Personalized eye behavior correction ai recommendation and real-time intervention system

Abstract

The application discloses a personalized eye behavior correction AI recommendation and real-time intervention system, and relates to the technical field of eye behavior correction, which comprises a data acquisition module, an AI analysis and processing module, a personalized recommendation module, a real-time intervention module, a data storage module and a man-machine interaction module, each module is bidirectionally linked through a communication link, and a closed-loop operation system of "collection-analysis-recommendation-intervention-feedback-optimization" is formed. Through the remote linkage module, linkage with ophthalmic medical institutions, parent terminals and teacher terminals is realized, a multiple correction system is formed, the problem of lack of professional medical support and disconnection of supervision in the prior art is solved, and the professionalism and effectiveness of correction are further improved.

Description

Technical Field

[0001] This invention relates to the field of eye health monitoring and behavior correction technology, and in particular to a personalized AI recommendation and real-time intervention system for correcting eye behavior. Background Technology

[0002] With the widespread use of electronic products and the diversification of eye-use scenarios, eye health problems such as myopia, eye fatigue, and dry eye syndrome are becoming increasingly common and showing a trend towards affecting younger people. Correcting eye-use behavior has become a key means of protecting eye health. Currently, existing technologies and products related to eye-use behavior correction generally have many technical shortcomings, making it difficult to meet the personalized correction needs of different users, and also failing to achieve accurate monitoring and efficient intervention of eye-use behavior.

[0003] In existing technologies, most eye behavior correction solutions adopt a homogenized design, failing to fully consider individual differences among users such as age, vision, eye habits, and eye tolerance. This results in poor targeting and ineffective correction. Furthermore, most correction systems intervene significantly late, often only after users exhibit serious poor eye behavior or exceed eye health risk limits. This fails to promptly curb the continued impact of poor eye behavior and makes it difficult to achieve the goal of "prevention before the disease occurs."

[0004] Regarding data collection, existing systems are mostly limited to single-dimensional data collection, such as monitoring only eye distance or duration, lacking comprehensive collection of multi-dimensional data such as eye physiological parameters, eye environment parameters, and eye behavior trajectories. This results in insufficient data support for AI analysis, inaccurate feature extraction, and consequently affects the rationality and suitability of correction solutions. In terms of AI analysis and processing, existing models mostly adopt traditional deep learning architectures, which suffer from slow convergence speed, inability to adapt to multi-dimensional heterogeneous data, and low feature extraction accuracy. Furthermore, the model parameters are fixed and cannot be optimized in real time according to the dynamic changes in user eye behavior, leading to a decline in adaptability after long-term use.

[0005] Furthermore, existing intervention methods are relatively simplistic, often relying solely on voice or visual reminders. These methods lack flexibility in adapting to user scenarios, preferences, and intervention priorities, easily leading to user resistance and reduced adherence. Simultaneously, the lack of remote collaboration with ophthalmology institutions, parents, and teachers prevents the formation of a comprehensive correction system that combines professional guidance with multi-party supervision, further impacting the effectiveness of the intervention. Regarding data storage, existing systems primarily employ centralized storage, resulting in data disorganization, low query efficiency, and poor security for user privacy data, making it difficult to guarantee the security and traceability of user data.

[0006] In response to the shortcomings of the existing technologies, there is an urgent need for an eye behavior correction system that can achieve multi-dimensional data collection, AI intelligent analysis, personalized solution recommendation, millisecond-level real-time intervention, and form a closed-loop optimization system. This system would improve the accuracy, effectiveness, and convenience of eye behavior correction and solve the core problems of existing technologies, such as homogenization, delayed intervention, and poor individual adaptability.

[0007] Therefore, there is a need to design a personalized AI recommendation and real-time intervention system for correcting eye behavior. Summary of the Invention

[0008] The purpose of this invention is to overcome the technical shortcomings of existing technologies, such as homogenization of eye behavior correction solutions, delayed intervention timing, lack of individual adaptability of recommendation strategies, and inability to achieve real-time monitoring and dynamic intervention linkage of eye behavior. This invention provides a personalized eye behavior correction AI recommendation and real-time intervention system. Through multi-dimensional data fusion analysis and AI intelligent learning, it achieves personalized correction solution recommendations and millisecond-level real-time intervention for different users, improving the accuracy and effectiveness of eye behavior correction. Simultaneously, it solves problems such as single data collection, poor AI model adaptability, single intervention methods, and insecure data storage in existing technologies.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: The personalized eye behavior correction AI recommendation and real-time intervention system includes a data collection module, an AI analysis and processing module, a personalized recommendation module, a real-time intervention module, a data storage module, and a human-computer interaction module. Each module is linked bidirectionally through a communication link to form a closed-loop operation system of "collection-analysis-recommendation-intervention-feedback-optimization". The system addresses the shortcomings of existing technologies, such as homogenization of eye behavior correction solutions, delayed intervention timing, lack of individual adaptability in recommendation strategies, and inability to achieve real-time monitoring and dynamic intervention of eye behavior. Through multi-dimensional data fusion analysis and AI intelligent learning, it enables personalized correction solution recommendations and millisecond-level real-time intervention for different users, thereby improving the accuracy and effectiveness of eye behavior correction.

[0010] As a further improvement of the present invention, the data acquisition module includes an image acquisition unit, a physiological parameter acquisition unit, an environmental parameter acquisition unit, and a behavior log acquisition unit. Each acquisition unit operates synchronously and the data is uploaded to the AI ​​analysis and processing module in real time. The image acquisition unit uses a combination of a high-definition camera and an infrared imaging component to capture the user's eye movements, facial posture, and viewing distance in real time. The sampling frequency is no less than 30 frames per second, and it can identify blink frequency, eyelid opening and closing, eyeball rotation angle, and head tilt angle. The physiological parameter acquisition unit collects physiological indicators related to eye fatigue of users through wearable sensors, including electrooculogram signals, intraocular pressure fluctuation data and tear secretion-related parameters. The sampling period can be dynamically adjusted to 1-5 minutes / time according to the user's eye use status. The environmental parameter acquisition unit is used to collect the light intensity, ambient humidity, ambient temperature and blue light intensity of the user's eye environment. The light intensity acquisition range is 0-10000 lux and the blue light intensity acquisition range is 380-450nm wavelength. It can identify adverse eye environments such as strong light, weak light and excessive blue light in real time. The behavior log collection unit automatically records the user's screen time, screen intervals, screen scenarios, and the execution of corrective behaviors, forming a raw database of the user's personal screen behavior. This provides full-dimensional data support for subsequent personalized analysis, which is different from the shortcomings of existing technologies that have a single data collection dimension and lack behavioral trajectory recording.

