A visual training adaptive management method and system

Abstract

The present application belongs to the field of digital health and human-computer interaction technology, and discloses a visual training adaptive management method and system. The method comprises the following steps: outputting audio to guide the user's visual line to switch to a physical visual target medium, verifying the real physical focusing switching through the corneal reflection light spot, the pupil center position and the micro eye movement characteristics; after the verification, recording and disassembling the total response time, constructing the time delay feature vector, inputting the adaptive regulation model to identify the abnormal dimension, dynamically adjusting the optical parameters, the visual target distance or the complexity; collecting the environmental illumination and the physical distance, compensating and guiding the voice according to the equivalent diopter model to reduce the training difficulty; calculating the individualized safety load threshold according to the historical data, suspending the training and supporting remote unlocking if the threshold is exceeded; encrypting and uploading the related data to the cloud to realize the terminal parameter remote overwrite, the configuration bidirectional synchronization and the version management. The present application effectively improves the data reliability, the difficulty adaptation accuracy and the physiological safety of the visual training.

Description

Technical Field

[0001] This invention relates to the field of digital health and human-computer interaction technology, and in particular to a visual training adaptive management method and system. Background Technology

[0002] Visual function training is an effective means of visual function rehabilitation, treatment of amblyopia, and relief of eye strain. Among them, reverse-viewing training is a commonly used visual rehabilitation training method to exercise the eye's accommodative and convergent functions, improve eye strain, and assist in the treatment of pseudomyopia and early presbyopia. The traditional training method is as follows: the patient holds a reverse-viewing device (a set of lenses with alternating positive and negative lenses), and after clearly seeing the near optotype card, flips the lenses and sees the optotype again, recording the number of cycles completed per minute.

[0003] Traditional reverse-grip training has the following technical shortcomings: 1. Poor adherence and difficulty in long-term commitment: The training process is tedious and repetitive, making it difficult for patients to persist. Although some auxiliary training apps have appeared on the market, most are simple timers and counters, lacking long-term incentive mechanisms. Patients are prone to giving up due to boredom, leading to training interruptions.

[0004] 2. Inaccurate data collection and lack of key indicator evaluation: Traditional training relies on manual observation and recording, which cannot accurately collect reaction time, a key physiological indicator. Reaction time reflects neural conduction speed and regulatory sensitivity, and is an important basis for evaluating training effectiveness and adjusting training difficulty. However, current technology lacks automated means of collecting this indicator.

[0005] 3. Lack of personalized difficulty adjustment and insufficient adaptability: Patients have individual differences in their visual acuity and recovery progress. Fixed difficulty settings (such as using ±2.00D lenses) cannot adapt to the actual abilities of different patients. Training that is too difficult leads to frustration, while training that is too easy fails to achieve the desired training effect. Current technology lacks a mechanism to dynamically adjust the training difficulty based on the patient's real-time performance.

[0006] 4. Lack of professional guidance during training, resulting in low efficiency: Reverse racket training requires patients to master correct training techniques to improve training efficiency and shorten the training cycle. However, for beginners or younger patients, it is difficult to master the relevant techniques in a short period of time. Current technology does not have targeted process guidance, and cannot provide real-time skill guidance based on patients' training feedback, leading to some patients using inappropriate training methods and affecting training results.

[0007] 5. Lack of accurate feedback and insufficient motivation: During training, the conscious effort to clearly see the visual targets when patients encounter bottlenecks is crucial for improving training effectiveness. Immediate accuracy assessment and feedback after answer selection can effectively stimulate patients' desire for challenge and enhance their training initiative. However, current technology lacks a robust mechanism for performance evaluation and immediate feedback, failing to provide timely positive encouragement and thus hindering the maintenance of patients' enthusiasm for training.

[0008] Existing digital vision training applications have the following shortcomings: 1. Excessive screen dependence: Some software requires patients to stare at the screen for a long time, which can increase fatigue and contradict the original purpose of training.

[0009] 2. Lack of safety boundaries: Visual training has a marginal effect; exceeding a certain dosage is not only ineffective but may even lead to adverse effects such as accommodative spasm.

[0010] 3. Fixed difficulty: The training difficulty cannot be dynamically adjusted according to the patient's real-time response, resulting in low training efficiency.

