two continuous numerical labels of brightness and color temperature.

[0116] In one embodiment, the attributes and original value examples of the behavior data of a sample are: current time (14.5, time coding format, indicating that the time is 14:30), use duration (120, unit: minute), activity type (2, reading, no unit, One-hot coding format), screen brightness (200, unit: candela per square meter), screen color temperature (5000, unit: Kelvin), ambient light sensor X-axis reading (0.8, normalized value, no unit), ambient light sensor Y-axis reading (0.6, normalized value, no unit), user preferred brightness level (70, percentage value, unit: %), user preferred color temperature level (5500, unit: Kelvin), historical average brightness (180, unit: candela per square meter), historical average color temperature (4500, unit: Kelvin), axial length (24.2, unit: millimeter), corneal curvature (43.5, unit: diopter), tear film quality index (85, percentage value, unit: %).

[0117] The label of this piece of data is the dynamic lighting mode parameter, specifically: target brightness L_"target" = 250 (candela per square meter), target color temperature T_"target" = 4800 (Kelvin).

[0118] In one embodiment, the dynamic feature fusion is based on refraction clustering:

[0119] The pupil diameter of a myopic user has a strong interaction effect with the environmental brightness. When processing normalized data, the conventional principal component analysis method is prone to lose such key discriminant information, and cannot eliminate the redundancy of highly correlated features such as spherical power and cylindrical power.

[0120] The present application realizes dynamic grouping and weighted fusion through the feature fusion layer guided by the refraction state. First, the samples are grouped based on the average spherical power, then a projection matrix is constructed for each cluster to eliminate feature redundancy, and finally the feature is fused combined with the cluster density weight, which reduces the feature dimension while retaining the key discriminant information, enhances the ability to capture the nonlinear interaction effect related to the refraction state, and the specific steps are as follows:

[0121] 1) Refraction state clustering and cluster assignment;

[0122] Using K-means algorithm, clustering based on the average spherical power of all training samples, defaulting to 3 clusters, calculating the Euclidean distance of each sample to the cluster center, and assigning the sample to the nearest cluster, determining the refractive state group to which each sample belongs, denoted as:

[0123]

[0124] In the formula, represents the number of clusters, and the average spherical power is clustered by K-means algorithm ;

[0125] represents the cluster index of the th sample, which identifies the cluster to which the sample belongs; represents the center of the cth cluster, that is, the cluster center, which is specifically calculated by iterative optimization of K-means algorithm, including initialization of random center, assignment of sample to nearest cluster, and update of center to average spherical power of cluster; is the cluster index; is the average spherical power of the th sample; is the L2 norm; represents the cluster index that minimizes the Euclidean distance .

[0126] where C=3 is an empirical default value, corresponding to common refractive state grouping, i.e. mild, moderate, and high myopia, and applied to the average spherical power of all training samples by K-means algorithm to automatically determine the cluster center.

[0127] 2) Cluster projection matrix and weight calculation:

[0128] For each cluster, use the normalized feature vector of the samples in the cluster to perform singular value decomposition, use the obtained left singular vector matrix and singular value matrix to calculate the projection matrix of the cluster, at the same time, calculate the average square distance of the samples in the cluster to the cluster center as the reciprocal of the cluster density, and based on the density reciprocal of all clusters, calculate the fusion weight of each cluster, obtain the cluster projection matrix for dimensionality reduction and redundancy elimination, and the fusion weight reflecting the importance of the cluster, denoted as:

[0129]

[0130]

[0131] In the formula, represents the projection matrix of the cth cluster, which is used to compress the feature dimension and retain the main variation within the cluster;

[0132] This represents the left singular vector matrix obtained by singular value decomposition of the normalized feature matrix of the sample in the c-th cluster; Let represent the singular value matrix of the c-th cluster. It is a diagonal matrix whose diagonal elements are singular values. Represents the singular value matrix Take each diagonal element The power of the power is the reciprocal square root; This represents the fusion weight of the c-th cluster; the larger the value, the greater the contribution of that cluster to the fusion result. This represents the weight adjustment factor, which controls the sensitivity of the density to the weights, such as... ; This represents the reciprocal of the sample density of the c-th cluster, which is the average squared distance from the sample to the cluster center. It is calculated as follows: ; This represents the reciprocal of the sample density of the k-th cluster; k is the cluster index that distinguishes it from c. This represents the number of samples in the c-th cluster; Indicates the first The normalized feature vector of each sample, where all feature values ​​are normalized values; For sample index; This represents the natural exponential function.