[0011] As a further improvement of the present invention, the AI ​​analysis and processing module includes a data preprocessing unit, a feature extraction unit, a user profile construction unit, a risk assessment unit, and a model optimization unit. The core is based on an improved deep learning model, which is based on the Transformer architecture and incorporates an attention mechanism and a residual network to solve the problems of slow convergence speed, inaccurate feature extraction, and inability to adapt to multi-dimensional heterogeneous data in existing AI models. The data preprocessing unit performs denoising, normalization, and outlier removal on the raw data uploaded by each acquisition module. It uses the Z-score normalization algorithm to normalize physiological and environmental parameters and employs a median filtering algorithm to remove noise points from the image acquisition data, ensuring data accuracy. The feature extraction unit extracts core features of eye behavior from the preprocessed data, including abnormal eye distance features, abnormal blinking features, abnormal head posture features, eye fatigue features, and abnormal environmental features. It strengthens the extraction weight of key features through an attention mechanism to improve the accuracy of feature recognition. The user profile building unit, based on the extracted core features and combined with the user's basic information and historical eye use data, constructs a personalized user profile that includes the user's eye use habits, eye use preferences, eye tolerance, and correction needs, achieving "one profile per person," which is different from the generalized profile building mode in the existing technology. The risk assessment unit uses a multi-dimensional risk assessment algorithm based on user profiles and real-time eye usage characteristics to classify and assess eye health risks such as myopia progression, eye fatigue, and dry eye syndrome into three levels: low risk, medium risk, and high risk. Each level corresponds to a clear risk judgment standard and intervention priority. The model optimization unit, based on user feedback data, correction effect data, and newly added eye use data, adopts an online incremental learning algorithm to update model parameters in real time, continuously optimize feature extraction accuracy and risk assessment accuracy, and ensure that the model always adapts to the dynamic changes in user eye use behavior, thus solving the shortcomings of existing AI models that are fixed and cannot adapt to changes in user behavior.

[0012] As a further improvement of the present invention, the personalized recommendation module generates a personalized eye behavior correction plan based on the user profile and risk assessment results output by the AI ​​analysis and processing module, combined with the eye health knowledge base. The eye health knowledge base integrates ophthalmological clinical data, eye behavior correction guidelines and personalized adaptation rules. The ophthalmological clinical data includes data on the correlation between eye behavior and eye health of users of different ages and vision conditions. The knowledge base construction logic of existing ophthalmological AI services is referenced and optimized to solve the problems of homogeneous recommendation plans and lack of clinical data support in the existing technology. The personalized correction plan includes basic correction rules, targeted correction training, suggestions for adjusting eye habits, and suggestions for optimizing the environment. The basic correction rules are set according to the user's age and vision condition. For teenagers, the focus is on rules related to myopia prevention and control; for adults, the focus is on rules related to relieving eye fatigue; and for elderly users, the focus is on rules related to vision protection and prevention of dry eye syndrome. The targeted corrective training is customized based on the user's core vision problems. For example, distance adjustment training is recommended for users who use their eyes at too close a distance, and eye relaxation training is recommended for users with abnormal blinking. For users with severe eye fatigue, a combination of warm compresses and eye exercises is recommended. The suggestions for adjusting eye habits include controlling the duration of eye use, setting the interval between eye use sessions, and guiding correct eye posture. The interval between eye use sessions follows the "20-20-20" principle and is dynamically adjusted based on the user's tolerance. The environmental optimization suggestions are generated based on the collected environmental parameters. For example, it is recommended to turn on the supplementary lighting when the light is insufficient, and to turn on the blue light protection mode when the blue light exceeds the standard, so as to ensure that the recommended solutions are targeted and feasible.

[0013] As a further improvement of the present invention, the real-time intervention module includes an intervention triggering unit, a multi-mode intervention unit and an intervention feedback unit, which is used to achieve millisecond-level real-time intervention when the user exhibits poor eye use behavior or exceeds the standard of eye health risk, thereby solving the problems of delayed intervention timing and single intervention method in the prior art. The intervention triggering unit receives real-time eye use characteristics and risk assessment results output by the AI ​​analysis and processing module in real time, sets multi-level intervention triggering thresholds, and immediately triggers an intervention command when it detects that the user's eye use behavior exceeds the normal range or the risk level reaches medium to high risk, with a trigger response time of no more than 100ms. The multi-mode intervention unit includes visual intervention, auditory intervention, tactile intervention, and scene-linked intervention. It can automatically select the intervention mode according to the user scenario, user preferences, and intervention priority. Visual intervention is achieved through the screen display of reminder information and changes in light in the human-computer interaction module. Auditory intervention is achieved through voice reminders and prompts. Tactile intervention is achieved through the vibration of wearable devices. Scene-linked intervention can be linked with smart devices such as electronic devices, lighting devices, and tables and chairs, such as automatically reducing the screen brightness of electronic devices, adjusting the light intensity of lighting devices, and reminding users to adjust the height of tables and chairs. The intervention feedback unit collects user response data to intervention operations in real time and uploads the feedback data to the AI ​​analysis and processing module for model optimization and recommendation adjustment, forming an intervention closed loop.

