Vision training system and method based on vibration function and myopia function association model

Abstract

The application discloses a vision training system and method based on a vibration function and myopia function association model, and belongs to the technical field of vision training and myopia prevention and control. The system comprises a whole-body sound wave vertical pulsation device, a myopia parameter measurement module and a central processing module, a quantitative association model of a multi-dimensional vibration function and a myopia function conforming to the physiological state of the eyes is established by constructing the two functions. The application adopts an inverse optimization algorithm combined with regularization constraints to automatically generate a personalized vibration training prescription adapted to the individual state of a user, and simultaneously relies on incremental learning to realize online updating of the model, and is matched with adaptive regulation and control, myopia trend prediction and a multi-mode vibration adaptation mechanism, and can support independent and parallel training of multiple people. The application solves the disadvantages of fixed parameters, poor targeting and no quantitative basis of traditional vibration vision training, realizes digitalization, personalization and closed-loop adaptive intervention of vision training, and can be applied to myopia prevention and control and vision fatigue rehabilitation scenes of people in the age group of adolescents.

Description

Technical Field

[0001] This invention belongs to the field of vision training and myopia prevention technology, specifically relating to a vision training system and method based on a correlation model between vibration function and myopia function. Background Technology

[0002] Myopia in adolescents is showing a trend of developing at younger ages and becoming more severe. The core pathogenesis of myopia is excessive elongation of the axial length, which is closely related to long-term ciliary muscle spasm and insufficient choroidal blood perfusion. Whole-body sound wave vertical rhythmic vision training devices, with their advantages of safety and convenience, have been widely used in myopia prevention and control and eye strain relief. However, existing devices relax eye muscles and improve ocular microcirculation through fixed mechanical vibrations, achieving only basic eye strain relief. They cannot precisely regulate the core pathogenesis of myopia, resulting in multiple technical shortcomings such as low training accuracy, poor individual adaptability, and insufficient long-term intervention effects.

[0003] Existing whole-body acoustic vertical rhythmic myopia intervention devices generally employ standardized vibration output modes with fixed parameters. Their core flaw lies in the lack of a quantitative mathematical correlation model between vibration parameters and myopia physiological indicators, resulting in a lack of intelligent and precise adaptive control capabilities throughout the training system. This technical deficiency prevents traditional devices from matching personalized optimal vibration parameters to the user's ocular physiological indicators and individual characteristics. They cannot form a closed-loop training mechanism for real-time feedback adjustment based on training results, nor can they rely on long-term training data to predict myopia progression trends and dynamically iterate and optimize intervention programs. The overall training method is rigid, individual adaptability is poor, and the intervention effect is difficult to precisely control, failing to meet the personalized, long-term, and precise myopia rehabilitation training needs of different users.

[0004] Therefore, existing whole-body acoustic vertical rhythmic vision training devices lack a correlation model between vibration function and myopia function, resulting in systemic technical problems such as unscientific training parameters, lack of individualized adaptation capabilities, lack of real-time closed-loop control, and lack of long-term dynamic optimization mechanisms. Consequently, their intelligence and intervention accuracy are severely insufficient. Based on this, this invention proposes a vision training method and system based on a correlation model between vibration function and myopia function to solve the aforementioned technical problems. Summary of the Invention

[0005] To address the aforementioned technical problems, in a first aspect, this invention provides a vision training system based on a correlation model between a vibration function and a myopia function, comprising a whole-body acoustic vertical vibration device, a myopia parameter measurement module, and a central processing module; the whole-body acoustic vertical vibration device generates vertical sinusoidal vibrations with a frequency range of 2-50Hz and an amplitude of 2-5mm; the frequency range and the amplitude can be adjusted in real time; the myopia parameter measurement module measures the user's axial length, refractive power, choroidal thickness, and accommodation amplitude; the central processing module establishes a vibration function by collecting data generated by the whole-body acoustic vertical vibration device, establishes a myopia function by collecting user data measured by the myopia parameter measurement module, establishes a correlation model between the vibration function and the myopia function, and generates a vibration prescription based on the correlation model.

[0006] The specific expression for the vibration function constructed in this invention is as follows:

[0007]

[0008] In the formula, Let be the vibration function. It is a frequency function. Let be the amplitude function. For intensity function, The duration parameter; where frequency Continuously adjustable within the 2-50Hz range, amplitude Continuously adjustable within the range of 2-5mm, strength Adjustable within the range of 0-120, and the duration T is adjustable within the range of 5-30 minutes;

[0009] The myopia function satisfies:

[0010]

[0011] In the formula, For myopia function, axial length, For refractive power, The thickness of the choroid. To adjust the amplitude.

[0012] The correlation model F constructed in this invention satisfies the quantization formula:

[0013]

[0014] In the formula, This represents the change in eye condition. This is a model relating the vibration function and the myopia function. It is a set of multidimensional vectors of vibration functions. For the user's eye physiological parameters. For the user's individual characteristic parameters, This is a correction term for the model fitting error; the correlation model is implemented by fusing one or more methods from linear regression, support vector regression, random forest regression, or neural networks.

[0015] This invention generates a vibration prescription using an inverse optimization algorithm: based on the desired myopia improvement target. The solution formula is:

[0016]

[0017] in, For the optimal combination of vibration prescription parameters, It is a set of multidimensional vectors of vibration functions. This is a model relating the vibration function and the myopia function. For the user's eye physiological parameters. For the user's individual characteristic parameters, To pre-determine the amount of change in the target for myopia improvement, The regularization coefficient is . For regularization terms, To constrain the range of vibration parameter values, and to ensure the safety and rationality of vibration parameters.