[0133] The term is used to calculate the first... The projection matrix of the cluster is used to compress the feature dimension, achieving dimensionality reduction while preserving the main variations within the cluster. It is the left singular vector of the singular value decomposition. Scaling singular values ​​balances dimensional contributions.

[0134] 3) Dynamic feature fusion:

[0135] For each sample, its normalized feature vector is transformed using the projection matrix of its cluster. Then, the transformation results of all clusters are weighted and summed according to their respective fusion weights to generate a fusion feature vector. This process fuses the feature information of each cluster, focusing on preserving key variations and nonlinear interaction effects within the refractive state group to which the sample belongs, thereby reducing the feature dimensionality. This is represented as:

[0136]

[0137] In the formula, Indicates the first The fusion feature vector of each sample is the computational dimension of a single sample, representing a compact representation of the adaptive refractive state.

[0138] The fusion feature vector uses the cluster to which the sample belongs. and The weighted sum combines the intra-cluster variance preserving and density weight, and adapts the modulation of the refractive state to the feature interaction, such as the nonlinear coupling of the ambient brightness and the pupil diameter of the myopic user, so as to make the generated fusion feature vector Reduce the feature dimension while preserving the key discriminative information.

[0139] It should be further pointed out that the first The sample is based on The projection matrix of the selected cluster And the fusion weight Realize the feature fusion adaptive to the refractive state.

[0140] In one embodiment, the conventional principal component analysis ignores the modulation of the refractive state to the feature correlation, such as the more sensitive pupil diameter of the myopic user to the ambient brightness, and the present application captures the local nonlinearity through clustering grouping and uses the projection matrix Eliminate redundancy, such as the spherical and cylindrical correlation redundant characteristics, while cooperating with the density weight Solve the data imbalance problem, give higher weight to the sample dense cluster, and then improve the expression ability of the fusion feature on the highly correlated features such as spherical and cylindrical degrees, and reduce the generalization error of the sparse cluster such as extreme ametropia.

[0141] In one specific embodiment, a deep neural network (prediction model or first prediction model) for dynamic lighting parameter prediction is constructed:

[0142] A fully connected deep neural network is constructed as the core prediction architecture, specifically including an input layer, a feature extraction module, and a dynamic lighting parameter prediction module.

[0143] The input layer dimension of the neural network is consistent with the fusion feature vector dimension, and receives the feature representation after refractive state adaptive processing.

[0144] The feature extraction module is the hidden layer structure of the neural network, which adopts a three-layer fully connected structure, and each layer adopts batch normalization operation to stabilize the training process, specifically including:

[0145] The first hidden layer contains 128 neurons, which realizes the primary abstraction of the environment-refractive interaction feature through the refractive perception adaptive activation function;

[0146] The second hidden layer is reduced to 64 neurons, which strengthens the non-linear feature transformation ability;

[0147] The third hidden layer is further compressed to 32 neurons, which extracts high-level discriminative patterns.

[0148] The dynamic lighting parameter prediction module is an output layer, adopts a double-branch structure, the first branch outputs a brightness prediction value through a linear full connection layer, and the second branch independently outputs a color temperature prediction value.

[0149] The network as a whole processes data in a forward propagation manner, and the fused feature vector is sequentially subjected to linear transformation and adaptive activation processing by each layer, and finally generates brightness and color temperature prediction values by the output layer.

[0150] Adaptive weight initialization of the deep neural network:

[0151] Xavier initialization and other conventional random initialization methods are sensitive to the skewness distribution of refractive data and extreme spherical power values, and are prone to cause unstable neural network training process and slow convergence.

[0152] The adaptive weight initialization mechanism is adopted, the influence of abnormal values is compressed through the skewness perception scaling matrix, the bias correction term is calculated based on the refractive clustering information to compensate for the distribution deviation, the initialization weight is generated by integrating the feature statistics and the correction term, and then the robustness of the model to the distribution characteristics of the refractive data is improved, the training convergence is accelerated, and the specific steps are as follows:

[0153] 1) Calculate the adaptive scaling matrix:

[0154] The skewness statistics of each dimension of the fused feature vector are calculated, and based on these skewness values and scaling adjustment hyperparameters, the scaling coefficients corresponding to each dimension are calculated, and then the adaptive scaling matrix is formed by these scaling coefficients, the influence of abnormal values with significant skewness distribution characteristics is compressed, and the initialization scale of each dimension is balanced, which is represented as:

[0155]

[0156] In the formula, The adaptive scaling matrix is used to adjust the weight initialization scale; The diagonal matrix generation function is represented as: is the feature dimension index, ; is the total dimension of the fused features; is the scaling coefficient of the first dimension, is the scaling coefficient of the first dimension, , correspondingly, is the scaling coefficient of the first dimension, is the scaling coefficient of the Dth dimension; is the scaling coefficient of the first dimension, is the skewness statistics of the first feature, and the spherical power is the main factor, because the far vision type of extreme positive value is less but the amplitude is large, so the refractive data is usually right skewed, that is, positive skewness, which causes the distribution to be dragged to the right side; is the scaling adjustment hyperparameter, which controls the compression strength of abnormal values, for example, .

