Eye vision light health management system

Abstract

The invention discloses an eye vision light health management system, and relates to the technical field of vision health management systems, the system specifically comprises data management and service management, the data management comprises a data acquisition module, a data processing module, a data identification module and a data transmission module, and the service management comprises tour service, store arrival service and remote service. According to the eye vision light health management system, through cooperation of the data acquisition module, the data processing module, the data identification module and the data transmission module, a detection mode of integrating intelligent wearable equipment and an eye vision light detection instrument to detect key indexes of vision data, intraocular pressure and diopter of a patient is adopted, a single detection mode is replaced, and meanwhile, the detection efficiency is improved. The integration of the intelligent wearable device and the eyesight light detection instrument also saves the step of manual recording of the detection data by the staff, thereby improving the accuracy of the eyesight detection data of the patient.

Description

Technical Field

[0001] This invention relates to the field of visual health management system technology, and in particular to an eye health management system. Background Technology

[0002] An eye health management system collects and statistically analyzes patient data, provides standard health data values ​​for comparison, allowing people to intuitively see their own vision, make improvements, and then perform further testing and comparisons to previous values ​​for further improvement. Its emergence makes it easier for people to see their own problems and make timely improvements.

[0003] However, in the field of optometry health management, existing technologies mostly rely on manual recording and analysis of customer information, which is inefficient and prone to errors. Health data collection depends on a single testing method, resulting in limited coverage. Data analysis capabilities are limited by traditional statistical methods, making it difficult to accurately predict vision change trends. When using sub-tables or simple database systems to manage customer information, and using basic ophthalmological examination equipment for data collection, followed by data analysis based on doctors' experience, the manual recording of data by staff requires a certain level of neatness in handwriting. Illegible handwriting is not convenient for others to refer to, and manual recording is also prone to errors in data transcription, which inconveniences the assessment of patients' vision. Furthermore, using a single vision testing method leads to incomplete data collection and inaccurate analysis, making it impossible to effectively predict the risk of vision decline. It also lacks personalized health management recommendations, affecting the service quality and efficiency of optometry centers. Summary of the Invention

[0004] This invention provides an eye health management system that solves the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an eye health management system, comprising data management and service management. The data management system includes a data acquisition module, a data processing module, a data recognition module, and a data transmission module. Through data exchange between these modules, an integrated smart wearable device and an eye examination instrument are used to detect key indicators of patient vision, intraocular pressure, and refractive error. This avoids relying solely on a single method of vision testing. Furthermore, the integration of the smart wearable device and the eye examination instrument eliminates the need for manual data recording by staff, preventing errors such as illegible handwriting or transcription mistakes, thereby improving the accuracy of patient vision test data. The eye health management system supports multi-terminal access, ensuring that users can access and update their personal historical data regardless of their location. The service management system includes mobile services, in-store services, and remote services. The data acquisition module specifically refers to the integration of smart wearable devices and optometry testing instruments to collect customer visual acuity, intraocular pressure, and refractive error indicators. The data processing module comprises four layers: the first layer uses edge computing technology to preprocess the raw data, removing noise and outliers; the second layer uses a deep learning model to analyze long-term vision change trends; and the third layer combines personalized characteristics of genetics, environment, and lifestyle to generate customized health management plans. The data identification module specifically refers to: using machine learning algorithms to label massive amounts of customer data, automatically identifying customer groups at risk of eye diseases, and issuing early warnings to doctors; The data transmission module specifically refers to: matching the received call request, separating structured information from image information, wherein the structured information includes patient ID and shooting date, and compressing and encrypting the separated image information using a custom encryption algorithm, then securely transmitting it to the cloud server via SSL protocol, and finally being parsed by professional software to generate a diagnostic report for doctors' reference.

[0006] The SSL protocol establishes an encrypted channel between the client and the server, ensuring the confidentiality, integrity, and authentication of data transmission. Confidentiality is achieved through symmetric encryption of data, with the key negotiated via asymmetric encryption, thus preventing eavesdropping. Integrity is ensured by using MAC (Message Authentication Code) or HMAC to prevent data tampering, thus preventing tampering. Authentication is achieved by the server proving its identity through a digital certificate (issued by a CA), with optional client certificate verification, thus preventing impersonation. Furthermore, the SSL protocol protects patient privacy information.

[0007] Optionally, the integrated smart device is equipped with several sensors, including but not limited to an accelerometer for detecting eye movements, an optical sensor for real-time monitoring of intraocular pressure, and a wireless communication module for data transmission; the optometry testing instrument includes but is not limited to an optometer and a fundus camera.