[0018] Furthermore, it also includes an adaptive adjustment module, which monitors the user's myopia function in real time or periodically, and calculates the deviation using the following formula:

[0019]

[0020] In the formula, The deviation between measured data and predicted data, To monitor the myopia function, To predict the myopia function, i.e. the predicted value of myopia status, when the deviation exceeds a preset threshold, the online update of the correlation model and the dynamic adjustment of the vibration prescription are triggered.

[0021] Furthermore, the association model employs an incremental learning method to achieve online updates, with the update formula being:

[0022]

[0023] In the formula, For the updated association model, The original association model. The learning rate for incremental learning. The gradient operator of the loss function with respect to the correlation model. For loss function, It is a set of multidimensional vectors of vibration functions. For the user's eye physiological parameters. This represents the actual change in myopia status measured after a single training session.

[0024] Furthermore, the system also includes a long-term trend prediction module, which predicts the long-term trend of the myopia function based on an association model:

[0025]

[0026] in, Let be the myopia function predicted at time t. At the initial moment, Let be the initial myopia function at the initial time. For integration time series variables, the integral term represents The cumulative changes in eye condition caused by vibration throughout the entire time period. for The vibration function vector at any moment, for The vector of the myopia function at any given time. For the user's individual characteristic parameters, This is a model relating the vibration function and the myopia function.

[0027] Furthermore, the frequency function f(t) supports one or more of the following modes: constant frequency mode, linear scanning mode, periodic modulation mode, and preset program mode.

[0028] Furthermore, the whole-body acoustic vertical rhythm device is a multi-user compatible device that can accommodate 2-6 people to use at the same time. The central processing module can independently establish a correlation model for each user and generate a vibration prescription.

[0029] Secondly, the present invention provides a vision training method for a vision training system based on a correlation model between vibration function and myopia function, comprising the following steps:

[0030] SS1. Initial Data Acquisition: Measure the user's initial myopia function, satisfying the following formula:

[0031]

[0032] In the formula, This is the initial myopia function. The initial axial length, Initial refractive power, The initial choroid thickness, To determine the initial adjustment range, the user's individual feature vector X is collected.

[0033] SS2. Goal Setting: Enter the user's myopia improvement goals and determine the target change in myopia function. ;

[0034] SS3. Prescription Generation: Based on a Pre-trained Association Model A personalized optimal vibration prescription is generated by inverse optimization. ;

[0035] SS4. Precision Training: Control the whole-body sound wave vertical rhythm device according to the vibration prescription. It delivers precise vibrational stimulation to complete a single training session;

[0036] SS5. Effect Detection: After training, collect the user's real-time myopia function. Calculate the measured change in myopia. ;

[0037] SS6. Model Iteration: Incorporate the current training data into the dataset and update the associated model through incremental learning. ;

[0038] SS7. Closed-loop iteration: Iterate the above steps repeatedly to achieve closed-loop adaptive vision training and dynamic myopia intervention.

[0039] Furthermore, the training method for the correlation model F is as follows: collect N sets of training data:

[0040] ,

[0041] In the formula, For training the model dataset, This represents the total number of training samples. For the first The vibration function vector corresponding to the group of samples, For the first The myopia vibration function vector corresponding to the group of samples, For the first The set of individual user characteristic parameters corresponding to the group samples. For the first The measured deviation value of myopia change corresponding to the group of samples;

[0042] The dataset is divided into training and test sets in a 7:3 ratio. The prediction error is minimized on the training set using the following formula:

[0043]

[0044] In the formula, The model F is the minimum value. For training sample numbers, This represents the total number of training samples. For the first The vibration function vector corresponding to the group of samples, For the first The myopia vibration function vector corresponding to the group of samples, For the first The set of individual user characteristic parameters corresponding to the group samples. For the first The measured deviation value of myopia change corresponding to the group of samples.

[0045] Furthermore, the inverse optimization solution process is subject to multiple constraints, including: , , files, .

[0046] Furthermore, the choroidal thickness C(t) is measured using frequency-domain optical coherence tomography, and the axial length... The adjustment amplitude was measured using an optical biometer. Measurements were taken using the proximity method or the negative lens method.

[0047] Furthermore, the vibration prescription includes frequency variation curves, intensity variation curves, total duration of a single training session, and daily training frequency recommendations.

[0048] The beneficial effects of this invention are:

[0049] 1. This invention constructs a vibration function corresponding to the vertical vibration parameters of whole-body sound waves and a myopia function corresponding to the physiological indicators of myopia in the eyes, and builds a quantitative coupling mathematical model for the two. It abandons the traditional, extensive training mode that relies on manual experience to preset fixed parameters, providing a rigorous basis for vibration intervention and eye function regulation. This makes the vision training process visible, calculable, and predictable, greatly improving the scientific rigor and predictability of myopia-assisted training.

[0050] 2. This invention, based on an inverse correlation model algorithm, can automatically determine the optimal vibration prescription by combining real-time ocular physiological states such as the user's axial length and refractive error, while simultaneously matching multi-dimensional individual characteristic parameters such as age, gender, height, weight, skeletal structure, degree of myopia, and ocular tolerance. This addresses the industry pain points of traditional devices—namely, a one-size-fits-all approach, fixed parameters, and poor adaptability—and enables differentiated, precise, and customized visual rhythm training interventions for different users.