[0157] 2) Calculate the bias correction term:

[0158] Calculate the global mean vector of all training sample fusion feature vectors, then calculate the difference between the centroid of each cluster and the global mean, and use the cluster fusion weight to weight the sum of the differences, and finally multiply the weighted sum result by the bias learning rate hyperparameter to obtain the bias correction term, which compensates for the initialization bias caused by the uneven distribution of refractive state data, represented as:

[0159]

[0160] wherein, represents the bias correction term, which is used to offset the clustering bias;

[0161] is the global sample mean vector, and the calculation method is represented as ; is the total number of training samples; is the bias learning rate hyperparameter, which controls the correction strength, such as, .

[0162] 3) Generate the initialization weight matrix:

[0163] Calculate the mean of all training sample fusion feature vectors, then multiply the mean vector by the adaptive scaling matrix (skewness-aware scaling matrix), and add the bias correction term to obtain the initialization weight matrix of the first layer of neural network, which integrates the overall information of fusion features, scaling adjustment of feature skewness, and bias compensation of data distribution imbalance, provides robust initial parameters for the first layer of neural network, represented as:

[0164]

[0165] wherein, represents the initialization weight matrix of the first layer of neural network;

[0166] It should be noted that, in order to handle the skewness distribution of refractive data, this initialization calculation method is only used for the first layer, and the weight matrix of other layers is initialized using the standard Xavier method.

[0167] It should also be noted that the initialization method of the initialization weight matrix of the first layer of neural network integrates the feature mean, adaptive scaling matrix and bias correction term , which especially enhances the prediction stability of abnormal samples such as high hyperopia samples.

[0168] In one specific embodiment, a feature extraction module based on an adaptive activation function is constructed:

[0169] High myopia users need stronger non-linear expression ability in low light environment, and the conventional ReLU activation function cannot dynamically respond to the change of refractive state, which easily leads to difficulty in capturing the complex interaction mode between light and refractive state by deep features.

[0170] The present application adopts a refractive perception adaptive activation function, dynamically adjusts the non-linear response characteristics by using the average spherical power, enhances the expression ability of the neural network to the sensitive features of the refractive state by modulating the steepness and threshold position of the activation curve, optimizes the modeling effect of the light-refractive interaction mode, and the specific steps are as follows:

[0171] 1) Perform linear transformation calculation:

[0172] The output feature vector of the previous layer is linearly transformed using the weight matrix and bias vector of the current layer to obtain the input value of the activation function, which is represented as:

[0173]

[0174] In the formula, represents the input value of the activation function, representing the linear transformation result; is the weight matrix of the lth layer of the neural network, which is a trainable parameter; is the layer index of the neural network; is the output feature vector of the l-1th layer of the neural network; is the bias vector of the lth layer of the neural network, which is a trainable parameter.

[0175] It should be noted that the input feature vector of the first layer of the neural network is , represents the fusion feature vector of the sample, which is the calculation dimension of a single sample.

[0176] 2) Apply the refractive perception activation function:

[0177] The linear transformation result is input into the adaptive activation function, which dynamically adjusts the non-linear response characteristics of the user's average spherical power, calculates the output feature vector of the current layer, so that the neural network can flexibly enhance or suppress the non-linear expression ability of the features according to the refractive state of the user, better capture the complex interaction mode between light and refractive state, represented as:

[0178]

[0179]

[0180] In the formula, represents the output feature vector of the lth layer of the neural network, which captures the deep mode of the interaction between the refractive state and the environment; An adaptive activation function; As a nonlinear steepness control parameter, it determines the slope of the function curve, such as... ; The threshold offset parameter controls the threshold position of the adaptive activation function, adjusting the starting point of the nonlinear response to adapt to different refractive states. Highly myopic users require a lower activation threshold, such as... .

[0181] It should be noted that the adaptive activation function, combined with the mean spherical power, dynamically modulates the nonlinear steepness control parameters. A low average spherical power indicates high myopia. The coefficient increases, enhancing the nonlinear expression capability under low light and improving the ability of deep features to capture the light-refractive interaction.