[0008] Optionally, the edge computing technology includes a curvature filtering method and an elliptical trajectory algorithm. The curvature filtering method is used to remove noise points appearing at sub-pixel edge points, and then mean filtering is performed. This curvature filtering method reduces the impact of random noise. After filtering by the curvature filtering method, not only can the impact of noise be reduced, but the connection of edge points can also be made smoother.

[0009] Since noise often exists in images, the obtained subpixel edge points are not accurate enough. In addition, if there are noise points in subpixel edge localization, directly performing least squares fitting of the curve will inevitably affect the accuracy of the fitting curve. Therefore, curvature filtering is used for filtering.

[0010] Curvature filtering and least squares trajectory fitting are used to process the ring image. By calculating the geometric parameters of the image, the relevant parameters of human eye indicators are obtained. The sub-pixel contour extraction method has high positioning accuracy and a certain ability to suppress interference noise introduced during the extraction process, thus achieving high-precision measurement of eye indicators.

[0011] Optionally, the curvature filtering method specifically refers to: sorting the curvature of each sub-pixel edge point in descending order, taking the curvature threshold as the 3rd × Nth element of the curvature value sequence from high to low, where N is the number of noise points, its value is the number of times δ=0 occurs during sub-pixel edge localization, and δ is the distance difference between the pixel-level edge point and the real edge point; The curvature of sub-pixel edge points is segmented using a curvature threshold. If the curvature of a sub-pixel edge point is greater than the curvature threshold and is also greater than the curvature of its neighboring points, then the sub-pixel edge point is considered a noise point and is removed.

[0012] Optionally, the curvature of each sub-pixel edge point is calculated, and the curvature calculation formula is as follows: Where (x1, y1), (x2, y2), and (x3, y3) are any three consecutive points in the same plane, and k is the curvature of point (x2, y2). The expressions for x0 and y0 are:

[0013] Optionally, the elliptical trajectory algorithm specifically refers to: a set of parameters used to minimize the distance metric between data points and the ellipse; The elliptical trajectory algorithm uses the least squares ellipse fitting method, which fits the ellipse equation to discrete data points. Its core idea is to find the optimal ellipse parameters that minimize the sum of the squares of the algebraic distances from the data points to the ellipse. The distance metric includes geometric distance and algebraic distance. The geometric distance represents the distance from a point to the nearest point on the curve. The algebraic distance specifically refers to the algebraic distance from a point (x0, y0) in the plane to the curve represented by the equation f(x0, y0) = 0, which is f(x0, y0).

[0014] Optionally, the deep learning model specifically refers to: using a pre-trained deep neural network model, inputting pre-processed vision data, and outputting a probability prediction of vision change trends; Deep neural network models are also known as convolutional neural network models. Convolutional neural network models extract effective representations of input data by establishing multiple convolutional kernels. These convolutional kernels perform convolution and pooling on the input data layer by layer, extracting high-level semantic information layer by layer, abstracting it layer by layer, and finally obtaining the translation and rotation invariant feature representation of the input data.

[0015] The convolutional layer in a convolutional neural network convolves the local receptive field of the input signal with a convolutional kernel. The convolutional kernel traverses the input once with a fixed stride. Each convolutional kernel extracts local features from the local receptive field of the input signal and constructs an output feature vector under the action of an activation function. The output feature vector of each layer is the result of the convolution of multiple input features.

[0016] After convolution, to avoid gradient saturation, activation functions provide the neural network with the ability to model nonlinearly, mapping the originally linearly inseparable multidimensional features to another space. In this space, the linear separability of the features will be enhanced. Furthermore, during the backpropagation of the network, shallow networks may not be able to be effectively trained due to excessively small residuals. Using activation functions can improve this phenomenon. Activation functions include hyperbolic tangent activation function (Tanh), sigmoid activation function, and Rectifield Linear Unit (ReLU). ReLU activation function is currently the most commonly used activation function in deep convolutional neural networks.

[0017] The deep neural network model specifically refers to one composed of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and a softmax layer.

[0018] The introduction of pooling layers is to downsample and abstract visual input objects in the same way as the human visual system. It has advantages such as feature invariance, reduction of network parameters of input features, and prevention of overfitting to a certain extent. Pooling methods include max pooling and mean pooling. Max pooling takes the maximum value of the receptive field of each feature surface as the final output, while mean pooling outputs the average value of the elements in the receptive field. After convolutional and pooling layers, two fully connected layers are connected. The first fully connected layer unfolds all feature vectors into a one-dimensional vector, and the second fully connected layer uses a Softmax regression classifier to solve the classification problem.