[0051] 3. This invention constructs a complete closed-loop control system of "training-detection-evaluation-adjustment," overcoming the shortcomings of traditional equipment's open-loop fixed operation and lack of dynamic feedback adjustment mechanisms. The system can collect changes in ocular physiological parameters before and after training in real time, dynamically capture deviations in ocular functional status, update the associated model online through incremental learning, continuously iterate and optimize the vibration prescription, and dynamically fine-tune training parameters according to the user's real-time ocular status, always ensuring that the training program is in a worry-free and adaptable state.

[0052] 4. This invention, based on a quantitative correlation model between vibration and myopia functions, can accumulate long-term time-series training data and ocular physiological change data, accurately predicting the evolution trend of myopia under different vibration prescriptions and anticipating risks such as excessive axial elongation and ciliary muscle dysfunction. It can provide users and managers with long-term, scientific training decision-making support, solving the problems of traditional equipment that only allows for single-session training, lacks long-term predictive capabilities, and has delayed intervention.

[0053] 5. This invention can visualize the myopia improvement effect, eye function optimization trend, and long-term training benefits derived from the model, allowing users to intuitively perceive the training value and expected effects. It solves the pain points of traditional equipment training effects that cannot be quantified, have weak user perception, and have poor long-term training compliance, effectively improving the compliance and continuity of long-term regular training and ensuring the long-term intervention effect of myopia prevention and control.

[0054] 6. Based on the correlation model between vibration function and myopia function, this invention can quickly complete parameter matching and optimal prescription selection without repeated manual adjustments. It can accurately lock the optimal combination of vibration parameters suitable for an individual, greatly reduce the time cost and training loss of manual trial and error, and significantly improve the overall efficiency and stability of personalized vision training. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the overall architecture of the vision training system based on the correlation model between vibration function and myopia function provided by the present invention.

[0056] Figure 2 This is a schematic diagram of the correlation model between the vibration function and the myopia function provided by the present invention.

[0057] Figure 3 This is a schematic diagram of the vibration function provided by the present invention.

[0058] Figure 4 This is a schematic diagram of the myopia function provided by the present invention.

[0059] Figure 5 A schematic diagram of the individual characteristic parameters provided by the present invention.

[0060] Figure 6 This is a flowchart illustrating the vision training method based on the correlation model between vibration function and myopia function provided by the present invention. Detailed Implementation

[0061] The following section provides a complete, detailed, and readily applicable description of the specific implementation methods of this invention, incorporating its complete system architecture, algorithm principles, hardware configuration, and training process. Those skilled in the art can replicate the entire process of hardware setup, data acquisition, function modeling, correlation model computation, personalized prescription generation, adaptive dynamic control, and closed-loop vision training without any creative effort.

[0062] This invention addresses four core shortcomings of existing vibration-based vision training equipment: fixed parameters, lack of a quantitative correlation system, blind training, inability to adaptively iterate, and lack of long-term predictive capabilities. It employs a comprehensive technical solution encompassing multidimensional mathematical modeling, data-driven training, intelligent inverse optimization, and closed-loop adaptive control to address each of these shortcomings individually. The following provides a detailed explanation of the core technical details, design logic, and complete implementation methods of this invention.

[0063] See Figure 1 The vision training system based on the correlation model between vibration function and myopia function consists of three core components: a whole-body sound wave vertical rhythm device, a myopia parameter measurement module, and a central processing module. The three modules work together in a closed loop to solve the technical defects of traditional vibration training equipment, such as fixed parameters, lack of quantitative modeling, blind training, and lack of personalized adaptation.

[0064] Among them, the whole-body acoustic vertical rhythm device is the only physical training execution terminal of the system. The hardware adopts a vertical servo drive structure, outputting only standard sinusoidal mechanical vibration in the vertical direction, completely eliminating ineffective stimulation and eye discomfort caused by horizontal noise and oblique vibration. The device supports real-time dynamic adjustment of vibration parameters, with the effective working parameter range strictly limited to a vibration frequency of 2-50Hz and a vibration amplitude of 2-5mm. The parameters are continuously adjustable and steplessly fine-tuned throughout the entire range. The output conditions can be changed in real time according to the user's eye condition, providing precise, gentle, and controllable mechanical vibration stimulation to the human eye muscles, choroidal tissue, and the whole-body visual-related nervous system, thereby improving eye microcirculation, regulating ciliary muscle activity, and optimizing the physiological state of the choroid.

[0065] The myopia parameter measurement module is the core unit for system data acquisition. It is equipped with a complete set of ophthalmic testing equipment and collects four core ocular physiological parameters of the user in a standardized manner, including axial length, refractive power, choroidal thickness, and accommodative amplitude. These four parameters correspond to myopia structural damage, degree of refractive abnormality, ocular blood supply and metabolism level, and ciliary muscle accommodation function, respectively. This comprehensively quantifies the user's true physiological state of myopia and provides basic data for subsequent modeling and prescription generation.

[0066] The central processing module serves as the system's intelligent computing and control hub. Equipped with an embedded processing chip, it possesses end-to-end functions including high-frequency data acquisition, function modeling, model computation, prescription solving, device control, and data storage iteration. During operation, the central processing module collects the dynamic vibration parameters of the whole-body acoustic vertical rhythm device in real time, constructing a unique vibration function tailored to the entire training session. Simultaneously, it receives real-time eye physiological data from the user collected by the myopia parameter measurement module, constructing a personalized myopia function for each user. Through algorithm training, it builds a quantitative correlation model between the two types of functions. Finally, based on this correlation model, it intelligently outputs a personalized vibration training prescription adapted to the individual eye condition of each user, achieving a technological upgrade from "fixed-mode uniform training" to "precise intervention tailored to each individual."