[0011] 4. Data silos between doctors and patients: Doctors cannot obtain real data on patients' home training process and can only rely on patients' subjective feedback, which leads to delays in adjusting the plan during follow-up visits and a lack of remote intervention methods.

[0012] Therefore, how to provide a visual training adaptive management method and system is an urgent problem to be solved. Summary of the Invention

[0013] This invention provides a visual training adaptive management method and system to solve the problems mentioned above in the prior art.

[0014] According to a first aspect of the present invention, a visual training adaptive management method is provided.

[0015] In one embodiment, the visual training adaptive management method includes: Output audio guidance commands to guide the user's gaze away from the display screen and focus on the physical target medium, acquire the user's eye image, and verify whether the user has completed the real physical space focus switch based on the relative position change of the corneal reflective spot and the pupil center combined with micro-eye movement feature analysis. After the physical focus switching verification is passed, the total response time from the end of the audio command to the user's interactive input confirmation time is recorded. The total response time is decomposed into multiple sub-time period vectors, and a delay feature vector containing multiple sub-time periods is constructed. Input the time delay feature vector into the preset difficulty adaptive control model, identify the time delay anomaly dimension, and dynamically adjust the suggested values ​​of optical parameters of the physical target medium, physical distance guidance instructions or target complexity according to the anomaly dimension; Collect ambient illuminance data and the actual distance from the user to the physical target medium, and calculate the effective pupil aperture and equivalent depth of field in the current environment based on the equivalent refractive power model. When it is detected that the training difficulty is artificially reduced due to excessively bright environment or too close distance, corresponding compensation and voice guidance are performed. Based on the reaction latency variance of users' historical training data, the system dynamically calculates the personalized safety load threshold for the day, accumulates the effective training load in real time, initiates progressive feedback adjustment when the threshold warning range is reached, and pauses the training entry and generates a recovery suggestion when the accumulated load exceeds the safety limit. It also supports remote authorization unlocking. The system encrypts and uploads latency feature vectors, environmental optical compensation data, and effective training load data to the cloud, and remotely overwrites the terminal training parameters. It also supports bidirectional synchronization and version management of training schemes, difficulty curves, and security thresholds.

[0016] In one embodiment, the visual training adaptive management system includes: The physical focus verification module is used to output audio guidance commands to guide the user's gaze away from the display screen and focus on the physical target medium, acquire the user's eye image, and verify whether the user has completed the real physical space focus switch based on the relative position change of the corneal reflective spot and the pupil center combined with micro-eye movement feature analysis. The latency feature construction module is used to record the total response time from the end of the audio command to the user's interactive input confirmation time after the physical focus switching verification is passed. The total response time is decomposed into multiple sub-time period vectors, and a latency feature vector containing multiple sub-time periods is constructed. The difficulty adaptive adjustment module is used to input the time delay feature vector into the preset difficulty adaptive adjustment model, identify the time delay anomaly dimension, and dynamically adjust the optical parameter suggestion value, physical distance guidance command or visual object complexity of the physical visual object medium according to the anomaly dimension. The environmental optics compensation module is used to collect environmental illuminance data and the actual distance between the user and the physical target medium, and calculate the effective pupil aperture and equivalent depth of field in the current environment based on the equivalent refractive power model. When it is detected that the training difficulty is artificially reduced due to excessively bright environment or too close distance, the corresponding compensation and voice guidance are executed. The physiological load safety management module is used to dynamically calculate the personalized safety load threshold for the day based on the reaction delay variance of the user's historical training data, accumulate the effective training load in real time, start progressive feedback adjustment when the threshold warning range is reached, suspend the training entry and generate recovery suggestions when the accumulated load exceeds the safety limit, and support remote authorization unlocking. The cloud synchronization and remote configuration module is used to encrypt and upload latency feature vectors, environmental optical compensation data, and effective training load data to the cloud, and to remotely overwrite the terminal training parameters. It supports bidirectional synchronization and version management of training schemes, difficulty curves, and security thresholds.

[0017] According to a third aspect of the present invention, a computer device is provided.

[0018] In some embodiments, the computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0019] According to a fourth aspect of the present invention, a computer-readable storage medium is provided.