[0019] Optionally, the mathematical expression for the convolutional layer is: In the formula: X i j M is the j-th element in the l-th layer. j X is the j-th convolutional region of the feature map of layer l-1. l-1 ij For the elements, w l ij Let b be the weight matrix corresponding to the convolution kernel. l j Here, f(.) is the bias term, and f(.) is the activation function. The mathematical expression for the maximum pooling layer is: The fully connected layer specifically refers to the process where, after the sound signal in the convolutional network is processed by the convolutional and pooling layers, it needs to undergo linear and nonlinear transformations through the fully connected layer to integrate the extracted multidimensional features into a feature vector and map it to the label space for classification. The Softmax layer specifically refers to selecting the category with the highest probability as the final output, which, after Softmax regression processing, becomes:

[0020] Optionally, the machine learning algorithm specifically refers to: dividing customer data into several groups through clustering algorithms, with each group representing a group of people with similar risks of vision decline. The module automatically identifies groups with rapid vision decline or high risk of eye diseases and generates warning reports to send to doctors. Clustering algorithm is an unsupervised learning technique used to divide samples in a dataset into several groups called "clusters," so that samples within the same cluster have high similarity, while samples between different clusters have low similarity. The clustering algorithm is selected based on data characteristics and business needs.

[0021] Optionally, the mobile service, the in-store service, and the remote service can be combined to provide users with continuous, full-cycle eye health management services, including functions such as disease early warning, chronic disease screening, and proactive intervention.

[0022] The present invention has the following beneficial effects: 1. This vision health management system, through the cooperation of data acquisition, data processing, data recognition and data transmission modules, adopts an integrated smart wearable device and vision testing instrument to detect key indicators of patients' vision data, intraocular pressure and refractive error, replacing the single detection method. At the same time, the integration of smart wearable device and vision testing instrument also saves staff from the manual recording of test data, thereby improving the accuracy of patients' vision test data.

[0023] 2. This vision health management system utilizes a pre-trained deep neural network model to input pre-processed vision data and output a probability prediction of vision change trends. The deep neural network model consists of an input layer, convolutional layer, pooling layer, fully connected layer, and softmax layer. Unlike traditional recognition algorithms, the deep neural network model can directly input data without feature extraction or reconstruction, and its weight-sharing network structure can reduce model complexity, making it suitable for training with large datasets.

[0024] 3. This vision health management system combines curvature filtering and elliptical trajectory algorithms. By using curvature filtering, noise points appearing at sub-pixel edges are removed. Then, mean filtering is performed to minimize the impact of random noise. This process not only reduces noise but also makes the connection of edge points smoother. Algorithmically, it combines sub-pixel edge detection and curvature filtering for image denoising. It uses the least squares method to fit an elliptical trajectory and calculates relevant parameters of the human eye index. Attached Figure Description

[0025] Figure 1 This is a flowchart of the present invention; Figure 2 This is a flowchart of the edge computing technology in the structure of this invention; Figure 3 This is a flowchart of the deep learning model in the structure of this invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figures 1 to 3 This invention provides a technical solution: an eye health management system, including data management and service management. The data management system includes a data acquisition module, a data processing module, a data recognition module, and a data transmission module. Through the cooperation between the data acquisition module, data processing module, data recognition module, and data transmission module, an integrated smart wearable device and an eye examination instrument are used to detect key indicators of patients' vision, intraocular pressure, and refractive error, replacing the single detection method. At the same time, the integration of smart wearable device and eye examination instrument also saves staff from the manual recording of test data, thereby improving the accuracy of patients' vision test data. The eye health management system supports multi-terminal access, ensuring that users can obtain and update their personal historical data no matter where they are. Service management includes mobile service, in-store service, and remote service. The data acquisition module specifically refers to the integration of smart wearable devices and optometry testing instruments to collect key indicators such as the customer's visual acuity, intraocular pressure, and refractive error. The data processing module consists of four layers: the first layer uses edge computing technology to preprocess the raw data, removing noise and outliers; the second layer uses deep learning models to analyze long-term vision change trends; and the third layer combines personalized characteristics of genetics, environment, and lifestyle to generate customized health management plans. The data identification module specifically refers to: using machine learning algorithms to label massive amounts of customer data, automatically identifying customer groups at risk of eye diseases, and issuing early warnings to doctors; The data transmission module specifically refers to: matching received call requests, separating structured information from image information, where the structured information includes patient ID and capture date; compressing and encrypting the separated image information using a custom encryption algorithm; and then securely transmitting it to the cloud server via SSL protocol. Finally, professional software parses the data and generates a diagnostic report for doctors' reference. SSL is a network security protocol used to protect the transmission of data over the Internet. The SSL / TLS protocol is an important tool for protecting the security of Internet communications. Through encryption, authentication, and data integrity checks, it ensures the secure transmission of data. The architecture of SSL / TLS includes a handshake protocol, an encryption protocol, and an alert protocol, each of which plays a key role in establishing and maintaining a secure connection.