[0182] Furthermore, the data is processed through forward propagation using a fully connected neural network, followed by activation via a layer-by-layer linear transformation activation function until the last layer of the fully connected neural network. The output feature vector of the last layer of the fully connected neural network is defined as... .

[0183] In one specific embodiment, a dynamic illumination parameter prediction module based on refractive perception is constructed:

[0184] Although the deep features after feature extraction contain environmental-refractive interaction information, direct linear mapping cannot adapt to the heterogeneous influence of refractive state on brightness and color temperature. Highly myopic users prefer low brightness to alleviate glare sensitivity, and conventional fully connected regression methods ignore the nonlinear modulation effect of refractive state in the output stage.

[0185] This invention employs a dual-path structure to predict target brightness and color temperature. The main path linearly maps the deep feature vector through a weight vector and a global bias term to obtain the basic predicted value. The modulation path, based on the deep feature vector and the average spherical power, calculates the refractive dynamic modulation factor for brightness prediction and the refractive dynamic modulation factor for color temperature prediction through refractive influence function scaling, feature projection, and interaction terms. The final predicted value is obtained by adding the product of the corresponding modulation factor and its fusion coefficient to the basic predicted value. An offset dynamically modulated by the refractive state is superimposed on the basic prediction, prioritizing the preference of highly myopic users for brightness reduction and color temperature adjustment, capturing the interaction effect between refractive state and deep features, expressed as:

[0186]

[0187] In the formula, Indicates the first The predicted brightness value for each sample, in candela per square meter; Indicates the first Predicted color temperature value of the i-th sample in Kelvin; is a trainable parameter, represents the weight vector for the luminance prediction; is the transpose of ; is a trainable parameter, represents the global bias term for the luminance prediction; is a trainable parameter, represents the weight vector for the color temperature prediction; is the transpose of ; is a trainable parameter, represents the global bias term for the color temperature prediction; is the fusion coefficient for the luminance modulation path, controlling the contribution intensity of , such as ; is the fusion coefficient for the color temperature modulation path, controlling the contribution intensity of , such as ; is the refractive dynamic modulation factor for the luminance prediction of the i-th sample, calculated based on and , enhancing the luminance down-regulation for highly myopic samples, and the calculation method is represented as ; ;

[0188] is the refractive dynamic modulation factor for the color temperature prediction, calculated based on and , dynamically adjusting the color temperature prediction value to respond to the nonlinear modulation effect of the refractive state on the color temperature perception, and the calculation method is represented as ;

[0189] is the luminance refractive influence function, and the calculation method is represented as ;

[0190] is the luminance modulation amplitude hyperparameter, controlling the function output range, such as ;

[0191] is the luminance modulation curvature hyperparameter, controlling the function sensitivity, such as ;

[0192] is the color temperature refractive influence function, and the calculation method is ;

[0193] is the color temperature modulation amplitude hyperparameter (such as ), which is a trainable parameter;

[0194] denote the color temperature modulation curvature super-parameter (e.g. ), which is a trainable parameter;

[0195] denote the luminance modulation weight vector, is the transpose of , used to project to generate the base offset, extract the feature subspace related to luminance modulation, which is a trainable parameter;

[0196] denote the color temperature modulation weight vector, is the transpose of , used to project to generate the base offset, extract the feature subspace related to color temperature modulation, which is a trainable parameter;

[0197] denote the luminance interaction weight vector, is the transpose of , used to extract the interaction weight vector for color temperature prediction from the deep feature vector , which is a trainable parameter;

[0198] denote the color temperature interaction weight vector, is the transpose of , used to extract the interaction weight vector for luminance prediction from the deep feature vector , which is a trainable parameter;

[0199] denote the luminance interaction gain coefficient, which controls the gain coefficient of the interaction term in the luminance prediction path, e.g., . ;

[0200] denote the color temperature interaction gain coefficient, which controls the gain coefficient of the interaction term in the color temperature prediction path, e.g., . .

[0201] It should be noted that and characterize the main path calculation, providing the basic prediction, and characterize the modulation path calculation, dynamically scaling the modulation amplitude based on , e.g., when , , thereby amplifying the luminance adjustment of highly myopic users. In addition, in the dual-path structure, the main path ensures global consistency, the modulation path combines local refractive dependence, and the bias correction term cooperates to improve the robustness to extreme refractive samples.

