Intelligent myopia prevention and control method and system based on artificial intelligence and multi-mode sensing

Abstract

The invention relates to an intelligent myopia prevention and control method and system based on artificial intelligence and multi-modal sensing, and the method comprises the steps: obtaining the three-dimensional data of the contour of the eyes of a user through an integrated microspur 3D scanning module after the user wears an intelligent eyeshade, and carrying out the processing through an AI algorithm, so as to precisely position the target acupoints around the eyes, and achieve the precise positioning of the target acupoints. A multi-mode sensing unit integrated on the intelligent eyeshade is used for collecting initial physiological data around the eyes of a user in real time, the physiological data comprise the oxyhemoglobin saturation, the pressing pressure and the electromyographic signals, and a central processing unit sends the three-dimensional data to the intelligent eyeshade according to the three-dimensional data, the initial physiological data and pre-stored user feature information. A set of personalized myopia prevention and control intervention scheme is obtained, after the intervention scheme is executed, simple vision testing is carried out through terminal equipment, a short-term effect evaluation report is generated in combination with sensor data, and the report is uploaded to a cloud platform to update a digital vision health file of a user.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of digital data technology, in particular to an intelligent myopia prevention and control method and system based on artificial intelligence and multi-modal sensing. BACKGROUND

[0002] At present, the problem of myopia is becoming increasingly serious, especially in the children and adolescent groups, which shows a high incidence and a low age trend. The existing myopia prevention and control means in the market, including eye massage devices and eye protection exercises, generally have obvious limitations.

[0003] Firstly, the traditional method relies on the user to locate the acupoint by himself, which has poor accuracy and seriously affects the intervention effect. Secondly, the existing device has single method parameters and lacks individualized adjustment ability according to the individual characteristics (acupoint sensitivity, fatigue degree, and constitution) of the user. Thirdly, the intervention effect relies on the subjective feeling of the user, which is difficult to be objectively quantified, resulting in low user compliance. Finally, these means cannot form continuous and long-term vision health data tracking, which is not conducive to the systematic evaluation and scheme optimization of the prevention and control effect.

[0004] Although there are some improved technologies, such as massage eye masks with acupoint positioning function, there are still deficiencies in intelligence, individualization, and data, which cannot realize real precise adaptation and closed-loop management. SUMMARY

[0005] In order to solve the problems existing in the prior art, the present application aims to provide an intelligent myopia prevention and control method and system based on artificial intelligence and multi-modal sensing.

[0006] The intelligent myopia prevention and control method based on artificial intelligence and multi-modal sensing described in the present application comprises the following steps:

[0007] S101, after the user wears the intelligent eye mask, the three-dimensional data of the user's eye contour is obtained through the integrated macro 3D scanning module of the intelligent eye mask, and an AI algorithm is used for processing, which is used for precise positioning of the target acupoint around the eye;

[0008] S102, the initial physiological data of the user's eye contour is collected in real time through the multi-modal sensing unit integrated on the intelligent eye mask, and the physiological data includes blood oxygen saturation, pressing pressure, and electromyographic signal;

[0009] S103, the central processing unit obtains a set of individualized myopia prevention and control intervention scheme according to the three-dimensional data, the initial physiological data, and the pre-stored user characteristic information;

[0010] S104, according to the individualized intervention scheme, the execution module in the intelligent eye mask is driven to massage and stimulate the target acupoint, and the stimulation parameters are dynamically adjusted based on the real-time physiological data fed back by the multi-modal sensing unit during the execution process;

[0011] S105, after the intervention scheme is completed, a simple visual acuity test is performed through the terminal device, and a short-term effect evaluation report is generated in combination with sensor data, and the report is uploaded to the cloud platform to update the digital visual health profile of the user.

[0012] Further, in the S101 step, the smart eyeshade obtains the user's eye contour through micro-distance 3D scanning to generate an original point cloud data set, processes the data using a point cloud segmentation algorithm to separate the key area, extracts features from the segmented point cloud through a convolutional neural network to generate a feature vector, and if the spatial distribution of the vector exceeds the matching degree threshold of the preset acupoint template library, the preliminary acupoint coordinates are determined, the deviation of the coordinates is corrected using a geometric transformation method, and the corrected coordinates are mapped to the eye anatomy to generate accurate three-dimensional position data of the target acupoint. Based on this data, an acupoint positioning path is generated through an interpolation algorithm, and the final positioning result is output.