[0020] In one embodiment, a computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the above method.

[0021] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: (1) This invention verifies the user’s physical focusing behavior through audio guidance and eye movement feature analysis to ensure the authenticity of training; it breaks down the complete interaction process into five time nodes: audio start, audio end, eye movement focusing, gaze return to screen, and interaction confirmation, and accurately calculates the delay of visual-motor adjustment and cognitive judgment to achieve targeted difficulty control.

[0022] (2) This invention eliminates optical baseline drift by real-time monitoring of ambient illuminance and target distance, performing brightness compensation distance fine-tuning and voice guidance based on an equivalent refractive power model; it introduces an instruction priority arbitration mechanism, and when local AI adaptive decision-making conflicts with remote professional intervention instructions, it executes an immediate interruption or smooth transition strategy based on the priority identifier of emergency intervention level or scheme configuration level; it calculates personalized safety load thresholds based on historical data, and guides users to train reasonably through progressive feedback adjustment; it encrypts and synchronizes structured data to the cloud, supports remote parameter configuration by professionals, thereby effectively improving the data credibility, difficulty adaptation accuracy and physiological safety of visual training, and realizing seamless collaboration between local intelligence and remote authority.

[0023] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0024] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0025] Figure 1 This is a flowchart illustrating a visual training adaptive management method according to an exemplary embodiment; Figure 2 This is a block diagram illustrating the principle of a visual training adaptive management system according to an exemplary embodiment; Figure 3 This is a schematic diagram of the structure of a computer device according to an exemplary embodiment; Figure 4 This is a system interaction timing diagram illustrated according to an exemplary embodiment. Detailed Implementation

[0026] The following description and accompanying drawings fully illustrate specific embodiments described herein to enable those skilled in the art to practice them. Some portions and features of certain embodiments may be included in or replace portions and features of other embodiments. The scope of the embodiments herein includes the entire scope of the claims and all available equivalents thereof. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0027] The modules in the apparatus or system of this application can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0028] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0029] Figure 1 An embodiment of a visual training adaptive management method of the present invention is shown.

[0030] In this optional embodiment, the visual training adaptive management method includes: S101. Output audio guidance commands to guide the user's gaze away from the display screen and focus on the physical target medium, acquire the user's eye image, and verify whether the user has completed the real physical space focus switch based on the relative position change of the corneal reflective spot and the center of the pupil combined with micro-eye movement feature analysis. S102. After the physical focus switching verification is passed, record the total response time from the end of the audio command to the user's interactive input confirmation time, decompose the total response time into multiple sub-time period vectors, and construct a time delay feature vector containing multiple sub-time periods. S103. Input the time delay feature vector into the preset difficulty adaptive control model, identify the time delay anomaly dimension, and dynamically adjust the optical parameter suggestion value, physical distance guidance command or target complexity of the physical target medium according to the anomaly dimension. S104. Collect ambient illuminance data and the actual distance from the user to the physical target medium, and calculate the effective pupil aperture and equivalent depth of field in the current environment based on the equivalent refractive power model. When it is detected that the training difficulty is artificially reduced due to excessively bright environment or too close distance, corresponding compensation and voice guidance are performed. S105. Based on the reaction latency variance of user's historical training data, dynamically calculate the personalized safety load threshold for the day, accumulate the effective training load in real time, start progressive feedback adjustment when the threshold warning range is reached, pause the training entry and generate recovery suggestions when the accumulated load exceeds the safety limit, and support remote authorization unlocking. S106. Encrypt and upload the time delay feature vector, environmental optical compensation related data, and effective training load related data to the cloud, and remotely overwrite the terminal training parameters. Support bidirectional synchronization and version management of training schemes, difficulty curves, and security thresholds.