[0014] As a further improvement of the present invention, the data storage module adopts a distributed storage architecture, which is divided into a raw database, a feature database, a user profile database, a recommendation scheme database, an intervention feedback database, and a model parameter database. It is used to classify and store various types of data during the operation of the system, thereby solving the problems of chaotic data storage, low query efficiency, and poor data security in the prior art. The raw database stores the unprocessed raw data uploaded by each acquisition module, and the storage period is no less than one year. The feature database stores the core feature data of eye use behavior output by the feature extraction unit, and establishes an association index with the original data to facilitate traceability and query. The user profile database stores basic user information, personalized user profiles, and dynamic update records; The recommended solution database stores personalized correction solutions generated for each user and solution update records. The intervention feedback database stores intervention trigger records, intervention methods, user response data, and intervention effect data; The model parameter database stores the initial parameters of the AI ​​analysis and processing module, the updated parameters after incremental learning, and the model version record; The data storage module supports encrypted data storage and hierarchical access control. It uses the AES encryption algorithm to encrypt user privacy data, allowing only authorized users to view and modify personal data, thus ensuring user data security and privacy.

[0015] As a further improvement of the present invention, the training method of the improved deep learning model includes the following steps: S1. Collect multi-dimensional sample data, including user eye image data, physiological parameter data, environmental parameter data and behavior log data of users of different ages, different vision conditions and different eye use scenarios, and construct a sample dataset with no less than 100,000 samples, of which 80% are used as training set and 20% are used as test set. S2. Preprocess the sample dataset using the same processing method as the data preprocessing unit to obtain standardized sample data. S3. Build an initial deep learning model based on the Transformer architecture, and introduce an attention mechanism and a residual network. The attention mechanism is used to enhance the extraction of key features of eye behavior, and the residual network is used to solve the gradient vanishing problem caused by the increase of model depth. S4. Input the preprocessed training set into the initial model for training. Use the cross-entropy loss function to calculate the model prediction error. Use the Adam optimization algorithm to adjust the model parameters. Set the number of training iterations to be no less than 100 rounds until the model prediction accuracy is no less than 95%. S5. Test the trained model using a test set. If the test accuracy is below 95%, adjust the model structure and parameters and retrain until the requirements are met. S6. Deploy the trained model to the AI ​​analysis and processing module, and combine it with the online incremental learning algorithm to continuously optimize the model parameters based on real-time user data to ensure the model's adaptability and accuracy.

[0016] As a further improvement of the present invention, the multi-level intervention trigger threshold of the real-time intervention module adopts a dynamic adjustment mechanism, which is dynamically updated based on user profile, historical intervention effect and eye health status. The specific adjustment logic is as follows: for low-risk users, increase the intervention trigger threshold, reduce the frequency of intervention, and avoid excessive intervention. For medium- to high-risk users, lower the intervention trigger threshold and increase the frequency of interventions to ensure timely intervention; for users who respond well to intervention recommendations, appropriately raise the intervention trigger threshold; for users who respond poorly to intervention recommendations, lower the intervention trigger threshold and optimize the intervention methods. Meanwhile, users can manually adjust the range of intervention trigger thresholds in the human-computer interaction module according to their own needs, so as to realize personalized adaptation of intervention thresholds and solve the problem that the intervention thresholds are fixed in the existing technology and cannot adapt to different user needs.

[0017] As a further improvement of the present invention, the human-computer interaction module includes a display unit, an input unit, a voice interaction unit and a feedback unit, supports multi-terminal adaptation, and solves the problems of single interaction mode and poor terminal adaptability in the prior art. The display unit is used to display user eye usage data, personalized correction plans, intervention reminder information, eye health risk assessment results and model optimization records. It uses clear and easy-to-understand visual charts to show the data change trends, making it easy for users to intuitively understand their own eye usage status. The input unit is used for users to input basic information, correction preferences, intervention method selection, and manual adjustment of intervention thresholds, and supports touch input, keyboard input, and handwriting input. The voice interaction unit supports voice command recognition and voice feedback. Users can use voice commands to query eye use data, start correction training, and adjust intervention modes. The system responds to user commands and provides relevant guidance through voice feedback. The feedback unit is used for users to provide feedback on the effectiveness of the correction plan, the applicability of the intervention method, and problems encountered during system operation. The feedback data is uploaded to the AI ​​analysis and processing module in real time for model optimization and plan adjustment.

[0018] As a further improvement of the present invention, the system also includes a remote linkage module for establishing remote linkage with ophthalmology medical institutions, parents' terminals, and teachers' terminals, thereby solving the problems of lack of professional medical support and disconnect between family and school supervision in the prior art; The remote linkage module can synchronize the user's eye usage data, risk assessment results, and correction plan implementation status to the associated ophthalmologist's terminal in real time. The ophthalmologist can provide the user with professional guidance and suggestions based on the relevant data, and can remotely adjust the correction plan when necessary. For adolescent users, eye use data and intervention status can be synchronized to the parent and teacher terminals, making it convenient for parents and teachers to monitor the user's eye use behavior and the implementation of the correction plan in real time, forming a multi-dimensional correction system of "systematic intervention + professional guidance + family supervision + school supervision"; Meanwhile, the remote linkage module allows users to initiate online consultations with ophthalmologists, enabling timely answers to eye health problems and further enhancing the professionalism and effectiveness of eye behavior correction. Attached Figure Description

[0019] Figure 1 The flowchart of the personalized eye behavior correction AI recommendation and real-time intervention system proposed by Wei in this invention.

[0020] The beneficial effects of this invention are: By constructing a closed-loop operation system to solve the problem of delayed intervention, this invention forms a closed-loop operation system of "collection-analysis-recommendation-intervention-feedback-optimization" through bidirectional linkage of various modules. This enables real-time collection, rapid analysis, accurate recommendation, millisecond-level intervention of eye use behavior, and dynamic optimization based on feedback. It completely solves the defects of delayed intervention timing and inability to achieve real-time linkage in existing technologies, and improves the timeliness and effectiveness of correction.

[0021] Multi-dimensional data collection enhances analysis accuracy: The data collection module covers four dimensions: image, physiology, environment, and behavior logs. The collected parameters are specific and comprehensive, solving the problems of single data collection dimensions and lack of behavior trajectory records in existing technologies. It provides full-dimensional data support for AI analysis and processing, ensuring the accuracy of feature extraction and risk assessment, while avoiding analysis bias caused by insufficient data in AI models.