[0013] Further, in the S102 step, the blood oxygen saturation, pressing pressure and electromyographic signals of the user's eye are collected in real time through the multi-modal sensing unit to generate an initial physiological data set, the data is preprocessed to eliminate noise using a time series analysis method to obtain a smoothed data set, and if the blood oxygen saturation signal exceeds the preset threshold, the frequency characteristics thereof are extracted through Fourier transform, and based on the feature set, a support vector machine algorithm is used to classify the blood oxygen and electromyographic signals to determine the eye physiological state category. The pressing pressure data is combined with the state category, a pressure distribution model of the eye is generated through an interpolation method, and if the local peak value in the model matches the preset anatomical position template, geometric correction is performed to determine the positioning coordinates of the target area of the eye. According to the coordinates, a dynamic adjustment path is generated using a path planning algorithm, and the positioning adjustment result is output.

[0014] Further, in the S103 step, the central processing unit first extracts the eye geometry features from the three-dimensional data to obtain the eye structure parameters, combines the initial physiological data, and extracts the key physiological indicators using a principal component analysis method to form a compressed physiological feature set. If the indicators in the feature set exceed the preset threshold, the physiological features are classified through a decision tree algorithm to determine the user's myopia risk level, and based on the risk level, the pre-stored user feature information is combined to generate personalized prevention and control parameters using an interpolation method to form a preliminary intervention scheme. The scheme is further combined with the user portrait to optimize the execution order thereof through a path planning algorithm to obtain a dynamic adjustment scheme, and according to the dynamic adjustment scheme, the corresponding intervention measures are matched from the pre-stored prevention and control strategy library to output the final personalized myopia prevention and control intervention scheme.

[0015] Further, in step S104, real-time physiological data is collected through a multimodal sensing unit, and noise is removed using a signal filtering method to obtain a purified physiological signal. Based on this signal, the features are classified using a support vector machine algorithm to determine the stimulation requirement level of the target acupoint. If the level is higher than a preset threshold, the corresponding massage mode is matched through a pre-stored acupoint stimulation template library to obtain initial massage parameters. Combined with real-time physiological data, a linear regression algorithm is used to predict the trend of massage effect, thereby generating optimized massage parameters. These parameters are used to drive the execution module to adjust the massage frequency and intensity to form a dynamic execution plan. Adaptation rules are extracted from a pre-stored feedback adjustment strategy library to obtain the final stimulation plan, which drives the execution module to continuously stimulate the acupoint and obtain real-time feedback data.

[0016] Further, in step S105, vision test data is collected through a terminal device and formatted using a standardized vision assessment scale to obtain a preliminary vision score. Based on this score, real-time physiological signals are acquired from a multimodal sensor and smoothed using a moving average filtering method to obtain a purified physiological signal. If the fluctuation amplitude of this signal exceeds a preset threshold, a decision tree algorithm is used to classify the preliminary vision score and the purified physiological signal to determine the short-term vision status level. Based on this level, a pre-stored vision health feature template is extracted to generate a user vision health feature vector. The cosine similarity algorithm is used to compare this vector with the health record data in the cloud platform to determine the record update requirement. If the requirement exceeds a threshold, the feature vector is transmitted through the cloud platform interface to update the digital vision health record. Based on the updated record data, a suitable intervention adjustment plan is matched from the pre-stored intervention strategy library to obtain personalized intervention parameters.

[0017] The intelligent myopia prevention and control system based on artificial intelligence and multimodal sensing described in this application includes:

[0018] The intelligent acupoint positioning eye mask module is designed to be worn around the user's eyes and integrates a macro 3D scanning unit and multiple execution units.

[0019] A multimodal sensing data acquisition module is integrated into the intelligent acupoint positioning eye mask module to collect real-time physiological data such as blood oxygen saturation, pressure, and electromyographic signals around the user's eyes.