[0067] See Figure 3 We construct a four-dimensional integrated vibration function system to quantify, finely control, and design differentiated modes for the core parameters of vibration training, thereby achieving standardized, personalized, and dynamic output of vibration stimulation.

[0068] The overall definition of the vibration function is:

[0069]

[0070] In the formula, For frequency function, For amplitude function, For intensity function, The duration parameter is used; among them, the vibration frequency is the core parameter determining the efficiency of vibration energy transmission to the eye. Extensive human experiments have verified that the optimal resonant frequency range for energy transmission from the human skeleton to the eye is 17-21 Hz, within which the transmission efficiency of vibration energy to the eye is highest. Based on this physiological characteristic, the range of values ​​for the frequency function is limited to... It adopts an ultra-high resolution continuously adjustable design of ≤0.5Hz, which can accurately locate the optimal resonance frequency for an individual, while adapting to the gradient stimulation needs of different myopia states.

[0071] This embodiment sets up four differentiated frequency working modes, each specifically adapted to different training scenarios, covering the intervention needs of the entire population: the first is the constant frequency mode, i.e. In the formula The first mode is a preset fixed vibration frequency value, suitable for daily steady-state maintenance training after the user's optimal resonance frequency has been determined. It outputs stable and gentle vibration stimulation to continuously improve eye microcirculation; the second mode is a linear scanning mode, i.e.

[0072]

[0073] In the formula, The scan start frequency, For frequency scan slope, For training time, this mode can achieve linear dynamic frequency change, used for accurate search of individual resonant frequencies during initial training and periodic parameter recalibration, solving the problem of optimal frequency deviation caused by individual physiological differences; the third is the periodic modulation mode, i.e.

[0074]

[0075] In the formula, To preset a fixed vibration frequency value, For frequency modulation amplitude, For modulating angular frequency, For training time, this mode effectively avoids habitual and adaptive fatigue of the ciliary muscle caused by single-frequency stimulation by periodically and dynamically fine-tuning the frequency, and continuously ensures the training sensitivity of the eye muscles; fourth, it has a preset program mode that supports custom time-frequency change curves, and can match the optimal frequency sequence combination according to different training goals such as rapid relaxation, deep repair, and long-term prevention and control, so as to achieve scenario-based precise intervention.

[0076] Vibration amplitude directly determines the total vibration energy in a single training session, and is a key parameter affecting the intensity of ocular stimulation and the depth of intervention. The amplitude function is limited to a specific range. It is continuously and steplessly adjustable throughout the entire process, taking into account both the safety and effectiveness of eye training, and avoiding the problems of insufficient intervention due to excessively low amplitude and eye fatigue and damage due to excessively high amplitude.

[0077] To meet users' needs for visual and convenient operation, vibration intensity levels with a linear and uniform distribution from 0 to 120 are set, establishing a precise quantitative mapping relationship between amplitude and intensity. The corresponding formula is as follows:

[0078]

[0079] In the formula, This represents the maximum amplitude of the equipment. This is the maximum intensity setting for the equipment. Adjusting the intensity setting essentially involves regulating the amount of driving current, thereby precisely controlling the output amplitude of the vibration platform and achieving a two-way adaptation between convenient user-defined intensity settings and precise system parameter control.

[0080] The duration of a single training session is a key dimension for balancing training effectiveness and eye tolerance; therefore, the training duration range is limited to [specific range]. The duration can be flexibly set according to the user's age, degree of myopia, and eye tolerance. Simultaneously, the system's associated model can intelligently recommend the optimal single training session length by combining the user's historical training data and dynamic changes in the myopia function, avoiding insufficient intervention in short-term training and overstimulation in long-term training, thus ensuring the safety and efficiency of the training.

[0081] See Figure 4A four-dimensional myopia function system was constructed, selecting four core physiological indicators: axial length, refractive power, choroidal thickness, and accommodative amplitude. These indicators correspond to structural damage to myopia, degree of refractive error, ocular blood supply status, and ciliary muscle accommodation function, respectively. This system comprehensively and sequentially quantifies the user's myopia status, providing core data support for accurate modeling and personalized prescription generation.

[0082] The overall definition of the myopia function is:

[0083]

[0084] In the formula, For axial length, For refractive power, For choroid thickness, To adjust the amplitude.

[0085] Axial length is a core indicator for assessing myopia progression, differentiating between true and false myopia, and evaluating structural myopia damage. Non-contact measurement using an optical biometer achieves an accuracy of ±0.01mm. The measurement frequency is adapted to the rate of myopia progression: monthly during the initial intervention phase, and every 3 months after the eye condition stabilizes. All data is automatically entered into the system to generate a time-series sequence of axial length changes. It accurately captures the abnormal growth trend of the axial length of the eye.

[0086] Refractive error directly reflects the overall degree of refractive abnormality of the eye and is a fundamental indicator for myopia classification and intervention target setting. This embodiment uses an automated computer-controlled refractometer or a comprehensive refractometer to complete the test under cycloplegic conditions, avoiding detection errors caused by accommodative spasm and obtaining accurate equivalent spherical refractive error data. The detection frequency is synchronized with the axial length of the eye, and the data is entered into the system in real time to form a time-series sequence of refractive error changes. The improvement and progress of quantifying refractive status.

[0087] Choroidal thickness is a highly sensitive biomarker for the short-term effectiveness of myopia intervention, showing a significant response to mechanical vibration stimuli and rapidly reflecting the effectiveness of training intervention. This embodiment uses frequency-domain optical coherence tomography (SD-OCT) to accurately detect the thickness of the subfoveal choroid, achieving an axial detection resolution of 1 μm. This detection is non-invasive, supports high-frequency monitoring, and in this embodiment, it is set to be detectable before and after a single training session, providing real-time feedback on the effect of each training session and generating a time-series sequence of choroidal thickness changes. This provides high-frequency data support for dynamic optimization of the model.