[0031] In this optional embodiment, when outputting audio guidance commands to guide the user's gaze away from the display screen and focus on the physical target medium, acquiring the user's eye image, and verifying whether the user has completed a real physical spatial focus switch based on the relative position change of the corneal reflective spot and the pupil center combined with micro-eye movement feature analysis, the client can be started and personalized training configuration can be loaded to initialize image acquisition, light sensing, and audio-related components; outputting audio commands containing background noise to guide the user's gaze away from the display screen and focus on the paper target card; acquiring the user's eye image at a frame rate no less than a preset frame rate, and identifying micro-tremor and micro-eye movement features based on the relative displacement sequence of the corneal reflective spot and the pupil center combined with frequency domain analysis to confirm the real physical focusing behavior; preset the coordinate range of the physical target area and the coordinate range of the display screen area, continuously monitoring the position of the corneal reflective spot, and determining that the user's gaze has moved back to the display screen when the spot position moves from the coordinate range of the physical target area to the coordinate range of the display screen area.

[0032] In this optional embodiment, after the physical focus switching verification is passed, the total response time from the end of the audio command to the user's interactive input confirmation time is recorded. The total response time is decomposed into multiple sub-time period vectors, and a time delay feature vector containing multiple sub-time periods is constructed. After the user completes the visual target recognition, a blind touch feedback operation is performed on the low-brightness interface. Key time nodes are recorded, including: the start time of the audio command, the end time of the audio command, the completion time of eye-tracking focus features, the completion time of cognitive decision-making, and the interaction confirmation time. Based on the recorded key time nodes, the auditory encoding delay, visual-motor accommodation delay, cognitive judgment delay, and interaction execution delay are calculated respectively. Based on the end time of the audio command and the interaction confirmation time, the total reaction delay is calculated. The auditory encoding delay, visual-motor accommodation delay, cognitive judgment delay, and interaction execution delay are combined to construct a time delay feature vector.

[0033] In this optional embodiment, when inputting the latency feature vector into a preset difficulty adaptive adjustment model to identify abnormal latency dimensions and dynamically adjust the suggested values ​​of optical parameters, physical distance guidance instructions, or target complexity of the physical target medium based on the abnormal dimensions, the latency feature vector, eye-tracking verification confidence, user historical training moving average features, and user profile embedding vector can be input into the difficulty adaptive adjustment model. The difficulty adaptive adjustment model performs multi-layer feature extraction and high-order feature abstraction on the input latency feature vector, eye-tracking verification confidence, user historical training moving average features, and user profile embedding vector. Based on the processed results, dimensional analysis is performed on the latency feature vector to identify abnormal latency dimensions within it. When an abnormal latency dimension is identified... When the visual-motor accommodation delay exceeds a preset range, a suggested value for the optical parameters of the physical target medium or a physical distance guidance instruction is generated. When a persistently high cognitive judgment delay is detected, a target complexity adjustment instruction is generated to simplify the target complexity or reduce interfering elements. The generated parameter adjustment content is validated using safety rules to constrain the magnitude of a single parameter adjustment within a preset safety range. The suggested values ​​for optical parameters, physical distance guidance instructions, or target complexity adjustment instructions generated by abnormal delays are validated using safety rules to constrain the magnitude of a single parameter adjustment within a preset safety range. Based on the instructions that pass the validation, the corresponding suggested values ​​for optical parameters, physical distance guidance instructions, or target complexity adjustment instructions are output, while keeping other dimensions of the delay characteristics stable.

[0034] It should be explained that the other dimension parameters refer to the latency dimensions in the latency feature vector (auditory encoding latency, visual-motor accommodation latency, cognitive judgment latency, and interaction execution latency) other than the dimension currently identified as abnormal and triggering adjustment. Specifically, when the visual-motor accommodation latency is abnormal, the other dimensions, namely auditory encoding latency, cognitive judgment latency, and interaction execution latency, remain stable; when the cognitive judgment latency is abnormal, the other dimensions, namely auditory encoding latency, visual-motor accommodation latency, and interaction execution latency, remain stable.