[0020] The central processing and AI algorithm module is connected to the intelligent acupoint positioning eye mask module and the multimodal sensing data acquisition module. It is used to receive three-dimensional data and the physiological data, and generate personalized myopia prevention and control intervention plans based on AI algorithms.

[0021] The execution control module is connected to the execution unit within the central processing and AI algorithm module and the intelligent acupoint positioning eye mask module. It is used to drive the execution unit to execute the intervention plan and dynamically adjust the stimulation parameters based on real-time physiological data.

[0022] The user interaction and terminal module is connected to the central processing and AI algorithm module and is used to provide a user interface for scheme selection, real-time data monitoring and vision testing.

[0023] The cloud platform and data management module, connected to the user interaction and terminal module and the central processing and AI algorithm module, is used to store the user's digital vision health records and provide support for the optimization of intervention programs based on the record data.

[0024] The intelligent myopia prevention and control method and system based on artificial intelligence and multimodal sensing described in this application has the advantages of achieving precise acupoint positioning through 3D scanning and AI algorithms, and dynamically adjusting parameters by combining user characteristics and real-time physiological feedback. This solves the pain points of inaccurate acupoint positioning and single manipulation techniques in traditional methods, and achieves personalized and precise intervention for each individual. By integrating multimodal sensors, physiological indicators such as periorbital blood oxygen and muscle tension are transformed into objective data, making the intervention effect measurable and verifiable, thereby enhancing the scientific nature of the plan and user compliance.

[0025] The system builds an ecosystem of "devices + cloud platform" to create long-term digital vision health records for each user, enabling full-cycle health management from immediate intervention to long-term trend tracking, and providing data support for personalized solution optimization and group research. It deeply integrates traditional Chinese medicine meridian theory with modern artificial intelligence and sensing technology, forming a unique technological barrier. At the same time, the built-in pressure and temperature monitoring and other safety designs ensure the safety and comfort of home use. Attached Figure Description

[0026] Figure 1 This application describes a process for an intelligent myopia prevention and control method based on artificial intelligence and multimodal sensing. Figure 1 ;

[0027] Figure 2 This application describes a process for an intelligent myopia prevention and control method based on artificial intelligence and multimodal sensing. Figure 2 . Detailed Implementation

[0028] like Figures 1-2 As shown, the intelligent myopia prevention and control method based on artificial intelligence and multimodal sensing described in this application includes:

[0029] like Figures 1-2As shown in S101, after the user wears the smart eye mask, the integrated macro 3D scanning module acquires the three-dimensional data of the user's eye contour and processes it with AI algorithms to accurately locate target acupoints around the eyes.

[0030] Further, in step S101, the three-dimensional data of the periorbital contour collected by the macro 3D scanning module when the user wears the smart eye mask is obtained, and the original point cloud dataset is generated.

[0031] The original point cloud dataset was processed using a point cloud segmentation algorithm to separate the key regions of the peri-eye contour, resulting in segmented point cloud data.

[0032] Feature extraction is performed on the segmented point cloud data using a convolutional neural network to generate feature vectors of the periorbital contour.

[0033] If the spatial distribution of the feature vector matches the preset acupoint template library by more than a preset threshold, then the preliminary acupoint coordinates are determined.

[0034] Based on the preliminary acupoint coordinates, the coordinate deviation is corrected using a geometric transformation method to obtain the corrected acupoint coordinates.

[0035] By mapping the corrected acupoint coordinates to the anatomical structures around the eyes, three-dimensional location data of the target acupoints are generated.

[0036] Based on the three-dimensional location data of the target acupoint, an interpolation algorithm is used to generate the acupoint location path, and the final acupoint location result is output.

[0037] Specifically, in step S101, after the user puts on the smart eye mask, its integrated macro 3D scanning module is immediately activated. It uses the built-in structured light projector to project an infrared light grid pattern with a wavelength of 850nm onto the area around the eyes. The camera captures the reflected light deformation image at a speed of 60 frames per second, thereby generating high-resolution point cloud data containing approximately 50,000 three-dimensional coordinate points, with each point having an accuracy of 0.1mm.