[0202] It should also be noted that, As deep features, they contain high-dimensional environment-refractive interaction information, such as brightness, color temperature, and pupil diameter, but not all dimensions are equally modulated by refractive states. Item and The two projection operations are performed using a trainable weight matrix. Mapping to a low-dimensional subspace preserves only features strongly correlated with refractive state, such as the ambient brightness sensitivity of highly myopic users, thereby reducing the modulation factor. and This enables efficient and targeted refractive adaptation, avoids noise interference, and improves prediction robustness.

[0203] It should also be noted that, Item and Item implementation and The interaction enhances nonlinearity, such as in high myopia. Increasing the negative value leads to negative saturation of the output, reducing its efficiency. At the same time, combined and accomplish and explicit coupling, thereby capturing the fused feature vector The nonlinear interaction characteristics emphasized in the text, such as the interaction effect between ambient brightness and pupil diameter in patients with high myopia.

[0204] It should also be noted that, Item and The term represents the weighted fusion modulation term, prioritizing optimization of highly myopic samples, such as... It exhibits higher brightness sensitivity, through and Dynamically adjust forecast values ​​and respond directly. , and based on the influence function of refractive state The normalized scaling operation and the adaptive activation operation based on the adaptive activation function form an end-to-end refractive adaptation chain.

[0205] In one embodiment, the prediction model dynamically adjusts the loss weights of brightness and color temperature prediction errors using a refractive state-weighted loss function and a weighting factor calculated using the average spherical power.

[0206] In one specific embodiment, highly myopic users are more sensitive to brightness errors, but conventional loss functions cannot prioritize optimizing such key parameters, and conventional mean squared error loss functions do not consider the heterogeneous impact of refractive state on prediction errors.

[0207] This invention employs a refractive state-weighted loss function, dynamically adjusting the loss weights for brightness and color temperature prediction errors through weighting factors calculated based on the average spherical power. This allows the model optimization process to focus on improving the accuracy of brightness prediction for highly myopic users. The specific steps are as follows:

[0208] 1) Calculate the refractive weighting factor:

[0209] Based on the user's average spherical power, the refractive weighting factor is calculated using a mapping function in the form of a sigmoid activation function, expressed as:

[0210]

[0211] In the formula, Indicates the first The refractive weighting factor for each sample is used to adjust the weight of the luminance error term;

[0212] As a slope control parameter, it adjusts the sensitivity to changes in weights, such as... .

[0213] It should be noted that the refractive weighting factor The calculation method is similar to the Sigmoid activation function, which can map the mean spherical power to... To ensure the high myopia sample Approaching 1 improves the accuracy of brightness prediction for highly myopic users.

[0214] 2) Calculate the refractive weighted loss value:

[0215] The squares of the brightness prediction error and color temperature prediction error are calculated. These errors are then weighted using a refractive weighting factor. The weighted error terms for all samples are then averaged to obtain the final loss value. This allows the loss function to automatically increase the accuracy requirement for brightness prediction based on the user's refractive state during the optimization process, especially prioritizing brightness prediction for highly myopic users. This is expressed as:

[0216]

[0217] In the formula, Represents the loss function;

[0218] For the first The predicted brightness value for each sample;

[0219] For the first The target brightness value for each sample;

[0220] For the first a predicted color temperature value of the sample;

[0221] a color temperature target value of the sample.

[0222] It should be noted that, in the calculation process of the refractive weight factor, the refractive weight factor helps the loss function to preferentially optimize the brightness prediction, characterizes high myopia, and the high myopia user is more sensitive to brightness error, for example, under the premise of setting when , when , , thereby increasing the weight of the item and preferentially optimizing the brightness prediction.

[0223] It should also be noted that the item characterizes the color temperature error weight item, when increases, the brightness weight is high, decreases, the loss function needs to keep the weight and constant as 1, and ensures the optimization balance.

[0224] In one specific embodiment, the parameter update is performed based on the refractive perception gradient optimization algorithm:

[0225] The standard Adam optimizer ignores the modulation effect of the refractive state on the parameter update, which is easy to cause the convergence speed of the high refractive error sample to be slow and easy to fall into a local optimal solution.

[0226] The present application amplifies the parameter update step by the average spherical power proportion according to the refractive perception dynamic learning rate mechanism, accelerates the convergence of the related parameters of the high refractive error sample, improves the model training efficiency, and the specific steps are as follows:

[0227] 1) Calculate the refractive perception learning rate:

[0228] Based on the absolute value of the average spherical power of the user, the basic learning rate is amplified in proportion to obtain the dynamic learning rate of the refractive perception, so that the model adopts a larger parameter update step in the optimization process for the user sample with more serious refractive degree, which is expressed as:

[0229]

[0230] In the formula, the dynamic learning rate of the refractive perception is used to adjust the parameter update step;

[0231] is the basic refractive perception dynamic learning rate, which controls the overall optimization speed, for example, ; and ​

[0232] is the refractive adjustment coefficient, which determines the influence intensity of the refractive state on the learning rate, such as, ;

[0233] is the absolute value of the average spherical degree.