[0022] An improved AI model enhances adaptability and accuracy: The AI ​​analysis and processing module is based on the Transformer architecture, introducing an attention mechanism and residual network to solve the problems of slow convergence speed, inaccurate feature extraction, and inability to adapt to multi-dimensional heterogeneous data in existing AI models. Combined with an online incremental learning algorithm, it enables real-time updates of model parameters, ensuring that the model always adapts to the dynamic changes in user eye behavior. At the same time, through standardized model training methods, it improves the model's accuracy and generalization ability, with a prediction accuracy of no less than 95%.

[0023] Personalized solutions to suit different user needs: Based on personalized user profiles and risk assessment results, combined with an eye health knowledge base, targeted correction solutions are generated. Solutions are customized according to different age groups, different vision conditions, and different eye problems to achieve "one solution per person". This solves the problems of homogeneity and lack of individual adaptability in existing correction solutions, and improves correction effect and user compliance.

[0024] Multi-mode real-time intervention enhances the flexibility and effectiveness of intervention: The real-time intervention module adopts multi-level trigger thresholds and multi-mode intervention methods with a response time of no more than 100ms. It can flexibly select the intervention mode according to user scenarios, preferences and risk levels. At the same time, it supports dynamic and manual adjustment of intervention thresholds, which solves the problems of single intervention methods and fixed thresholds in existing technologies, reduces user resistance and improves intervention effects.

[0025] Secure and efficient storage to protect user privacy: Adopting a distributed storage architecture, it realizes the classified storage and correlation tracing of various types of data, improving data query efficiency; using AES encryption algorithm and hierarchical permission management, it ensures the security of user privacy data and solves the problems of chaotic data storage and poor security in existing technologies.

[0026] Multi-faceted collaboration to enhance corrective support: Through the remote collaboration module, collaboration with ophthalmology medical institutions, parents, and teachers is achieved to form a multi-faceted corrective system, solving the problems of lack of professional medical support and lack of supervision in existing technologies, and further improving the professionalism and effectiveness of correction. Detailed Implementation

[0027] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0028] The personalized eye behavior correction AI recommendation and real-time intervention system includes a data collection module, an AI analysis and processing module, a personalized recommendation module, a real-time intervention module, a data storage module, and a human-computer interaction module. Each module is linked bidirectionally through a communication link to form a closed-loop operation system of "collection-analysis-recommendation-intervention-feedback-optimization". The Personalized Eye Behavior Correction AI Recommendation and Real-Time Intervention System addresses the shortcomings of existing technologies, such as homogenized eye behavior correction solutions, delayed intervention timing, lack of individualized recommendation strategies, and inability to achieve real-time monitoring and dynamic intervention of eye behavior. Through multi-dimensional data fusion analysis and AI intelligent learning, it enables personalized correction solution recommendations and millisecond-level real-time intervention for different users, improving the accuracy and effectiveness of eye behavior correction.

[0029] The data acquisition module of the personalized eye behavior correction AI recommendation and real-time intervention system includes an image acquisition unit, a physiological parameter acquisition unit, an environmental parameter acquisition unit, and a behavior log acquisition unit. Each acquisition unit operates synchronously and the data is uploaded to the AI ​​analysis and processing module in real time. The personalized eye behavior correction AI recommendation and real-time intervention system's image acquisition unit uses a combination of a high-definition camera and an infrared imaging component to capture the user's eye movements, facial posture, and eye distance in real time, with a sampling frequency of no less than 30 frames per second. It can identify blink frequency, eyelid opening and closing degree, eyeball rotation angle, and head tilt angle. The personalized eye behavior correction AI recommendation and real-time intervention system's physiological parameter acquisition unit collects users' eye fatigue-related physiological indicators through wearable sensors, including electrooculogram signals, intraocular pressure fluctuation data, and tear secretion-related parameters. The sampling cycle can be dynamically adjusted to 1-5 minutes / time according to the user's eye use status. The personalized eye behavior correction AI recommendation and real-time intervention system's environmental parameter acquisition unit is used to collect the light intensity, ambient humidity, ambient temperature, and blue light intensity of the user's eye environment. The light intensity acquisition range is 0-10000 lux, and the blue light intensity acquisition range is 380-450nm wavelength. It can identify poor eye environments such as strong light, weak light, and excessive blue light in real time. The AI ​​recommendation and real-time intervention system for personalized eye use behavior correction automatically records the user's eye use duration, intervals, scenarios, and the execution of corrective behaviors through its behavior log collection unit. This forms a raw database of the user's personal eye use behavior, providing comprehensive data support for subsequent personalized analysis. This differs from existing technologies that suffer from single data collection dimensions and a lack of behavioral trajectory recording.

[0030] The AI ​​analysis and processing module of the personalized eye behavior correction AI recommendation and real-time intervention system includes a data preprocessing unit, a feature extraction unit, a user profile construction unit, a risk assessment unit, and a model optimization unit. Its core is based on an improved deep learning model. This improved deep learning model is based on the Transformer architecture, incorporating an attention mechanism and a residual network to address the problems of slow convergence speed, inaccurate feature extraction, and inability to adapt to multi-dimensional heterogeneous data in existing AI models. The data preprocessing unit performs denoising, normalization, and outlier removal on the raw data uploaded by each acquisition module. It uses the Z-score normalization algorithm to normalize physiological and environmental parameters and employs a median filtering algorithm to remove noise points from the image acquisition data, ensuring data accuracy. The feature extraction unit of the personalized eye behavior correction AI recommendation and real-time intervention system extracts core features of eye behavior from preprocessed data, including abnormal eye distance features, abnormal blinking features, abnormal head posture features, eye fatigue features, and abnormal environmental features. It strengthens the extraction weight of key features through the attention mechanism to improve the accuracy of feature recognition. The user profile building unit of the personalized eye behavior correction AI recommendation and real-time intervention system is based on the extracted core features and combined with the user's basic information and historical eye use data to build a personalized user profile that includes the user's eye use habits, eye use preferences, eye tolerance, and correction needs, achieving "one profile per person", which is different from the generalized profile building mode in the existing technology. The risk assessment unit of the personalized eye behavior correction AI recommendation and real-time intervention system uses a multi-dimensional risk assessment algorithm based on user profiles and real-time eye characteristics to classify and assess users' eye health risks such as myopia progression, eye fatigue, and dry eye syndrome into three levels: low risk, medium risk, and high risk. Each level corresponds to a clear risk judgment standard and intervention priority. The personalized eye behavior correction AI recommendation and real-time intervention system model optimization unit is based on user feedback data, correction effect data, and new eye behavior data. It adopts an online incremental learning algorithm to update model parameters in real time, continuously optimize feature extraction accuracy and risk assessment accuracy, and ensure that the model always adapts to the dynamic changes in user eye behavior. This solves the shortcomings of existing AI models that are fixed and cannot adapt to changes in user behavior.