[0038] Through AI algorithm processing, a Gaussian filter with a standard deviation of σ=2.0 is first used to smooth the noise of the point cloud to eliminate scanning interference. Then, the principal component analysis (PCA) algorithm is applied to extract the main feature vectors around the eye, reduce the data to three-dimensional space, calculate the relative position of the eye center coordinates (0,0,0), and identify the contour boundary of the temple TS23 with a depth of 5.2mm and a width of 3.8mm.

[0039] Next, a convolutional neural network (CNN) model is introduced. This model is pre-trained on a dataset containing 10,000 labeled images around the eyes. The Adam optimizer is used to iterate for 50 cycles with a learning rate of 0.001 to perform semantic segmentation on the contour data and output acupoint probability heatmaps. Among them, the temple peak is analyzed to meet user needs. After the user wears the smart eye mask, its integrated macro 3D scanning module will collect three-dimensional data of the contour around the eyes.

[0040] The probability of the value reached 0.95, and the peak value of EX-HN4 at the Yuyao acupoint was 0.88. The location of the target acupoint was confirmed by filtering with a threshold of 0.8.

[0041] The algorithm integrates a K-nearest neighbor (KNN) classifier to calculate the Euclidean distance between acupoints with k=5. For example, the distance from the Taiyang acupoint to the Jingming acupoint (BL1) is 12.4mm, ensuring that the positioning error is less than 0.5mm. This provides precise coordinate guidance for subsequent stimulation of acupoints around the eyes. The entire process is completed within 0.5 seconds, forming a closed-loop logical chain from data acquisition to analysis output, supporting real-time health monitoring applications.

[0042] like Figures 1-2 As shown in S102, the user's initial physiological data around the eyes is collected in real time through the multimodal sensing unit integrated on the smart eye mask. The physiological data includes blood oxygen saturation, pressure, and electromyography signals.

[0043] Furthermore, in step S102, the user's blood oxygen saturation, pressure, and electromyography signals around the eyes are collected in real time by a multimodal sensing unit to generate an initial physiological dataset.

[0044] The initial physiological dataset was preprocessed using time series analysis to remove noise interference, resulting in a smoothed physiological dataset.

[0045] If the blood oxygen saturation signal of the smoothed physiological dataset exceeds the preset threshold, the frequency features of the signal are extracted by Fourier transform to obtain the frequency feature set.

[0046] Based on the frequency feature set, the support vector machine algorithm is used to classify blood oxygen saturation and electromyography signals to determine the category of periocular physiological state;

[0047] By mapping the pressure data to the physiological state categories around the eyes, an interpolation method is used to generate a pressure distribution model of the area around the eyes, and pressure distribution data is obtained.

[0048] If the local peak value of the pressure distribution data matches the preset anatomical location template, the pressure distribution data is adjusted by geometric correction method to determine the positioning coordinates of the target area around the eye.

[0049] Based on the positioning coordinates, a path planning algorithm is used to generate a dynamic adjustment path for the target area around the eye, resulting in the final positioning adjustment result.

[0050] Specifically, in step S102, after the user wears the smart eye mask, its integrated multimodal sensing unit collects blood oxygen saturation, pressure, and electromyographic signal data around the eyes in real time.

[0051] The sensing unit uses an optical blood oxygen sensor, which illuminates the skin around the eyes with dual-wavelength infrared light of 660nm and 940nm. The reflected light intensity is captured by a photodiode to calculate blood oxygen saturation. The sampling frequency is 100Hz, generating 100 data points per second with an accuracy of 0.5%.

[0052] Meanwhile, the pressure sensor detects the pressure applied around the eyes with a range of 10 kPa and a resolution of 0.01 kPa, and the sampling frequency is 50 Hz. It records the peak pressure of each press, for example, an average of 2.3 kPa.

[0053] The electromyography sensor captures electrical signals from the muscles around the eye at a resolution of 1 mV, with a sampling rate of 500 Hz, generating a timing signal containing approximately 5,000 data points.

[0054] The collected data were integrated using a preprocessing algorithm. First, Kalman filtering was applied to the blood oxygen data, with the process noise covariance set to 0.01 to eliminate ambient light interference, resulting in a smooth blood oxygen curve with a mean of 95.2%.