[0234] It should be noted that under the premise of setting , when , increases by 12%, because the high hyperopia sample parameters need a larger update step to accelerate convergence, and continue to adjust can make when ,

[0235] 2) Perform refractive perception parameter update:

[0236] Using the improved Adam optimization rule, the first moment estimate and the second moment estimate after bias correction are calculated using the Adam optimizer, then the dynamic learning rate of refractive perception is used to scale the standard update rule of Adam, update the parameters of the neural network, and accelerate the convergence speed of the high hyperopia sample related parameters, which is expressed as:

[0237]

[0238] In the formula, represents the parameter value of the th iteration, which specifically refers to the trainable parameters of the neural network, including all trainable parameters of the neural network;

[0239] represents the parameter value of the th iteration;

[0240] is the first moment estimate vector of the Adam optimizer, representing the mean of the gradient after bias correction;

[0241] is the second moment estimate vector of the Adam optimizer, representing the variance of the gradient after bias correction, represents the element-wise square root of ;

[0242] is a numerical stability constant to prevent division by zero errors, such as, ;

[0243] represents the unit matrix, achieves the construction of a vector with the same dimension as .

[0244] In one specific embodiment, iterative training and early stopping strategy:

[0245] The model training employs a mini-batch gradient descent strategy, combined with a refractive sensing gradient optimization algorithm to dynamically update network parameters. Each iteration consists of three steps, specifically:

[0246] Forward propagation calculates the predicted brightness and color temperature for the current batch;

[0247] Call the loss function to calculate the batch loss value;

[0248] Backpropagation is used to calculate the gradient and update the refractive sensing parameters.

[0249] The performance of the validation set is continuously monitored during the training process. After each training cycle, the validation set loss is calculated and the optimal loss value is recorded.

[0250] If the verification loss does not decrease for 10 consecutive periods, that is, the current loss is higher than the historical best value, the early stopping mechanism is triggered, and the model parameters before that period are saved as the final model.

[0251] At the same time, set a maximum training cycle limit, such as 2000 cycles, to prevent infinite iteration.

[0252] Early stopping conditions and maximum period limits work together to prevent overfitting and ensure that the model terminates training when the performance on the validation set peaks.

[0253] This invention provides a dynamic lighting adjustment method based on refractive state adaptation, comprising: a light source controller and a lighting source;

[0254] The luminance prediction value and color temperature prediction value are obtained based on the above dynamic illumination prediction method based on refractive state.

[0255] The predicted brightness and color temperature values ​​are input to the light source controller;

[0256] The brightness and color temperature of the lighting source are dynamically adjusted by the light source controller.

[0257] In one embodiment, the light source controller controls the lighting source through instructions, computer programs, or signals, including one or more of the following: wireless Bluetooth control, wired control, and local area network control.

[0258] In one specific embodiment, dynamic lighting parameter prediction and lighting adjustment:

[0259] 1) Real-time data preprocessing: Collect new user raw data in the required format and perform adaptive normalization scaling based on the dynamic data processing method of eye refractive state;

[0260] 2) Feature fusion: based on the pre-stored clustering centroid and projection matrix, the refractive cluster index of the user is calculated, and the corresponding projection matrix and fusion weight are applied to generate a fusion feature vector;

[0261] 3) Neural network prediction: the fusion feature is input into the trained deep neural network, and the luminance prediction value and color temperature prediction value are output through forward propagation.

[0262] Further, based on the prediction result, the lighting is adjusted, specifically, the prediction parameters are sent to the lighting controller to drive the LED light source to dynamically adjust the output luminous flux and color temperature, and the system continuously monitors the changes of the ambient light and user feedback. When the environmental sensor data or user activity type is updated, the above process is triggered again to realize closed-loop adaptive adjustment.