[0028] The SSL protocol establishes an encrypted channel between the client and server to ensure the confidentiality, integrity, and authentication of data transmission. Confidentiality is achieved through symmetric encryption, with the key negotiated via asymmetric encryption, thus preventing eavesdropping. Integrity is ensured by using MAC (Message Authentication Code) or HMAC to prevent data tampering, thus preventing tampering. Authentication is achieved through the server verifying its identity using a digital certificate (issued by a CA), with optional client certificate verification, thus preventing impersonation. Furthermore, the SSL protocol protects patient privacy. The integrated smart device is equipped with several sensors, including an accelerometer for detecting eye movements, an optical sensor for real-time intraocular pressure monitoring, and a wireless communication module for data transmission. Among them, the accelerometer for detecting eye movements is used to measure the minute acceleration changes generated by eye movements; the optical sensor for monitoring intraocular pressure is used to monitor changes in intraocular pressure in patients with diseases such as glaucoma in real time or continuously. It calculates intraocular pressure by spraying air to flatten the cornea and using infrared light or optical sensors to detect the reflected light when the cornea is deformed, without direct contact with the eyeball. Optical testing instruments include, but are not limited to, refractometers and fundus cameras. A fundus camera is a medical optical device specifically designed to photograph the fundus of the eye, including the retina, optic disc, macula, and blood vessels. It is widely used in ophthalmic examinations, disease diagnosis, and health management. Fundus cameras illuminate the fundus through special optical designs, such as a coaxial illumination system, and use a high-resolution camera to capture reflected light. An refractometer examines the convergence of light after it enters the eye. Using emmetropia as a standard, it measures the difference in convergence and divergence between the examined eye and the emmetropia. Refractometer refraction can be used for diagnostic refraction of soft contact lenses, quickly determining the refractive power without requiring pupil dilation. Refractometer results are all automatically printed, requiring no conversion. A patient can typically be tested in a few seconds to a few minutes, and the refractive error can be quickly determined, providing a relatively accurate refractive power and interpupillary distance for lens correction.

[0029] Edge computing technology includes curvature filtering and elliptical trajectory algorithms. By combining these two methods, the subpixel contour extraction method achieves high positioning accuracy and robustness, enabling high-precision measurement of eye indexes. Noise points appearing at subpixel edge points are removed using curvature filtering, followed by mean filtering. The combination of curvature filtering and elliptical trajectory algorithms, through the specific settings of curvature filtering, minimizes the impact of random noise. This process not only reduces noise but also smooths the connection of edge points. Algorithmically, it combines subpixel edge detection and curvature filtering for image denoising, using least squares to fit an elliptical trajectory and calculate relevant parameters of the human eye index.

[0030] The curvature filtering method specifically refers to: sorting the curvature of each sub-pixel edge point in descending order, taking the curvature threshold as the 3rd × Nth element of the curvature value from high to low in the sequence, where N is the number of noise points, its value is the number of times δ=0 occurs during sub-pixel edge localization, and δ is the distance difference between the pixel-level edge point and the real edge point; The curvature of sub-pixel edge points is segmented using a curvature threshold. If the curvature of a sub-pixel edge point is greater than the curvature threshold and is also greater than the curvature of its neighboring points, then the sub-pixel edge point is considered a noise point and is removed.