[0055] The pressure data was obtained by using a sliding window algorithm with a window size of 0.2 seconds to extract the dynamic features of each press and calculate the average pressure change rate as 0.15 kPa / s.

[0056] The electromyographic signals were converted from the time domain to the frequency domain using Fast Fourier Transform (FFT), and the dominant frequency in the 0-50Hz band was extracted. The average value was 12.7Hz, which reflects the intensity of periorbital muscle activity.

[0057] The fusion algorithm utilizes a support vector machine (SVM) model, pre-trained on a set containing 5000 sets of labeled physiological data, with an RBF kernel function and a penalty parameter C=1.0. It analyzes the correlation between blood oxygen, pressure, and electromyography characteristics, and outputs the probability of a healthy state, such as a fatigue state probability of 0.82.

[0058] Based on time series analysis, the system calculates the correlation coefficient between blood oxygen and electromyography signals to be 0.75. Combined with stress data, it generates a real-time health monitoring report. The entire process takes 0.3 seconds, providing data support for subsequent personalized adjustments.

[0059] like Figures 1-2 As shown, S103 and the central processing unit obtain a personalized myopia prevention and control intervention plan based on the three-dimensional data, initial physiological data and pre-stored user characteristic information.

[0060] Further, in step S103, the central processing unit extracts eye geometric features from the three-dimensional data to obtain eye structural parameters;

[0061] Based on ocular structural parameters and initial physiological data, principal component analysis was used to extract key physiological indicators and obtain a compressed physiological feature set.

[0062] If the indicators in the compressed physiological feature set exceed the preset threshold, the physiological features are classified using a decision tree algorithm to determine the myopia risk level.

[0063] Based on the myopia risk level and combined with pre-stored user characteristic information, an interpolation method is used to generate personalized prevention and control parameters to obtain a preliminary intervention plan.

[0064] Based on the initial intervention plan and user profile, a path planning algorithm is used to optimize the execution order of the intervention plan, resulting in a dynamic adjustment plan.

[0065] Based on the dynamic adjustment plan, corresponding intervention measures are matched from the pre-stored prevention and control strategy library to obtain the final personalized myopia prevention and control intervention plan.

[0066] Specifically, in step S103, the central processing unit receives three-dimensional data, initial physiological data and pre-stored user characteristic information, and generates a personalized myopia prevention and control intervention plan through multi-dimensional analysis;

[0067] First, the processing unit extracts eye structure features from the three-dimensional data, uses a three-dimensional point cloud registration algorithm, and reconstructs the periocular geometric model with a spatial resolution of 0.1mm based on the iterative nearest point method (ICP). The radius of curvature of the eyeball is calculated, with an average value of 7.8mm and a deviation controlled within 0.05mm.

[0068] Combining heart rate data from initial physiological data, with a sampling frequency of 10Hz and an average heart rate of 72 beats / minute, an adaptive filtering algorithm was used to remove high-frequency noise by setting the filter bandwidth to 0.5Hz, extracting heart rate variability (HRV) features, and calculating a standard deviation of 45ms to reflect the user's autonomic nervous state.

[0069] The pre-stored user characteristics include age 15, myopia degree -3.5D, and average daily screen time of 6 hours. The data is input into a logistic regression model, pre-trained on a dataset of 10,000 users, with a weight parameter of 0.8 and a bias of 0.2. The model predicts the risk of myopia progression and outputs a risk probability of 0.65.

[0070] Next, by integrating the analysis of three-dimensional data and physiological data, principal component analysis (PCA) algorithm was used to extract the correlation features between heart rate and eye curvature. After retaining 95% of the variance, the correlation coefficient was calculated to be 0.68, indicating that there is a weak correlation between eye structure and heart rate fluctuations.

[0071] Based on this, the system optimizes the intervention plan through a genetic algorithm, sets the population size to 50, iterates 20 times, and generates a plan that includes a limit on screen time (5 hours per day), screen brightness adjustment (300 cd / m²), and rest intervals (5 minutes of rest every 40 minutes).

[0072] The solution is stored in JSON format and takes 0.2 seconds to process, and is then executed by smart devices.