[0263] In one embodiment, the prediction accuracy under different refractive states is compared and analyzed to evaluate the prediction performance of the present technology in different refractive state user groups. By comparing the prediction error distribution of five types of users (high myopia: average spherical power < -6 diopters; moderate myopia: -6 ~ -3 diopters; mild myopia: -3 ~ 0 diopters; emmetropia: 0 ~ +1 diopters; hyperopia: > +1 diopters) when using the present technology, the advantages of the present invention in adaptive processing of refractive states are verified, especially the optimization effect for high myopia users. The error distribution of five types of refractive state users (high myopia: average spherical power < -6 diopters; moderate myopia: -6 ~ -3 diopters; mild myopia: -3 ~ 0 diopters; emmetropia: 0 ~ +1 diopters; hyperopia: > +1 diopters) in the two key indicators of luminance prediction and color temperature prediction is compared. The unit of luminance prediction error is candela per square meter (measuring light intensity), and the unit of color temperature prediction error is Kelvin (measuring light source color characteristics). The violin plot is used to show the distribution characteristics of the prediction error, and the overall distribution range, density and key statistics of the data are also shown. In the luminance prediction error distribution graph, as shown in Figure 5 The experimental data shows that the prediction error distribution of high myopia users is the most concentrated (the violin shape is the narrowest), the error median is the lowest, and as the degree of myopia decreases, the error distribution range gradually widens (the violin shape becomes wider), and the error median gradually increases. The error distribution characteristics of hyperopia users are similar to those of mild myopia users, and the error distribution is positively skewed (the upper tail is longer), indicating that there are a small number of large prediction errors. In the color temperature prediction error distribution graph, as shown in Figure 6As shown, the abscissa is also five types of refractive state user groups, and the ordinate is the color temperature prediction error value (unit: Kelvin), and the distribution trend is similar to the brightness prediction. The error distribution of high myopia users is the most concentrated and the median is the lowest. The error distribution range of emmetropia users is the widest and the median is the highest. The error distribution of moderate myopia and mild myopia users is between the two. The error distribution of all groups is positively skewed. The experimental results show that the error of high myopia users in brightness and color temperature prediction is the smallest and most stable, which verifies the targeted optimization effect of key technologies such as refractive state influence function, adaptive activation function and dynamic feature fusion on high myopia users. The violin shape of the high myopia group in the figure is obviously narrower and the median line is lower, which intuitively shows this advantage. In addition, the error distribution changes regularly and gradiently with the change of refractive state, indicating that the present application can dynamically adjust the model parameters and processing strategy according to the refractive state of the user, especially the obvious performance difference from high myopia to moderate myopia, which reflects the sensitivity of the technology to the subtle changes of refractive state.

[0264] In one specific embodiment, the present application analyzes the nonlinear influence of refractive state on light perception, explores the complex nonlinear relationship between refractive state (quantified by average spherical degree) and light perception characteristics, and verifies the superiority of the present application in capturing these relationships compared with traditional methods. Three key indicators are investigated: light sensitivity factor, preferred brightness value and light change response time. The experiment is divided into three parts: the relationship between refractive state and light sensitivity factor, the relationship between refractive state and preferred brightness value, and the comparison between traditional methods and the present application in the influence of refractive state on response time, such as Figure 7 、 Figure 8 、 Figure 9The horizontal coordinates of all the graphs are the average spherical power (unit: diopter), and the vertical coordinates are, respectively, the light sensitivity factor (dimensionless relative value), the preferred brightness value (unit: candela per square meter), and the response time (unit: millisecond). In the light sensitivity factor analysis, the scatter plot shows the relationship between the diopter and the sensitivity of each sample, and the color depth represents the sample density (the dark color area has a high sample density). The red dashed trend line shows an obvious S-shaped curve characteristic. When the diopter is less than -4 diopter, the sensitivity factor significantly increases, and when the diopter is greater than 0 diopter, the sensitivity factor tends to be stable. In the preferred brightness value analysis, the scatter distribution shows that the diopter is negatively correlated with the preferred brightness, and the purple dashed trend line shows an inverse S-shaped characteristic. The preferred brightness value of the highly myopic user (diopter less than -6) is obviously lower than that of other groups, and the sample density shows that most users have a diopter in the range of -8 to +2 diopter. In the response time comparison, the blue scatter points and the trend line represent the traditional method, and the orange scatter points and the trend line represent the present application. Both the two technologies show that the response time is shortened when the diopter decreases (myopia deepens). The experimental results show that the S-shaped trend line perfectly verifies the rationality of the design of the diopter state influence function (based on the hyperbolic tangent function) proposed in the technical document, and proves that the present application can accurately model the complex nonlinear relationship between the diopter state and the light perception. In addition, the present application is superior to the traditional method in the response time index, especially in the highly myopic area (diopter less than -6), which verifies the effectiveness of the diopter perception gradient optimization algorithm and the adaptive activation function, and shows that the present application can dynamically adjust the learning rate and the nonlinear response characteristics according to the diopter state of the user.