[0031] The personalized eye use behavior correction AI recommendation and real-time intervention system's personalized recommendation module generates personalized eye use behavior correction plans based on user profiles and risk assessment results output by the AI ​​analysis and processing module, combined with an eye health knowledge base. The eye health knowledge base of the personalized eye use behavior correction AI recommendation and real-time intervention system integrates ophthalmological clinical data, eye use behavior correction guidelines, and personalized adaptation rules. The ophthalmological clinical data includes data on the correlation between eye use behavior and eye health for users of different ages and vision conditions. It references and optimizes the knowledge base construction logic of existing ophthalmological AI services to solve the problems of homogenized recommendation plans and lack of clinical data support in existing technologies. The personalized eye behavior correction AI recommendation and real-time intervention system includes basic correction rules, targeted correction training, suggestions for adjusting eye habits, and suggestions for environmental optimization. The basic correction rules are set according to the user's age and vision status. For teenagers, the system mainly includes rules related to myopia prevention and control; for adults, the system mainly includes rules related to relieving eye fatigue; and for elderly users, the system mainly includes rules related to vision protection and dry eye prevention. The personalized eye behavior correction AI recommendation and real-time intervention system provides targeted correction training based on the user's core eye problems. For example, for users who use their eyes at too close a distance, distance adjustment training is recommended; for users with abnormal blinking, eye relaxation training is recommended. For users with severe eye fatigue, a combination of warm compresses and eye exercises is recommended. The personalized eye behavior correction AI recommendation and real-time intervention system provides suggestions for adjusting eye habits, including controlling the duration of eye use, setting the interval between eye use sessions, and guiding correct eye posture. The interval between eye use sessions follows the "20-20-20" principle and is dynamically adjusted based on the user's tolerance. The personalized eye behavior correction AI recommendation and real-time intervention system generates environmental optimization suggestions based on the collected environmental parameters. For example, it recommends turning on supplemental lighting when there is insufficient light and turning on blue light protection mode when blue light exceeds the standard, ensuring that the recommended solutions are targeted and feasible.

[0032] The AI ​​recommendation and real-time intervention system for personalized eye behavior correction includes an intervention triggering unit, a multi-mode intervention unit, and an intervention feedback unit. It is used to achieve millisecond-level real-time intervention when users exhibit poor eye behavior or exceed the standard for eye health risks, thus solving the problems of delayed intervention timing and single intervention methods in existing technologies. The AI ​​recommendation and real-time intervention system for personalized eye behavior correction receives real-time eye characteristics and risk assessment results from the AI ​​analysis and processing module in real time. It sets multi-level intervention trigger thresholds and immediately triggers intervention instructions when it detects that the user's eye behavior exceeds the normal range or the risk level reaches medium to high risk. The trigger response time does not exceed 100ms. The personalized eye behavior correction AI recommendation and real-time intervention system has a multi-mode intervention unit that includes visual intervention, auditory intervention, tactile intervention, and scene-linked intervention. It can automatically select the intervention mode according to the user scenario, user preferences, and intervention priority. Visual intervention is achieved through the human-computer interaction module's screen display reminder information and light changes; auditory intervention is achieved through voice reminders and prompts; tactile intervention is achieved through the vibration of wearable devices; and scene-linked intervention can be linked with smart devices such as electronic devices, lighting devices, and tables and chairs, such as automatically reducing the brightness of electronic device screens, adjusting the light intensity of lighting devices, and reminding users to adjust the height of tables and chairs. The personalized eye behavior correction AI recommendation and real-time intervention system's intervention feedback unit collects user response data to intervention operations in real time and uploads the feedback data to the AI ​​analysis and processing module for model optimization and recommendation adjustment, forming an intervention closed loop.

[0033] The data storage module of the personalized eye behavior correction AI recommendation and real-time intervention system adopts a distributed storage architecture, which is divided into a raw database, a feature database, a user profile database, a recommendation scheme database, an intervention feedback database, and a model parameter database. This is used to classify and store various types of data during the system operation, solving the problems of chaotic data storage, low query efficiency, and poor data security in existing technologies. The raw database of the personalized eye behavior correction AI recommendation and real-time intervention system stores the raw data uploaded by each collection module without preprocessing, and the storage period is no less than 1 year. The feature database of the personalized eye behavior correction AI recommendation and real-time intervention system stores the core feature data of eye behavior output by the feature extraction unit, and establishes an association index with the original data to facilitate traceability and query. The user profile database of the personalized eye behavior correction AI recommendation and real-time intervention system stores basic user information, personalized user profiles, and dynamic update records. The personalized eye behavior correction AI recommendation and real-time intervention system's recommendation solution database stores the personalized correction solutions generated for each user and the solution update records; The personalized eye behavior correction AI recommendation and real-time intervention system's intervention feedback database stores intervention trigger records, intervention methods, user response data, and intervention effect data; The model parameter database of the personalized eye behavior correction AI recommendation and real-time intervention system stores the initial parameters of the AI ​​analysis and processing module, the updated parameters after incremental learning, and the model version record. The data storage module of the personalized eye behavior correction AI recommendation and real-time intervention system supports encrypted data storage and hierarchical access control. It uses the AES encryption algorithm to encrypt user privacy data, and only authorized users can view and modify personal data, ensuring user data security and privacy.