[0031] The curvature of each sub-pixel edge point is calculated, and the curvature calculation formula is as follows: Where (x1, y1), (x2, y2), and (x3, y3) are any three consecutive points in the same plane, and k is the curvature of point (x2, y2). The expressions for x0 and y0 are:

[0032] Elliptical trajectory algorithm specifically refers to: finding a set of parameters and minimizing the distance metric between data points and the ellipse. Least squares ellipse fitting is a commonly used ellipse fitting method. Least squares is an optimal estimation technique derived from the maximum likelihood method when the random error is normally distributed. It can minimize the sum of squares of measurement errors, and is therefore regarded as one of the most reliable methods to find a set of unknowns from a set of measurements. Among them, distance measurement includes geometric distance and algebraic distance. Geometric distance represents the distance from a point to the nearest point on a curve, while algebraic distance specifically refers to the algebraic distance from a point (x0, y0) in a plane to the curve represented by the equation f(x0, y0) = 0, which is f(x0, y0).

[0033] Deep learning models specifically refer to: using a pre-trained deep neural network model, taking pre-processed vision data as input, and outputting a probability prediction of vision change trends; A deep neural network model specifically refers to a system composed of an input layer, convolutional layers, pooling layers, fully connected layers, and a softmax layer. Convolutional and pooling layers are non-fully connected neuron structures used for feature extraction. Unlike traditional recognition algorithms, convolutional neural networks can directly accept data without feature extraction or reconstruction. Furthermore, their weight-sharing network structure reduces model complexity, making them suitable for training large datasets. The functions of the input layer, convolutional layers, pooling layers, fully connected layers, and softmax layers are as follows: the input layer takes in data; convolutional and pooling layers are used alternately for feature extraction; the fully connected layer is used for classification; and the softmax layer outputs the probability distribution. Convolutional Neural Networks (CNNs) are a type of neural network with a special network structure that has achieved great success in fields such as image recognition, speech recognition, and natural language processing. They are characterized by high temporal and spatial invariance to translation and scaling of input signals, and also feature local perception, weight sharing, and multiple convolutional kernels. Convolutional layers extract effective representations of input data by establishing several convolutional kernels. These kernels perform convolution and pooling operations layer by layer, extracting high-level semantic information layer by layer and abstracting it progressively to ultimately obtain translation- and rotation-invariant feature representations of the input data. Within a convolutional layer, the kernels perform convolution operations, and then an activation function is used to obtain the feature map for that layer. Pooling layers are typically located between convolutional layers and are used to reduce matrix size, decrease fully connected layer parameters, and merge features. Pooling layers accelerate network computation and avoid overfitting. Average pooling layers use an averaging operation, while max pooling layers are selected... The maximum value within the region, the fully connected layer, is where the sound signal in the convolutional network, after being processed by convolutional and pooling layers, undergoes linear and nonlinear transformations. It integrates the extracted multidimensional features into a feature vector and maps it to a label space for classification. The fully connected layer is similar to the feedforward output of a traditional neural network, used to integrate feature vectors and complete the classification task. The Softmax output layer, in sound recognition tasks, is typically used as the activation function of the output layer to output the probability of the sound belonging to each category. The category with the highest probability is selected as the final output. The mathematical expression for a convolutional layer is: In the formula: X i j M is the j-th element in the l-th layer. j X is the j-th convolutional region of the feature map of layer l-1. l-1 ij For the elements, w lij Let b be the weight matrix corresponding to the convolution kernel. l j Here, f(.) is the bias term, and f(.) is the activation function. The mathematical expression for the max pooling layer is: The fully connected layer specifically refers to the process where, after the sound signal in the convolutional network is processed by the convolutional and pooling layers, it needs to undergo linear and nonlinear transformations through the fully connected layer. This process integrates the extracted multidimensional features into a feature vector and maps it to the label space for classification. The Softmax layer specifically refers to selecting the category with the highest probability as the final output. After Softmax regression processing, the output is:

[0034] Specifically, machine learning algorithms refer to using clustering algorithms to divide customer data into several groups, each group representing a group of people with similar risks of vision decline. The module automatically identifies groups with rapid vision decline or higher risk of eye diseases and generates warning reports to send to doctors. Clustering algorithms are an unsupervised learning method in machine learning that aims to divide samples in a dataset into different categories or "clusters" so that the similarity of data within the same category is as high as possible, while the similarity of data between different categories is as low as possible.

[0035] The combination of mobile services, in-store services, and remote services is used to provide users with continuous, full-cycle eye health management services, including disease early warning, chronic disease screening, and proactive intervention.