[0073] like Figures 1-2 As shown, in step S104, the execution module in the smart eye mask is driven to massage and stimulate the target acupoints, and the stimulation parameters are dynamically adjusted based on the real-time physiological data fed back by the multimodal sensing unit during the execution process.

[0074] Furthermore, in step S104, real-time physiological data is acquired through a multimodal sensing unit, and noise interference is removed by signal filtering to obtain purified physiological signals;

[0075] Based on the physiological signals of purification, the support vector machine algorithm is used to classify the signal features and determine the stimulation requirement level of the target acupoints;

[0076] If the stimulation requirement level is higher than the preset threshold, the corresponding massage mode is matched by the pre-stored acupoint stimulation template library to obtain the initial massage parameters;

[0077] Based on the initial massage parameters and real-time physiological data, a linear regression algorithm is used to predict the trend of massage effect and obtain optimized massage parameters.

[0078] By optimizing massage parameters, the execution module of the smart eye mask is driven to adjust the massage frequency and intensity, resulting in a dynamic execution plan;

[0079] Based on the dynamic execution plan, the appropriate adjustment rules are extracted from the pre-stored feedback adjustment strategy library to obtain the final stimulus plan;

[0080] The final stimulation scheme drives the execution module to continuously massage and stimulate the target acupoints, obtaining real-time feedback data.

[0081] Specifically, in step S104, after receiving the personalized intervention plan, the execution module of the smart eye mask drives the massage unit through the embedded microprocessor to stimulate target acupoints such as Zanzhu, Yuyao and Taiyang.

[0082] The microprocessor first loads the preset acupoint coordinates, based on a three-dimensional head model with a spatial resolution of 0.2 mm, and uses the Fast Fourier Transform (FFT) algorithm to locate the acupoints, with the calculation deviation controlled within 0.1 mm;

[0083] The massage unit initiates vibration stimulation at a frequency of 3Hz and a pressure of 0.5N, lasting for 30 seconds per acupoint, repeated 3 times, for a total duration of 90 seconds;

[0084] Meanwhile, the multimodal sensing unit acquires periocular electromyography (EMG) and skin conductance (GSR) signals in real time at a sampling frequency of 50 Hz, with an average EMG signal intensity of 0.15 mV and a GSR change rate of 0.02 S / s.

[0085] The collected data was denoised using a Kalman filter algorithm with a filter gain of 0.9. After removing environmental interference, the time-domain features of the electromyography signal were extracted, and the root mean square value was calculated to be 0.12mV, reflecting the degree of muscle fatigue.

[0086] Using GSR data, a support vector machine (SVM) model was pre-trained on 5000 samples with a radial basis function (RBF) kernel function and parameter C set to 1.0 to predict user relaxation level, with an output value of 0.75.

[0087] Based on this, the system dynamically adjusts the massage parameters. If the relaxation level is below 0.7, the vibration frequency is increased to 4Hz and the pressure is increased to 0.6N; if it is above 0.8, the frequency is decreased to 2.5Hz and the pressure is reduced to 0.4N.

[0088] The adjustment process is optimized using the gradient descent algorithm with a learning rate of 0.01 and 10 iterations to ensure that the stimulation parameters match the physiological feedback.

[0089] The adjusted parameters are stored in the device memory in binary format, and the processing time is 0.15 seconds, which is then used for the next loop execution.

[0090] like Figures 1-2 As shown in step S105, after the intervention plan is completed, a simple vision test is conducted through the terminal device, and a short-term effect evaluation report is generated in combination with sensor data. The report is then uploaded to the cloud platform to update the user's digital vision health record.

[0091] Further, in step S105, visual acuity test data is collected through a terminal device, and the data is formatted using a standardized visual acuity assessment scale to obtain a preliminary visual acuity score.

[0092] Based on the preliminary vision score, real-time physiological signals are acquired from multimodal sensors, and the signals are smoothed using a moving average filtering method to obtain purified physiological signals.

[0093] If the fluctuation range of the purification physiological signal is higher than the preset threshold, the decision tree algorithm is used to classify the preliminary vision score and the purification physiological signal to determine the short-term vision status level.