[0034] The training method for the improved deep learning model of the personalized eye behavior correction AI recommendation and real-time intervention system includes the following steps: S1. Collect multi-dimensional sample data, including user eye image data, physiological parameter data, environmental parameter data and behavior log data of users of different ages, different vision conditions and different eye use scenarios, and construct a sample dataset with no less than 100,000 samples, of which 80% are used as training set and 20% are used as test set. S2. Preprocess the sample dataset using the same processing method as the data preprocessing unit to obtain standardized sample data. S3. Build an initial deep learning model based on the Transformer architecture, and introduce an attention mechanism and a residual network. The attention mechanism is used to enhance the extraction of key features of eye behavior, and the residual network is used to solve the gradient vanishing problem caused by the increase of model depth. S4. Input the preprocessed training set into the initial model for training. Use the cross-entropy loss function to calculate the model prediction error. Use the Adam optimization algorithm to adjust the model parameters. Set the number of training iterations to be no less than 100 rounds until the model prediction accuracy is no less than 95%. S5. Test the trained model using a test set. If the test accuracy is below 95%, adjust the model structure and parameters and retrain until the requirements are met. S6. Deploy the trained model to the AI ​​analysis and processing module, and combine it with the online incremental learning algorithm to continuously optimize the model parameters based on real-time user data to ensure the model's adaptability and accuracy.

[0035] The AI ​​recommendation and real-time intervention system for personalized eye behavior correction adopts a dynamic adjustment mechanism for the multi-level intervention trigger threshold of the real-time intervention module. It is dynamically updated based on user profile, historical intervention effects and eye health status. The specific adjustment logic is as follows: for low-risk users, the intervention trigger threshold is increased, the frequency of intervention is reduced, and over-intervention is avoided. For medium- to high-risk users, lower the intervention trigger threshold and increase the frequency of interventions to ensure timely intervention; for users who respond well to intervention recommendations, appropriately raise the intervention trigger threshold; for users who respond poorly to intervention recommendations, lower the intervention trigger threshold and optimize the intervention methods. Meanwhile, users can manually adjust the range of intervention trigger thresholds in the human-computer interaction module according to their own needs, so as to realize personalized adaptation of intervention thresholds and solve the problem that the intervention thresholds are fixed in the existing technology and cannot adapt to different user needs.

[0036] The AI ​​recommendation and real-time intervention system for personalized eye behavior correction includes a human-computer interaction module comprising a display unit, an input unit, a voice interaction unit, and a feedback unit. It supports multi-terminal adaptation and solves the problems of single interaction methods and poor terminal adaptability in existing technologies. The display unit of the personalized eye behavior correction AI recommendation and real-time intervention system is used to display user eye data, personalized correction plans, intervention reminders, eye health risk assessment results and model optimization records. It uses clear and easy-to-understand visual charts to show the data change trends, making it easy for users to intuitively understand their own eye use status. The input unit of the personalized eye behavior correction AI recommendation and real-time intervention system is used by users to input basic information, correction preferences, intervention method selection, and manual adjustment of intervention thresholds. It supports touch input, keyboard input, and handwriting input. The personalized eye behavior correction AI recommendation and real-time intervention system's voice interaction unit supports voice command recognition and voice feedback. Users can use voice commands to query eye use data, start correction training, and adjust intervention modes. The system responds to user commands and provides relevant guidance through voice feedback. The feedback unit of the personalized eye behavior correction AI recommendation and real-time intervention system is used for user feedback on the effectiveness of the correction plan, the applicability of the intervention method, and problems encountered during system operation. The feedback data is uploaded to the AI ​​analysis and processing module in real time for model optimization and plan adjustment.

[0037] The personalized eye behavior correction AI recommendation and real-time intervention system also includes a remote linkage module, which is used to establish remote linkage with ophthalmology medical institutions, parents, and teachers to solve the problems of lack of professional medical support and disconnect between family and school supervision in existing technologies. The personalized eye behavior correction AI recommendation and real-time intervention system remote linkage module can synchronize the user's eye data, risk assessment results, and correction plan implementation status to the associated ophthalmologist terminal in real time. Ophthalmologists can provide users with professional guidance and suggestions based on relevant data, and can remotely adjust the correction plan when necessary. For adolescent users, eye use data and intervention status can be synchronized to the parent and teacher terminals, making it convenient for parents and teachers to monitor the user's eye use behavior and the implementation of the correction plan in real time, forming a multi-dimensional correction system of "systematic intervention + professional guidance + family supervision + school supervision"; Meanwhile, the remote linkage module allows users to initiate online consultations with ophthalmologists, enabling timely answers to eye health problems and further enhancing the professionalism and effectiveness of eye behavior correction.

[0038] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A personalized AI recommendation and real-time intervention system for correcting eye-use behavior, characterized in that, It includes a data acquisition module, an AI analysis and processing module, a personalized recommendation module, a real-time intervention module, a data storage module, and a human-computer interaction module. Each module is linked bidirectionally through a communication link to form a closed-loop operation system of "collection-analysis-recommendation-intervention-feedback-optimization". The system addresses the shortcomings of existing technologies, such as homogenization of eye behavior correction solutions, delayed intervention timing, lack of individual adaptability in recommendation strategies, and inability to achieve real-time monitoring and dynamic intervention of eye behavior. Through multi-dimensional data fusion analysis and AI intelligent learning, it enables personalized correction solution recommendations and millisecond-level real-time intervention for different users, thereby improving the accuracy and effectiveness of eye behavior correction.