[0088] Accommodative amplitude directly reflects the ciliary muscle's ability to contract and relax, and is a core indicator for assessing visual fatigue, accommodative dysfunction, and pseudomyopia. This embodiment uniformly employs the ophthalmological standard near-point method or negative lens method to detect the accommodative near point, using a standardized testing environment (500 lx standard illumination, fixed testing distance and body position) to avoid environmental and human operational errors. Standardized testing is performed once a week to generate a temporal sequence of accommodative amplitude changes. To accurately evaluate the repair effect on the eye's accommodation function.

[0089] See Figure 5 Individual characteristic parameters specifically include user age, gender, height, weight, skeletal structure, degree of myopia, and eye tolerance, improving the accuracy of personalized adaptation.

[0090] See Figure 2 Constructing a data-driven, multidimensional quantitative correlation model and establishing a precise mapping relationship between vibration stimulation, individual characteristic parameters, and changes in myopia status is the core mathematical foundation for realizing personalized prescription generation, adaptive regulation, and long-term trend prediction.

[0091] The core formula of the correlation model F is:

[0092]

[0093] In the formula, This represents the change in eye condition. This is a model relating the vibration function and the myopia function. It is a set of multidimensional vectors of vibration functions. For the user's eye physiological parameters. For the user's individual characteristic parameters, This is a term for correcting model fitting errors. To adapt to application scenarios with different data volumes and varying complexity of associations, this embodiment is compatible with multiple machine learning algorithms and can be flexibly switched for adaptation: Linear regression models are suitable for simple scenarios with mild myopia and approximately linear parameter associations, offering high computational efficiency and strong stability; SVR (Short Vector Regression) is suitable for small sample sizes and nonlinear association scenarios, adapting to modeling specific myopia populations; Random Forest Regression is suitable for scenarios with high-dimensional features and significant multi-parameter interaction effects, accurately capturing the coupling effects of multiple parameters; Neural Network models are suitable for large-scale datasets and highly nonlinear association scenarios, maximizing model prediction accuracy. This multi-algorithm compatible architecture balances model computational efficiency and fitting accuracy, adapting to training needs across all scenarios.

[0094] This embodiment defines the core algorithm for optimizing vibration prescriptions, achieving "goal-oriented" automatic generation of intelligent prescriptions. It overcomes the shortcomings of traditional training schemes, such as homogeneity, lack of specificity, and inability to match personalized improvement goals. The generation of the optimal personalized prescription is a solution to the inverse problem, calculated using the following formula:

[0095]

[0096] In the formula, For the optimal combination of vibration prescription parameters, V is the set of multidimensional vectors of vibration functions. This is a model relating the vibration function and the myopia function. For the user's eye physiological parameters. For the user's individual characteristic parameters, To pre-determine the amount of change in the target for myopia improvement, The regularization coefficient is . For regularization terms, To constrain the range of vibration parameter values, ensuring their safety and rationality, the system employs a gradient descent iterative algorithm to converge the optimal solution during the iterative solution process. After initializing the candidate set of vibration parameters, the system continuously iterates backwards to update the parameter combinations with the core objective of minimizing the error. Each iteration verifies the validity of the parameters through a regularization term, eliminating invalid parameter solutions that exceed limits or exhibit abrupt changes, until the error converges to a preset accuracy threshold. Finally, the globally optimal vibration prescription is output. This inverse optimization generation method requires no human intervention and relies entirely on the user's measured eye data and quantified improvement goals to intelligently generate prescriptions. It adapts to individual differences among users of different ages, degrees of myopia, and eye habits, truly achieving personalized intelligent and precise vision training intervention.

[0097] The adaptive control module employs a closed-loop working logic of "real-time monitoring - deviation calculation - threshold determination - dynamic adjustment". The system collects the user's latest ocular physiological data in real-time or periodically at a preset frequency to construct a monitoring myopia function; simultaneously, it outputs a predicted myopia function for the same period through a correlation model. The module compares the differences between the measured data and the predicted data parameter by parameter, accurately calculating the dynamic deviation. The calculation formula is as follows:

[0098]

[0099] In the formula, The deviation between measured data and predicted data, To monitor the myopia function, To predict the myopia function, i.e. the predicted value of myopia status, when the deviation exceeds a preset threshold, the online update of the correlation model and the dynamic adjustment of the vibration prescription are triggered.

[0100] The system has a built-in unified preset deviation threshold. When the calculated dynamic deviation δ(t) does not exceed the threshold, it is determined that the user's eye physiological state has no significant fluctuations, and the system maintains the original model weights and training prescription unchanged. When the deviation δ(t) exceeds the preset threshold, it proves that the user's eye state has changed significantly, and the original model parameters and training scheme can no longer adapt to the current physiological state. The module immediately triggers two core actions: first, it starts the online update and iteration of the associated model to adapt to the latest eye data features; second, it re-optimizes and solves the new vibration prescription to achieve dynamic adaptive adjustment of the training scheme and ensure the accuracy and effectiveness of long-term training.