[0035] In this optional embodiment, when collecting ambient illuminance data and the actual distance between the user and the physical target medium, and calculating the effective pupil aperture and equivalent depth of field in the current environment based on the equivalent refractive power model, if it is detected that the training difficulty is artificially reduced due to excessively bright environment or excessively close distance, corresponding compensation and voice guidance are executed. Ambient illuminance values ​​can be collected, the anchor point position of the physical target medium can be identified, and the actual physical distance between the user and the physical target medium can be estimated by combining focal length parameters. Based on the pupil accommodation model, the pupil size is estimated according to the current environmental conditions, and the depth of field compensation term is calculated. Simultaneously, a brightness compensation term is calculated based on the collected ambient illuminance values, standard illuminance values, standard training distance, and preset brightness compensation coefficients. The actual physical distance, depth of field compensation term, and brightness compensation term are input into the equivalent refractive power model. The system calculates the equivalent refractive power using a diopter model and infers the effective pupil aperture and equivalent depth of field corresponding to the current environment based on the equivalent refractive power. It determines whether the ambient illuminance exceeds the preset standard illuminance range and whether the actual physical distance deviates from the sum of the standard training distance and the compensation distance, to detect whether the training difficulty is artificially reduced due to excessively bright environments or excessively close distances. When the ambient illuminance exceeds the preset range, it outputs a voice prompt to guide the user to adjust the lighting environment. When the actual physical distance is less than the sum of the standard training distance and the compensation distance, it outputs a voice prompt to guide the user to adjust the position of the physical target medium. Based on the degree of deviation between the ambient illuminance and the actual physical distance, it fine-tunes the distance between the user and the physical target medium to achieve training difficulty compensation adjustment, and constrains all compensation adjustment operations within a preset safety range.

[0036] In this optional embodiment, based on the reaction latency variance of the user's historical training data, the personalized safety load threshold for the day is dynamically calculated, the effective training load is accumulated in real time, and progressive feedback adjustment is initiated when the threshold warning interval is reached. When the accumulated load exceeds the safety limit, the training entry is paused and a recovery suggestion is generated. Remote authorization unlocking is also supported. The system can obtain the user's most recent effective training reaction latency data, calculate the latency mean and latency standard deviation, and dynamically calculate the personalized elastic safety load threshold for the day based on the mean and standard deviation. A corresponding warning interval threshold is set according to the elastic safety load threshold, and the user's reaction latency is accumulated in real time during training. The current effective training load; when the cumulative effective training load is in the normal range below the warning range, virtual stimulus resources are generated according to standard rules and regular interactive feedback is maintained; when the cumulative effective training load enters the warning range but does not exceed the elastic safety load threshold, progressive feedback adjustment is initiated, reducing the probability of stimulus resource generation and increasing the interactive delay duration; when the cumulative effective training load reaches or exceeds the elastic safety load threshold, the training entry is directly suspended and corresponding eye recovery suggestions are generated; after the training entry is suspended, training permissions can be unlocked through remote authorization via a professional terminal, or the training entry can be automatically restored the next day.

[0037] In this optional embodiment, when encrypting and uploading latency feature vectors, environmental optical compensation related data, and effective training load related data to the cloud, and remotely overwriting terminal training parameters to support bidirectional synchronization and version management of training schemes, difficulty curves, and security thresholds, the latency feature vectors, environmental optical compensation related data, and effective training load related data can be obtained, and the latency feature vectors, environmental optical compensation related data, and effective training load related data can be desensitized and encrypted. After the training session ends, the encrypted latency feature vectors, environmental optical compensation related data, and effective training load related data are encapsulated into a data packet and uploaded to the cloud collaborative platform via encrypted transmission. During startup or timed heartbeat synchronization, the terminal queries the latest data from the cloud. The system prioritizes policy versions and remote configuration commands. When a new version of the policy or remote configuration command exists in the cloud, the terminal downloads and updates the locally cached priority policy and training parameter configuration. It performs digital signature verification on remotely issued training schemes, difficulty curves, and security threshold-related configurations, and executes parameter overwriting after successful verification. For ordinary training parameters, a gradual transition method is used to execute remote configuration overwriting; for critical security parameters, the changes take effect directly after verification. During parameter transition and overwriting, the system monitors the user's training status in real time; if an anomaly occurs, parameter adjustment is paused and reverted to secure parameters. It records the remote command arbitration results, configuration overwriting process, and execution logs, and encrypts and uploads the execution logs to the cloud to achieve bidirectional synchronization and version management of training schemes, difficulty curves, and security thresholds.

[0038] Figure 2An embodiment of a visual training adaptive management system according to the present invention is shown.