2. The personalized eye behavior correction AI recommendation and real-time intervention system according to claim 1, characterized in that, The data acquisition module includes an image acquisition unit, a physiological parameter acquisition unit, an environmental parameter acquisition unit, and a behavior log acquisition unit. Each acquisition unit operates synchronously and the data is uploaded to the AI ​​analysis and processing module in real time. The image acquisition unit uses a combination of a high-definition camera and an infrared imaging component to capture the user's eye movements, facial posture, and viewing distance in real time. The sampling frequency is no less than 30 frames per second, and it can identify blink frequency, eyelid opening and closing, eyeball rotation angle, and head tilt angle. The physiological parameter acquisition unit collects physiological indicators related to eye fatigue of users through wearable sensors, including electrooculogram signals, intraocular pressure fluctuation data and tear secretion-related parameters. The sampling period can be dynamically adjusted to 1-5 minutes / time according to the user's eye use status. The environmental parameter acquisition unit is used to collect the light intensity, ambient humidity, ambient temperature and blue light intensity of the user's eye environment. The light intensity acquisition range is 0-10000 lux and the blue light intensity acquisition range is 380-450nm wavelength. It can identify adverse eye environments such as strong light, weak light and excessive blue light in real time. The behavior log collection unit automatically records the user's screen time, screen intervals, screen scenarios, and the execution of corrective behaviors, forming a raw database of the user's personal screen behavior. This provides full-dimensional data support for subsequent personalized analysis, which is different from the shortcomings of existing technologies that have a single data collection dimension and lack behavioral trajectory recording.

3. The personalized eye behavior correction AI recommendation and real-time intervention system according to claim 1, characterized in that, The AI ​​analysis and processing module includes a data preprocessing unit, a feature extraction unit, a user profile construction unit, a risk assessment unit, and a model optimization unit. Its core is based on an improved deep learning model, which uses the Transformer architecture and incorporates an attention mechanism and residual network to address the problems of slow convergence speed, inaccurate feature extraction, and inability to adapt to multi-dimensional heterogeneous data in existing AI models. The data preprocessing unit performs denoising, normalization, and outlier removal on the raw data uploaded by each acquisition module. It uses the Z-score normalization algorithm to normalize physiological and environmental parameters and employs a median filtering algorithm to remove noise points from the image acquisition data, ensuring data accuracy. The feature extraction unit extracts core features of eye behavior from the preprocessed data, including abnormal eye distance features, abnormal blinking features, abnormal head posture features, eye fatigue features, and abnormal environmental features. It strengthens the extraction weight of key features through an attention mechanism to improve the accuracy of feature recognition. The user profile building unit, based on the extracted core features and combined with the user's basic information and historical eye use data, constructs a personalized user profile that includes the user's eye use habits, eye use preferences, eye tolerance, and correction needs, achieving "one profile per person," which is different from the generalized profile building mode in the existing technology. The risk assessment unit uses a multi-dimensional risk assessment algorithm based on user profiles and real-time eye usage characteristics to classify and assess eye health risks such as myopia progression, eye fatigue, and dry eye syndrome into three levels: low risk, medium risk, and high risk. Each level corresponds to a clear risk judgment standard and intervention priority. The model optimization unit, based on user feedback data, correction effect data, and newly added eye use data, adopts an online incremental learning algorithm to update model parameters in real time, continuously optimize feature extraction accuracy and risk assessment accuracy, and ensure that the model always adapts to the dynamic changes in user eye use behavior, thus solving the shortcomings of existing AI models that are fixed and cannot adapt to changes in user behavior.

4. The personalized eye behavior correction AI recommendation and real-time intervention system according to claim 1, characterized in that, The personalized recommendation module generates personalized eye behavior correction plans based on user profiles and risk assessment results output by the AI ​​analysis and processing module, combined with an eye health knowledge base. The eye health knowledge base integrates ophthalmological clinical data, eye behavior correction guidelines, and personalized adaptation rules. The ophthalmological clinical data includes data on the correlation between eye behavior and eye health for users of different ages and vision conditions. The knowledge base construction logic of existing ophthalmological AI services is referenced and optimized to solve the problems of homogeneous recommendation plans and lack of clinical data support in existing technologies. The personalized correction plan includes basic correction rules, targeted correction training, suggestions for adjusting eye habits, and suggestions for optimizing the environment. The basic correction rules are set according to the user's age and vision condition. For teenagers, the focus is on rules related to myopia prevention and control; for adults, the focus is on rules related to relieving eye fatigue; and for elderly users, the focus is on rules related to vision protection and prevention of dry eye syndrome. The targeted corrective training is customized based on the user's core vision problems. For example, distance adjustment training is recommended for users who use their eyes at too close a distance, and eye relaxation training is recommended for users with abnormal blinking. For users with severe eye fatigue, a combination of warm compresses and eye exercises is recommended. The suggestions for adjusting eye habits include controlling the duration of eye use, setting the interval between eye use sessions, and guiding correct eye posture. The interval between eye use sessions follows the "20-20-20" principle and is dynamically adjusted based on the user's tolerance. The environmental optimization suggestions are generated based on the collected environmental parameters. For example, it is recommended to turn on the supplementary lighting when the light is insufficient, and to turn on the blue light protection mode when the blue light exceeds the standard, so as to ensure that the recommended solutions are targeted and feasible.

5. The personalized eye behavior correction AI recommendation and real-time intervention system according to claim 1, characterized in that, The real-time intervention module includes an intervention triggering unit, a multi-mode intervention unit, and an intervention feedback unit. It is used to achieve millisecond-level real-time intervention when users exhibit poor eye-use behavior or exceed the standard for eye health risks, thus solving the problems of delayed intervention timing and single intervention methods in the existing technology. The intervention triggering unit receives real-time eye use characteristics and risk assessment results output by the AI ​​analysis and processing module in real time, sets multi-level intervention triggering thresholds, and immediately triggers an intervention command when it detects that the user's eye use behavior exceeds the normal range or the risk level reaches medium to high risk, with a trigger response time of no more than 100ms. The multi-mode intervention unit includes visual intervention, auditory intervention, tactile intervention, and scene-linked intervention. It can automatically select the intervention mode according to the user scenario, user preferences, and intervention priority. Visual intervention is achieved through the screen display of reminder information and changes in light in the human-computer interaction module. Auditory intervention is achieved through voice reminders and prompts. Tactile intervention is achieved through the vibration of wearable devices. Scene-linked intervention can be linked with smart devices such as electronic devices, lighting devices, and tables and chairs, such as automatically reducing the screen brightness of electronic devices, adjusting the light intensity of lighting devices, and reminding users to adjust the height of tables and chairs. The intervention feedback unit collects user response data to intervention operations in real time and uploads the feedback data to the AI ​​analysis and processing module for model optimization and recommendation adjustment, forming an intervention closed loop.