[0101] The incremental learning algorithm formula and core implementation logic for online updating of the association model enable lightweight, efficient, and overfit-free continuous iterative optimization, providing the core algorithmic support for the system's adaptive capability. The core formula for online updating of the association model is:

[0102]

[0103] in, For the updated association model, The original association model. The learning rate for incremental learning. The gradient operator of the loss function with respect to the correlation model. Let V be the loss function, and let V be the set of multidimensional vectors of vibration functions. For the user's eye physiological parameters. This represents the actual change in myopia status obtained from a single training session. This incremental learning method does not require loading massive full historical datasets; it only relies on the latest training data to complete local weight iterations. It has low computational cost, fast response speed, and enables real-time online model updates. Simultaneously, the small-gradient iteration update mechanism effectively avoids overfitting caused by anomalous data in a single session, allowing the model to continuously adapt to the user's personalized eye condition changes and steadily improve long-term prediction accuracy and prescription fit.

[0104] The long-term trend prediction module breaks through the limitations of traditional myopia intervention, which can only passively correct vision, and achieves proactive and predictive myopia prevention and control. Based on a correlation model, the long-term trend prediction module predicts the long-term changing trend of the myopia function. The core time-series integral formula for long-term trend prediction is:

[0105]

[0106] In the formula, Let be the myopia function predicted at time t. At the initial moment, Let be the initial myopia function at the initial time. For integration time series variables, the integral term represents The cumulative changes in eye condition caused by vibration throughout the entire time period. for The vibration function vector at any moment, for The vector of the myopia function at any given time. For the user's individual characteristic parameters, This is a model linking the vibration function and the myopia function. The formula, through time-series integration, quantifies and integrates the continuous improvement effect of vibration training on ocular physiological parameters over all periods. It can accurately predict the four-dimensional myopia parameter change trends for users over the next 7, 30, and 90 days, outputting quantitative predicted values ​​and dynamic change curves. Users can rely on the prediction results to anticipate the risk of myopia progression, adjust long-term training frequency and parameters, and develop scientific long-term intervention plans, achieving an upgrade in myopia intervention from passive treatment to proactive prevention and precise prediction.

[0107] The whole-body sound wave vertical rhythm device is a multi-person compatible device that can accommodate 2-6 users to conduct vision training simultaneously. It balances device efficiency with personalized training accuracy and is suitable for large-scale use in schools, rehabilitation institutions, and homes.

[0108] The equipment features 2-6 independent training stations, each equipped with its own dedicated drive circuit, vibration signal output channel, and data acquisition port. This ensures complete physical isolation between multiple users' vibration output, data acquisition, and signal transmission, preventing interference between stations. The central processing module boasts multi-threaded parallel computing capabilities, enabling it to independently complete the entire data processing workflow for each user simultaneously: individually collecting personal myopia parameters, independently constructing a dedicated myopia function, independently performing correlation model calculations, independently generating vibration prescriptions through inverse optimization, and independently implementing model iteration and adaptive adjustment. Multiple users can train simultaneously, with each person's vibration parameters, training plan, model weights, and intervention effect data remaining completely independent. This truly achieves "multi-user parallel processing, individualized prescriptions, and independent precise intervention," significantly improving equipment utilization while strictly guaranteeing personalized training effects for each user. This addresses the shortcomings of traditional single-user training equipment, such as low efficiency and inability to perform batch interventions.

[0109] See Figure 6 Based on the above training system, the closed-loop adaptive vision training method includes a complete standardized process from SS1 to SS7, forming a full-link intelligent intervention system of "data collection - target setting - prescription generation - precision training - effect quantification - model iteration - closed-loop cycle". The specific step-by-step implementation is as follows.

[0110] SS1. Initial Data Acquisition: Before the user's first training session, four core eye parameters are standardized and tested using the myopia parameter measurement module to construct an initial myopia function. The formula is as follows:

[0111] In the formula, This is the initial myopia function. For the initial axial length, For initial refractive power, For the initial choroid thickness, The initial adjustment range is set; individual feature vectors X, such as age, gender, height, weight, skeletal structure, degree of myopia, and eye tolerance, are simultaneously entered to establish a personal electronic file and basic database, providing basic data support for subsequent modeling and prescription solving.

[0112] SS2. Goal Setting: Users or medical staff input feasible, phased myopia improvement goals based on the user's myopia level, age, eye tolerance, and intervention needs. The system automatically quantifies the ambiguous goals into standardized target myopia function changes. This clarifies the precise iterative direction for prescription optimization.

[0113] SS3, Prescription Generation: The system calls the pre-trained baseline correlation model F, and under multiple safety parameter constraints, obtains a personalized optimal vibration prescription that adapts to the user's current eye condition and improvement goals through an inverse optimization iterative algorithm. .

[0114] SS4 Precision Training: The central processing module drives the whole-body sound wave vertical rhythm device, outputting vertical sinusoidal vibration stimulation in strict accordance with the timing parameters of the optimal vibration prescription. Throughout the process, the closed-loop feedback circuit of the equipment corrects parameter errors in real time, ensuring that the training stimulation is precise, stable, and controllable, and completing a single standardized vision training session.

[0115] SS5. Effect Detection: After a single training session, the user rests for 10 minutes to allow the eye's physiological state to stabilize before the system re-collects the real-time myopia function. By comparing the differences in parameters before and after training, the measured change in myopia during this training can be accurately calculated. To quantify the actual intervention effect of a single training session.

[0116] SS6, Model Iteration: The system incorporates all vibration prescription parameters, initial eye data, measured improvement data, and individual characteristic data from this training into the training sample set. It then uses an incremental learning algorithm to iteratively update the associated model F, continuously optimizing the model's fitting accuracy and personalized adaptation capabilities.