[0039] In this optional embodiment, the visual training adaptive management system includes: The physical focus verification module 201 is used to output audio guidance instructions to guide the user's gaze away from the display screen and focus on the physical target medium, acquire the user's eye image, and verify whether the user has completed the real physical space focus switch based on the relative position change of the corneal reflective spot and the center of the pupil combined with micro-eye movement feature analysis. The delay feature construction module 202 is used to record the total response time from the end of the audio command to the user's interactive input confirmation time after the physical focus switching verification is passed, decompose the total response time into multiple sub-time period vectors, and construct a delay feature vector containing multiple sub-time periods. The difficulty adaptive adjustment module 203 is used to input the time delay feature vector into the preset difficulty adaptive adjustment model, identify the time delay anomaly dimension, and dynamically adjust the optical parameter suggestion value, physical distance guidance command or visual object complexity of the physical visual object medium according to the anomaly dimension. The environmental optical compensation module 204 is used to collect environmental illuminance data and the actual distance from the user to the physical target medium, and calculate the effective pupil aperture and equivalent depth of field in the current environment based on the equivalent refractive power model. When it is detected that the training difficulty is artificially reduced due to excessively bright environment or too close distance, corresponding compensation and voice guidance are performed. The physiological load safety management module 205 is used to dynamically calculate the personalized safety load threshold for the day based on the reaction delay variance of the user's historical training data, accumulate the effective training load in real time, start progressive feedback adjustment when the threshold warning range is reached, suspend the training entry and generate recovery suggestions when the accumulated load exceeds the safety limit, and support remote authorization unlocking. The cloud synchronization and remote configuration module 206 is used to encrypt and upload latency feature vectors, environmental optical compensation related data, and effective training load related data to the cloud, and to remotely overwrite the terminal training parameters. It supports bidirectional synchronization and version management of training schemes, difficulty curves, and security thresholds.

[0040] To facilitate understanding of the above technical solutions of the present invention, the following further explains the above technical solutions of the present invention from the perspective of architecture and principle, such as... Figure 4 As shown, the details are as follows: I. Training Interaction and Data Acquisition.

[0041] 1. System initialization: The user starts the client, loads the personalized training configuration, and initializes the image acquisition, light sensing, and audio modules.

[0042] 2. Physical focus guidance: The system outputs audio instructions containing background noise (such as "Please observe the direction of the third line of visual targets") to guide the user's gaze away from the screen and focus on the paper visual target card.

[0043] 3. Eye movement feature verification: The terminal camera captures eye images at a frame rate of ≥60fps. By analyzing the relative displacement sequence between the corneal reflective spot and the center of the pupil, and combining it with frequency domain analysis, micro-eye movement features (3-5Hz micro-tremors) are identified to confirm the real physical focusing behavior. At the same time, the system continuously monitors changes in the direction of the gaze: a preset physical target area coordinate range A and a display screen area coordinate range B are used. When the position of the corneal reflective spot moves from area A to area B, it is determined that the user's gaze has moved back to the screen.

[0044] 4. Delay Vector Decomposition: After the user completes visual target recognition, blind touch feedback is provided on a low-brightness interface. The system records key time points: : The start time of the audio command.

[0045] : The time when the audio command ends.

[0046] Eye-tracking focusing feature completion time: The moment when the displacement amplitude of the corneal reflected light spot is lower than a preset threshold (e.g., 0.5°) and remains stable for more than a preset duration (e.g., 100ms) indicates that physical adjustment is complete.

[0047] Cognitive decision completion moment: Based on the completion of eye-tracking focusing, detect the moment when the position of the corneal reflected light spot deviates from the visual target area.

[0048] Interaction confirmation moment: The moment when the user touches the screen with their finger and triggers a click event.

[0049] Calculate: Auditory coding delay ; Visual adjustment delay Cognitive judgment delay Interactive execution latency Total response time .

[0050] 5. Training performance data generation: Accuracy determination: The direction of the user's blind click is compared with the system's preset standard direction. If they match, it is marked as correct; otherwise, it is marked as incorrect.

[0051] Effective reaction time delay extraction: If the accuracy is determined to be correct, then extract the vision-motion accommodation delay of this training session. With cognitive judgment delay The sum of these values ​​serves as the effective response time delay. If the error is detected, the reaction delay is recorded as invalid.