[0265] The present application also discloses a computer program product or system, comprising a computer program which, when executed by a processor, implements the steps of the above-mentioned dynamic data processing method based on the eye diopter state.

[0266] Figure 2 The present application provides a dynamic data processing system based on the eye diopter state, which specifically comprises:

[0267] The acquisition unit acquires the initial diopter data of the eyes of the to-be-tested person and environmental data, including environmental brightness and environmental color temperature.

[0268] The processing unit pre-processes the initial diopter data of the eyes and the environmental data to obtain pre-processed data.

[0269] The calculation unit calculates the diopter data of the eyes from the pre-processed data by using a diopter state influence function.

[0270] The present application provides a dynamic lighting prediction system based on the diopter state, which comprises:

[0271] an acquisition module configured to acquire eye refraction data of a subject;

[0272] a feature module configured to cluster and fuse features of the refraction data to obtain fused features;

[0273] a prediction module configured to input the fused features into a trained prediction model to obtain luminance prediction values and color temperature prediction values;

[0274] the prediction model is initialized by compressing abnormal values of the refraction data and compensating distribution deviation of refraction clusters.

[0275] An embodiment of the present application provides a dynamic lighting adjustment system based on an adaptive refraction state, comprising:

[0276] a light source controller and a lighting source;

[0277] a prediction module configured to obtain luminance prediction values and color temperature prediction values according to the dynamic lighting prediction method based on the refraction state;

[0278] a transmission module configured to input the luminance prediction values and the color temperature prediction values into the light source controller;

[0279] a control module configured to dynamically adjust luminance and color temperature of the lighting source through the light source controller.

[0280] Figure 3 An embodiment of the present application provides a computer device schematic diagram, specifically comprising:

[0281] a memory and a processor; the memory is configured to store program instructions; the processor is configured to call the program instructions, and when the program instructions are executed, any one of the dynamic data processing methods based on the eye refraction state is executed.

[0282] The present application discloses an embodiment of a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program is executed by a processor to execute any one of the dynamic data processing methods based on the eye refraction state.

[0283] The verification result of the verification embodiment shows that assigning inherent weights to the indications can improve the performance of the method compared with the default setting. It can be clearly understood by those skilled in the art that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be described here. In the several embodiments provided by the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of the units is only a logical function division. There can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms. The units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e. can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme. In addition, each functional unit in the various embodiments of the present application can be integrated in one processing unit, or each unit can be a physically independent unit, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware, or in the form of software functional units. Those skilled in the art can understand that all or part of the steps of the various methods in the above embodiments can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable storage medium, which can include read only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc.

[0284] Those skilled in the art can understand that all or part of the steps of the above-mentioned embodiment methods can be completed by programs instructing relevant hardware, and the programs can be stored in a computer readable storage medium, and the above-mentioned medium storage can be read only memory, magnetic disk or optical disk, etc.

[0285] The computer device provided by the present application has been described in detail above. For those skilled in the art, according to the idea of the embodiment of the present application, there will be changes in specific implementation and application range. In view of the above, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A dynamic data processing method based on the refractive state of the eye, characterized in that, The method comprises the following steps: obtaining initial refraction data of the eye of a subject and environmental data; preprocessing the initial refraction data of the eye and the environmental data to obtain preprocessed data; calculating the refraction data of the eye from the preprocessed data by using a refraction state influence function; wherein the refraction state influence function is obtained by calculating the dynamic change relationship between the initial refraction data of the eye and the environmental data; the dynamic change relationship of the refraction state influence function is obtained based on the nonlinear response of the individual refraction state to the light sensitivity; the calculation formula of the refraction state influence function is as follows: wherein, adjusting the hyperparameters for the first refractive state controls the magnitude of the scaling factor, adjusting the hyperparameters for the second refractive state controls the curvature of the scaling factor, denotes the mean sphere power, is the calculated dimension of a single sample.

2. A computer program product comprising a computer program or instructions embodied thereon, characterized in that, the computer program or instruction is executed by the processor to realize the dynamic data processing method based on the refraction state of the eye according to claim 1.

3. A computer device comprising a memory, a processor, and a computer program or instructions stored on the memory, wherein, the computer program or instruction is executed by the processor to realize the dynamic data processing method based on the refraction state of the eye according to claim 1.

4. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, the computer program or instruction is executed by the processor to realize the dynamic data processing method based on the refraction state of the eye according to claim 1.

Images