[0117] SS7. Closed-loop iteration: Execute all the above steps in a loop, continuously and dynamically monitor changes in the user's eye condition, iteratively optimize the vibration prescription and correlation model, realize long-term, dynamic, and adaptive closed-loop myopia intervention and vision training, and continuously optimize the prevention and control effect.

[0118] The pre-training methods and dataset construction standards for the correlation model ensure that the model has excellent fitting accuracy and generalization ability. The model pre-training stage first involves constructing a standardized multi-dimensional dataset:

[0119] In the formula, For training the model dataset, This represents the total number of training samples. For the first The vibration function vector corresponding to the group of samples, For the first The myopia vibration function vector corresponding to the group of samples, For the first The set of individual user characteristic parameters corresponding to the group samples. For the first The dataset contains measured myopia change deviation values ​​corresponding to the sample groups. The dataset covers all age groups (6-18 years old) and includes data on pseudomyopia, low-grade myopia, moderate myopia, and high myopia. It contains a large amount of data on vibration parameters, eye baseline parameters, user individual characteristics, and training effect changes from real training scenarios, ensuring sample diversity and comprehensiveness. After the dataset was constructed, it was divided into training and testing sets in a 7:3 ratio. The two sets of data are completely independent and have no overlap, preventing model overfitting. The model training phase focuses on minimizing the overall prediction error, calculated using a loss function. The calculation formula is as follows:

[0120]

[0121] In the formula, The model F is the minimum value. For training sample numbers, This represents the total number of training samples. For the first The vibration function vector corresponding to the group of samples, For the first The myopia vibration function vector corresponding to the group of samples, For the first The set of individual user characteristic parameters corresponding to the group samples. For the first The measured myopia change deviation values ​​corresponding to the sample groups. Multiple rounds of iterative training were completed based on the calculation formula to continuously optimize the model weight parameters. After training, the model's generalization ability was verified using a test set. Models with satisfactory prediction accuracy and excellent stability were selected as the initial benchmark models for formal online use, ensuring the accuracy of subsequent prescription solving and state prediction.

[0122] The prescription inverse optimization solution process has multiple safety parameter constraints to prevent out-of-limit parameter outputs from the algorithm level, ensuring the safety and compliance of all training processes and adapting to human eye tolerance standards.

[0123] Throughout the entire iterative solution process for the vibration prescription, the system enforces strict parameter boundary constraints. All candidate parameters generated in the iterations must meet the following limits: vibration frequency 2Hz≤f≤50Hz, vibration amplitude 2mm≤A≤5mm, vibration intensity 0≤I≤120 levels, and single training duration 5min≤T≤30min. During the algorithm iteration process, if an invalid parameter solution exceeding the above limits is generated, the system will automatically truncate, backtrack, and iterate again, retaining only the optimal parameter combination within the valid limits as the final prescription. This multi-constraint mechanism perfectly matches the physiological tolerance limits of the human eye, avoiding the risks of eye fatigue and accommodation disorders caused by high-frequency, high-amplitude, and excessively long training sessions, while also avoiding the problem of ineffective intervention from low-frequency, low-amplitude training, thus achieving a dual guarantee of safety and effectiveness.

[0124] The complete output system of vibration prescriptions ensures that the prescription content is standardized, practical, and directly executable, providing users with comprehensive and standardized training and guidance programs.

[0125] The vibration prescription ultimately generated by the system includes four core, actionable components: First, a frequency variation curve, which displays the temporal dynamic changes in vibration frequency throughout a single training session, adapting to the parameter change logic of the selected working mode; second, an intensity variation curve, which synchronously matches the dynamic intensity level changes of frequency and amplitude parameters, ensuring smooth adaptation of training stimuli; third, the total duration of a single training session, accurately determined based on the user's eye condition and improvement goals, strictly adhering to safe duration constraints; and fourth, a daily training frequency recommendation, which intelligently recommends the number of daily training sessions based on the user's myopia severity, eye tolerance, and age characteristics, forming a standardized long-term training plan.

[0126] The complete prescription content has clear parameters, a well-defined time sequence, and uniform standards. It can directly drive the device to complete precise training without the need for secondary adjustments by the user. At the same time, it can serve as a standardized basis for formulating long-term myopia intervention programs, realizing the standardization, precision, and intelligence of vision training.

[0127] The above-described embodiments are merely one implementation of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A vision training system based on a correlation model between vibration function and myopia function, characterized in that, It includes a whole-body sound wave vertical rhythm device, a myopia parameter measurement module, and a central processing module; The whole-body acoustic vertical rhythm device generates vertical sinusoidal vibrations with continuously adjustable frequency and amplitude. The myopia parameter measurement module measures the user's axial length, refractive power, choroidal thickness, and accommodative amplitude. The central processing module establishes a vibration function by collecting data generated by the whole-body sound wave vertical rhythm device, establishes a myopia function by collecting user data measured by the myopia parameter measurement module, establishes a correlation model between the vibration function and the myopia function, and generates a vibration prescription based on the correlation model.