[0036] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. The terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, unless otherwise explicitly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. Moreover, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0037] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An eye health management system, characterized in that: It includes data management and service management. The data management includes a data acquisition module, a data processing module, a data identification module, and a data transmission module. The service management includes mobile service, in-store service, and remote service. The data acquisition module specifically refers to the integration of smart wearable devices and optometry testing instruments to collect customer visual acuity, intraocular pressure, and refractive error indicators. The data processing module comprises four layers: the first layer uses edge computing technology to preprocess the raw data, removing noise and outliers; the second layer uses a deep learning model to analyze long-term vision change trends; and the third layer combines personalized characteristics of genetics, environment, and lifestyle to generate customized health management plans. The data identification module specifically refers to: using machine learning algorithms to label massive amounts of customer data, automatically identifying customer groups at risk of eye diseases, and issuing early warnings to doctors; The data transmission module specifically refers to: matching the received call request, separating structured information from image information, wherein the structured information includes patient ID and shooting date, and compressing and encrypting the separated image information using a custom encryption algorithm, then securely transmitting it to the cloud server via SSL protocol, and finally being parsed by professional software to generate a diagnostic report for doctors' reference.

2. The eye health management system according to claim 1, characterized in that: The integrated smart device is equipped with several sensors, including but not limited to an accelerometer for detecting eye movements, an optical sensor for real-time monitoring of intraocular pressure, and a wireless communication module for data transmission. The aforementioned optometry testing instruments are used to capture ocular indicators such as visual acuity, intraocular pressure, and refractive error, including but not limited to optometers and fundus cameras.

3. The eye health management system according to claim 1, characterized in that: The edge computing technology includes a curvature filtering method and an elliptical trajectory algorithm. Noise points appearing at sub-pixel edge points are removed using the curvature filtering method, and then mean filtering is performed.

4. The eye health management system according to claim 3, characterized in that: The curvature filtering method specifically refers to: sorting the curvature of each sub-pixel edge point in descending order, taking the curvature threshold as the 3rd × Nth element of the curvature value from high to low in the sequence, where N is the number of noise points, its value is the number of times δ=0 occurs during sub-pixel edge localization, and δ is the distance difference between the pixel-level edge point and the real edge point. The curvature of sub-pixel edge points is segmented using a curvature threshold. If the curvature of a sub-pixel edge point is greater than the curvature threshold and is also greater than the curvature of its neighboring points, then the sub-pixel edge point is considered a noise point and is removed.

5. The eye health management system according to claim 4, characterized in that: The curvature of each sub-pixel edge point is calculated, and the curvature calculation formula is as follows: Where (x1, y1), (x2, y2), and (x3, y3) are any three consecutive points on the same plane, and k is the curvature of point (x2, y2); the expressions for x0 and y0 are:

6. The eye health management system according to claim 3, characterized in that: The elliptical trajectory algorithm specifically refers to: a set of parameters used to minimize the distance metric between data points and the ellipse; The distance metric includes geometric distance and algebraic distance. The geometric distance represents the distance from a point to the nearest point on the curve. The algebraic distance specifically refers to the algebraic distance from a point (x0, y0) in the plane to the curve represented by the equation f(x0, y0) = 0, which is f(x0, y0).

7. The eye health management system according to claim 1, characterized in that: The deep learning model specifically refers to: using a pre-trained deep neural network model, inputting pre-processed vision data, and outputting a probability prediction of vision change trends; The deep neural network model specifically refers to one composed of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and a softmax layer.

8. The eye health management system according to claim 7, characterized in that: The mathematical expression for the convolutional layer is: In the formula: X i j M is the j-th element in the l-th layer. j X is the j-th convolutional region of the feature map of layer l-1. l-1 ij For the elements, w l ij Let b be the weight matrix corresponding to the convolution kernel. l j Here, f(.) is the bias term, and f(.) is the activation function. The mathematical expression for the maximum pooling layer is: The fully connected layer specifically refers to the process where, after the sound signal in the convolutional network is processed by the convolutional and pooling layers, it needs to undergo linear and nonlinear transformations through the fully connected layer to integrate the extracted multidimensional features into a feature vector and map it to the label space for classification. The Softmax layer specifically refers to selecting the category with the highest probability as the final output, which, after Softmax regression processing, becomes:

9. The eye health management system according to claim 1, characterized in that: The machine learning algorithm specifically refers to: using a clustering algorithm to divide customer data into several groups, each group representing a group of people with similar risks of vision decline. The module automatically identifies groups with rapid vision decline or high risk of eye diseases and generates an early warning report to send to the doctor.

10. An eye health management system according to claim 1, characterized in that: The combination of the mobile service, the in-store service, and the remote service is used to provide users with continuous, full-cycle eye health management services, including functions such as disease early warning, chronic disease screening, and proactive intervention.

Images