[0094] Based on the short-term visual status level, extract the pre-stored visual health feature template and generate the user's visual health feature vector;

[0095] By using the user's vision health feature vector, the cosine similarity algorithm is used to compare the health record data stored in the cloud platform to determine the need for record updates.

[0096] If the need for updating the records exceeds the preset threshold, the user's vision health feature vector will be transmitted through the cloud platform interface to update the digital vision health record and obtain the updated record data.

[0097] Based on the updated archival data, suitable intervention adjustment plans are matched from the pre-stored vision intervention strategy library to obtain personalized intervention parameters.

[0098] Specifically, in step S105, after the intervention plan is completed, the smart eye mask initiates a simple vision test through the terminal device. The test module loads a preset vision detection algorithm based on the Landolt C test method. The screen displays C-shaped patterns in four directions with a resolution of 1080p, a pixel pitch of 0.1mm, and a test distance of 0.5m.

[0099] The system generates 10 sets of random patterns, with the pattern size increasing from 0.5 arcminutes to 5 arcminutes and the interval being 0.5 arcminutes. The user confirms the direction of the C-shaped opening through voice input or eye tracking. The sampling frequency is 60Hz, the eye tracking accuracy is 0.05 degrees, and the system records the reaction time for each time, with an average of 0.8 seconds.

[0100] The test data was dimensionality reduced using principal component analysis (PCA) algorithm, retaining 95% of the variance, and visual acuity and reaction speed features were extracted to calculate a visual acuity score ranging from 0 to 1, with an example score of 0.85.

[0101] Multimodal sensors were used to collect periocular oxygen saturation (SpO2) and heart rate variability (HRV) after intervention. The sampling frequency was 100 Hz, the mean SpO2 was 98%, and the standard deviation of HRV was 45 ms.

[0102] Sensor data was denoised using wavelet transform with Daubechies4 as the wavelet basis function and 5 decomposition levels. The frequency domain features of SpO2 and the time domain features of HRV were extracted, and the mean HRV was calculated to be 50ms, reflecting the short-term relaxation effect.

[0103] The system combines vision scores and sensor data, employs a random forest model, pre-trained on 10,000 samples, with a tree depth of 10, and feature importance weights of 0.6 for vision, 0.25 for SpO2, and 0.15 for HRV. It generates a short-term effect evaluation report with a quantization value of 0.88, stored in JSON format, and occupying approximately 50KB of space.

[0104] The report was encrypted using the AES-256 encryption algorithm with a key length of 256 bits, uploaded to the cloud platform at a speed of 10Mbps, and took 0.2 seconds.

[0105] The cloud platform analyzes the report and updates the user's digital vision health record. The record is stored in a NoSQL database with an update latency of less than 0.1 seconds. Fields include vision score, mean SpO2, HRV standard deviation, and timestamp, ensuring the record's real-time performance and consistency.

[0106] The intelligent myopia prevention and control system based on artificial intelligence and multimodal sensing described in this application includes:

[0107] The intelligent acupoint positioning eye mask module is designed to be worn around the user's eyes and integrates a macro 3D scanning unit and multiple execution units.

[0108] A multimodal sensing data acquisition module is integrated into the intelligent acupoint positioning eye mask module to collect real-time physiological data such as blood oxygen saturation, pressure, and electromyographic signals around the user's eyes.

[0109] The central processing and AI algorithm module is connected to the intelligent acupoint positioning eye mask module and the multimodal sensing data acquisition module. It is used to receive three-dimensional data and the physiological data, and generate personalized myopia prevention and control intervention plans based on AI algorithms.

[0110] The execution control module is connected to the execution unit within the central processing and AI algorithm module and the intelligent acupoint positioning eye mask module. It is used to drive the execution unit to execute the intervention plan and dynamically adjust the stimulation parameters based on real-time physiological data.

[0111] The user interaction and terminal module is connected to the central processing and AI algorithm module and is used to provide a user interface for scheme selection, real-time data monitoring and vision testing.

[0112] The cloud platform and data management module, connected to the user interaction and terminal module and the central processing and AI algorithm module, is used to store the user's digital vision health records and provide support for the optimization of intervention programs based on the record data.