2. The vision training system based on the correlation model between vibration function and myopia function according to claim 1, characterized in that, The vibration function satisfies: In the formula, Let be the vibration function. It is a frequency function. Let be the amplitude function. For intensity function, This is the duration parameter; where The adjustment range is 2-50Hz. The adjustment range is 2-5mm. The adjustment range for t is 0-120 levels, and the adjustment range for T is 5-30 minutes; The myopia function satisfies: In the formula, For the myopia function, axial length, For refractive power, The thickness of the choroid. To adjust the amplitude. The association model satisfies: In the formula, This represents the change in eye condition. For the aforementioned association model, This is a multidimensional vector set of the vibration function. For the user's eye physiological parameters. For the user's individual characteristic parameters, This is a correction term for the model fitting error; the correlation model is implemented by fusing one or more methods from linear regression, support vector regression, random forest regression, or neural networks. The method for generating the vibration prescription is as follows: based on the desired myopia improvement target. Solve the inverse problem to obtain the optimal combination of vibration prescription parameters, and select the vibration prescription from it; the formula is: In the formula, For the optimal combination of vibration prescription parameters, This is a multidimensional vector set of the vibration function. For the aforementioned association model, For the user's eye physiological parameters. For the user's individual characteristic parameters, To pre-determine the amount of change in the target for myopia improvement, The regularization coefficient is . For regularization terms, To constrain the range of vibration parameter values, and to ensure the safety and rationality of vibration parameters.

3. The vision training system based on the correlation model between vibration function and myopia function according to claim 2, characterized in that, It also includes an adaptive adjustment module, which monitors the user's myopia function in real time or periodically, and calculates the deviation using the following formula: In the formula, The deviation between measured data and predicted data, To monitor the myopia function, To predict the myopia function, i.e. the predicted value of myopia state, when the deviation exceeds a preset threshold, the online update of the correlation model and the dynamic adjustment of the vibration prescription are triggered.

4. The vision training system based on the correlation model between vibration function and myopia function according to claim 3, characterized in that, The online update of the association model adopts an incremental learning method, and the calculation formula of this method is as follows: In the formula, For the updated association model, The original association model. The learning rate for incremental learning. The gradient operator of the loss function with respect to the correlation model. Let V be the loss function, and let V be the set of multidimensional vectors of vibration functions. For the user's eye physiological parameters. For the user's individual characteristic parameters, This represents the actual change in myopia status measured after a single training session.

5. The vision training system based on the correlation model between vibration function and myopia function according to claim 2, characterized in that, It also includes a long-term trend prediction module, which predicts the long-term trend of the myopia function based on the correlation model, using the following formula: In the formula, Let be the myopia function predicted at time t. At the initial moment, Let be the initial myopia function at the initial time. For integration time series variables, the integral term represents The cumulative changes in eye condition caused by vibration throughout the entire time period. For time sequence The corresponding vibration function vector, For time sequence The corresponding myopia function vector, For the user's individual characteristic parameters, This refers to the association model.

6. The vision training system based on the correlation model between vibration function and myopia function according to claim 2, characterized in that, The frequency function supports one or more of the following modes: constant frequency mode, linear scanning mode, periodic modulation mode, and preset program mode.

7. The vision training system based on the correlation model between vibration function and myopia function according to claim 1, characterized in that, The whole-body acoustic vertical rhythm device is a multi-user device that can accommodate 2-6 people to use at the same time. The central processing module can independently establish the association model for each user and generate a personalized vibration prescription.

8. A vision training method using the vision training system based on the correlation model between vibration function and myopia function as described in any one of claims 2-6, characterized in that, Includes the following steps: SS1. Initial Data Acquisition: Measure the user's initial myopia function, satisfying the following formula: In the formula, This is the initial myopia function. The initial axial length, Initial refractive power, The initial choroid thickness, To determine the initial adjustment range, the user's individual feature vector X is collected. SS2. Goal Setting: Enter the user's myopia improvement goals and determine the target change in myopia function; SS3. Prescription Generation: Based on a pre-trained association model, a personalized optimal vibration prescription is generated through inverse optimization. SS4. Precision Training: Control the whole-body sound wave vertical rhythm device to output precise vibration stimulation according to the personalized optimal vibration prescription to complete a single training session; SS5. Effect Detection: After training is completed, the user's real-time myopia function is collected, and the measured change in myopia is calculated; SS6. Model Iteration: Incorporate the current training data into the dataset and update the associated model through incremental learning; SS7. Closed-loop iteration: Iterate the above steps repeatedly to achieve closed-loop adaptive vision training and dynamic myopia intervention.

9. The vision training method based on the correlation model between vibration function and myopia function according to claim 8, characterized in that, The training method for the associated model in the model iteration of step SS6 is as follows: collect N sets of training data: in, For training the model dataset, This represents the total number of training samples. For the first The vibration function vector corresponding to the group of samples, For the first The myopia vibration function vector corresponding to the group of samples, For the first The set of individual user characteristic parameters corresponding to the group samples. For the first The measured deviation value of myopia change corresponding to the group of samples; The dataset is divided into training and test sets in a 7:3 ratio. The prediction error is minimized on the training set, and the calculation formula is as follows: In the formula, The model F is the minimum value. For training sample numbers, This represents the total number of training samples. For the first The vibration function vector corresponding to the group of samples, For the first The myopia vibration function vector corresponding to the group of samples, For the first The set of individual user characteristic parameters corresponding to the group samples. For the first The measured deviation value of myopia change corresponding to the group of samples.

10. The vision training method based on the correlation model between vibration function and myopia function according to claim 8, characterized in that, In the prescription generation process of step SS3, the inverse optimization solution process is subject to multiple constraints, including: , , files, .

11. The vision training method according to claim 10, characterized in that, The choroid thickness C(t) was measured using frequency domain optical coherence tomography, the axial length L(t) was measured using an optical biometer, and the accommodation amplitude R(t) was measured using the proximity method or the negative lens method.

12. The vision training method based on the correlation model between vibration function and myopia function according to claim 8, characterized in that, The vibration prescription includes frequency variation curves, intensity variation curves, total duration of a single training session, and daily training frequency recommendations.

Images