6. The personalized eye behavior correction AI recommendation and real-time intervention system according to claim 1, characterized in that, The data storage module adopts a distributed storage architecture, which is divided into a raw database, a feature database, a user profile database, a recommendation scheme database, an intervention feedback database, and a model parameter database. It is used to classify and store various types of data during the operation of the system, and solves the problems of chaotic data storage, low query efficiency, and poor data security in the existing technology. The raw database stores the unprocessed raw data uploaded by each acquisition module, and the storage period is no less than one year. The feature database stores the core feature data of eye use behavior output by the feature extraction unit, and establishes an association index with the original data to facilitate traceability and query. The user profile database stores basic user information, personalized user profiles, and dynamic update records; The recommended solution database stores personalized correction solutions generated for each user and solution update records. The intervention feedback database stores intervention trigger records, intervention methods, user response data, and intervention effect data; The model parameter database stores the initial parameters of the AI ​​analysis and processing module, the updated parameters after incremental learning, and the model version record; The data storage module supports encrypted data storage and hierarchical access control. It uses the AES encryption algorithm to encrypt user privacy data, allowing only authorized users to view and modify personal data, thus ensuring user data security and privacy.

7. The personalized eye behavior correction AI recommendation and real-time intervention system according to claim 3, characterized in that, The training method for the improved deep learning model includes the following steps: S1. Collect multi-dimensional sample data, including user eye image data, physiological parameter data, environmental parameter data and behavior log data of users of different ages, different vision conditions and different eye use scenarios, and construct a sample dataset with no less than 100,000 samples, of which 80% are used as training set and 20% are used as test set. S2. Preprocess the sample dataset using the same processing method as the data preprocessing unit to obtain standardized sample data. S3. Build an initial deep learning model based on the Transformer architecture, and introduce an attention mechanism and a residual network. The attention mechanism is used to enhance the extraction of key features of eye behavior, and the residual network is used to solve the gradient vanishing problem caused by the increase of model depth. S4. Input the preprocessed training set into the initial model for training. Use the cross-entropy loss function to calculate the model prediction error. Use the Adam optimization algorithm to adjust the model parameters. Set the number of training iterations to be no less than 100 rounds until the model prediction accuracy is no less than 95%. S5. Test the trained model using a test set. If the test accuracy is below 95%, adjust the model structure and parameters and retrain until the requirements are met. S6. Deploy the trained model to the AI ​​analysis and processing module, and combine it with the online incremental learning algorithm to continuously optimize the model parameters based on real-time user data to ensure the model's adaptability and accuracy.

8. The personalized eye behavior correction AI recommendation and real-time intervention system according to claim 5, characterized in that, The multi-level intervention trigger threshold of the real-time intervention module adopts a dynamic adjustment mechanism, which is dynamically updated based on user profile, historical intervention effects and eye health status. The specific adjustment logic is as follows: for low-risk users, increase the intervention trigger threshold, reduce the frequency of intervention, and avoid over-intervention. For medium- to high-risk users, lower the intervention trigger threshold and increase the frequency of interventions to ensure timely intervention; for users who respond well to intervention recommendations, appropriately raise the intervention trigger threshold; for users who respond poorly to intervention recommendations, lower the intervention trigger threshold and optimize the intervention methods. Meanwhile, users can manually adjust the range of intervention trigger thresholds in the human-computer interaction module according to their own needs, so as to realize personalized adaptation of intervention thresholds and solve the problem that the intervention thresholds are fixed in the existing technology and cannot adapt to different user needs.

9. The personalized eye behavior correction AI recommendation and real-time intervention system according to claim 1, characterized in that, The human-computer interaction module includes a display unit, an input unit, a voice interaction unit, and a feedback unit, and supports multi-terminal adaptation, solving the problems of single interaction methods and poor terminal adaptability in the prior art. The display unit is used to display user eye usage data, personalized correction plans, intervention reminder information, eye health risk assessment results and model optimization records. It uses clear and easy-to-understand visual charts to show the data change trends, making it easy for users to intuitively understand their own eye usage status. The input unit is used for users to input basic information, correction preferences, intervention method selection, and manual adjustment of intervention thresholds, and supports touch input, keyboard input, and handwriting input. The voice interaction unit supports voice command recognition and voice feedback. Users can use voice commands to query eye use data, start correction training, and adjust intervention modes. The system responds to user commands and provides relevant guidance through voice feedback. The feedback unit is used for users to provide feedback on the effectiveness of the correction plan, the applicability of the intervention method, and problems encountered during system operation. The feedback data is uploaded to the AI ​​analysis and processing module in real time for model optimization and plan adjustment.

10. The personalized eye behavior correction AI recommendation and real-time intervention system according to any one of claims 1-9, characterized in that, The system also includes a remote linkage module, which is used to establish remote linkage with ophthalmology medical institutions, parents' terminals, and teachers' terminals, to solve the problems of lack of professional medical support and disconnect between family and school supervision in existing technologies. The remote linkage module can synchronize the user's eye usage data, risk assessment results, and correction plan implementation status to the associated ophthalmologist's terminal in real time. The ophthalmologist can provide the user with professional guidance and suggestions based on the relevant data, and can remotely adjust the correction plan when necessary. For adolescent users, eye use data and intervention status can be synchronized to the parent and teacher terminals, making it convenient for parents and teachers to monitor the user's eye use behavior and the implementation of the correction plan in real time, forming a multi-dimensional correction system of "systematic intervention + professional guidance + family supervision + school supervision"; Meanwhile, the remote linkage module allows users to initiate online consultations with ophthalmologists, enabling timely answers to eye health problems and further enhancing the professionalism and effectiveness of eye behavior correction.

Images