[0113] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.

Claims

1. A method and system for intelligent myopia prevention and control using artificial intelligence and multimodal sensing, characterized in that, include: S101. After the user wears the smart eye mask, the integrated macro 3D scanning module obtains the three-dimensional data of the user's eye contour and processes it with AI algorithm to accurately locate the target acupoints around the eyes. S102. Through the multimodal sensing unit integrated on the smart eye mask, the initial physiological data around the user's eyes is collected in real time, including blood oxygen saturation, pressure, and electromyography signals. S103. The central processing unit obtains a personalized myopia prevention and control intervention plan based on three-dimensional data, initial physiological data and pre-stored user characteristic information. S104. According to the personalized intervention plan, drive the execution module in the smart eye mask to massage and stimulate the target acupoints, and dynamically adjust the stimulation parameters based on the real-time physiological data fed back by the multimodal sensing unit during the execution process. S105. After the intervention program is completed, a simple vision test is conducted through the terminal device, and a short-term effect evaluation report is generated in combination with sensor data. The report is then uploaded to the cloud platform to update the user's digital vision health record.

2. The intelligent myopia prevention and control method and system based on artificial intelligence and multimodal sensing according to claim 1, characterized in that, Step S101 includes: The three-dimensional data is segmented into point clouds to separate the key area around the eyes. Features are extracted from the segmented point clouds using a convolutional neural network to generate feature vectors. The feature vectors are matched with a preset acupoint template library. If the matching degree exceeds a threshold, preliminary acupoint coordinates are determined. Geometric transformation methods are used to correct the deviation of the preliminary acupoint coordinates to generate accurate three-dimensional position data of the target acupoint.

3. The intelligent myopia prevention and control method and system based on artificial intelligence and multimodal sensing according to claim 1, characterized in that, Step S102 includes: Time series analysis was used to preprocess the data to remove noise. The frequency features of the blood oxygen saturation signal were extracted by Fourier transform. The support vector machine algorithm was used to classify the blood oxygen saturation and electromyography signals to determine the category of the physiological state around the eyes. Combined with the pressure data, a pressure distribution model around the eyes was generated and the location coordinates of the target area were determined.

4. The intelligent myopia prevention and control method and system based on artificial intelligence and multimodal sensing according to claim 1, characterized in that, Step S103 includes: Eye geometric features are extracted from the three-dimensional data to obtain eye structural parameters. Combined with the initial physiological data, key physiological indicators are extracted using principal component analysis. The physiological features are classified using a decision tree algorithm to determine the user's myopia risk level. Based on the myopia risk level and pre-stored user feature information, a genetic algorithm is used to optimize and generate a dynamic adjustment scheme that includes massage parameters and eye behavior suggestions.

5. The intelligent myopia prevention and control method and system based on artificial intelligence and multimodal sensing according to claim 1, characterized in that, Step S104 includes: Based on the support vector machine, real-time physiological signals are classified to determine the stimulation demand level of target acupoints. The trend of massage effect is predicted by the linear regression algorithm to obtain optimized massage parameters. Based on the optimized massage parameters, the massage frequency and intensity of the execution module are dynamically adjusted.

6. The intelligent myopia prevention and control method and system based on artificial intelligence and multimodal sensing according to claim 1, characterized in that, Step S105 includes: The visual acuity test data were formatted using a standardized visual acuity assessment scale to obtain a preliminary visual acuity score. The physiological signals after intervention were smoothed using a moving average filtering method. The preliminary visual acuity score and the purified physiological signals were classified using a decision tree algorithm to determine the short-term visual acuity status level. Based on the level, a random forest model was used to generate a quantitative effect evaluation report.

7. A system for implementing the method according to any one of claims 1 to 6, characterized in that, include: The intelligent acupoint positioning eye mask integrates a micro-3D scanning module for performing step S101, a multimodal sensing unit for performing step S102, and an execution module for performing step S104. The central processing unit is used to execute the decision and adjustment logic in steps S103 and S104; User interaction terminal, used to execute step S105; The cloud platform is used to execute step S105.

Images