[0052] Performance score calculation: Based on the number of effective training sessions and accuracy per unit time, a performance score (Score) is calculated for each training session. The calculation formula is as follows: ; in, For the correct number of times, Total session duration For the average effective response time, This is the standard baseline latency for this user age group. and These are weighting coefficients, exemplarily set to 0.6 and 0.4.

[0053] Data encapsulation: Performance scores, accuracy, and effective reaction time are encapsulated into training performance data for subsequent incentive generation and difficulty adjustment.

[0054] II. Difficulty Adaptive Adjustment Model Based on Deep Neural Networks

[0055] 1. Construct a difficulty adaptive adjustment model based on deep neural networks (hereinafter referred to as the difficulty adaptive model) to predict the optimal training difficulty adjustment parameters based on real-time physiological feedback data of users.

[0056] 1.1 Model Architecture: This model employs a deep neural network structure combining a multilayer perceptron and feature embedding. The overall model architecture includes an input layer, a feature extraction layer, a fusion decision layer, and an output layer, with the specific connections as follows: 1.1.1 Input Layer: Receives the preprocessed feature vectors, with dimensions of... .

[0057] It includes two types of input channels: Real-time physiological channel: Inputs the micro-delay vector of the current training round. and eye-tracking verification confidence .

[0058] Historical status channel: Access the moving average features of the user's past N training sessions (such as the average reaction time and fatigue growth rate in the past 7 days) and the user profile embedding vector (age, initial vision level).

[0059] 1.1.2 Feature Extraction Layer: Contains 3 fully connected layers, each followed by Batch Normalization (BN) and Rectified Linear Activation Function (ReLU).

[0060] The first fully connected layer has 64 nodes and is used to extract low-dimensional features from the time delay vector.

[0061] The second fully connected layer has 32 nodes and is used to fuse historical state information.

[0062] The third fully connected layer has 16 nodes and is used for high-order feature abstraction.

[0063] Connection relationship: Input layer → Fully connected layer 1 - BN - ReLU - Fully connected layer 2 - BN - ReLU - Fully connected layer 3 - BN - ReLU.

[0064] 1.1.3 Fusion Decision Layer: Combines the output vector of the feature extraction layer with the current environmental optical compensation coefficient. The data is then stitched together. An attention mechanism module dynamically weights physiological and environmental characteristics to address fluctuations in user performance under different lighting conditions.

[0065] 1.1.4 Output Layer: Employs a linear activation function, with an output dimension of... eigenvectors The output data is directly mapped to the physical parameter adjustment amounts of the rehabilitation training equipment.

[0066] 1.2 Input and Output Data Definitions and Domain Relevance: To ensure the specific integration of the algorithm with the visual rehabilitation scenario, the input and output data definitions of the model are shown in Table 1.

[0067] Table 1 defines the input and output data of the model. 1.3 Model Training Process: The model is trained offline on a cloud server, and the trained weights are then deployed to the terminal or cloud inference engine. The specific training steps are as follows: 1.3.1 Construction of training dataset: Data source: Collected no less than 10,000 historical user training session data, covering different ages, different visual acuity, and different ambient lighting conditions.

[0068] Labels are generated using a combination of expert rules and manual review: Initial labels are manually adjusted by senior optometrists based on the user's average reaction time and accuracy during the training session, according to clinical guidelines (e.g., +0.25D, -0.25D, or remain unchanged).

[0069] For subsequent data, reinforcement learning feedback is used: if a user's completion rate significantly improves in the next training session and there are no fatigue alarms, the current data is marked as a "positive sample"; if a user makes consecutive errors or quits midway, it is marked as a negative sample. Z-score standardization is applied to the input features to eliminate the influence of unit dimensions.

[0070] 1.3.2 Loss Function: The mean squared error loss function is used as the main loss to measure the difference between the prediction adjustment and the expert label.

[0071] ; in, This represents the mean squared error loss value, in units of non-negative real numbers. The smaller the value, the more accurate the prediction. N This indicates the number of samples in the training batch. The unit / value range is a positive integer, and for example, it is set to 64, which is consistent with the batch size. i This represents the sample index, with units / value ranges from 1 to... N An integer is used to iterate through each training sample in the batch. The model represents the first i The predicted output vector for each sample, in units / range: a three-dimensional vector. These correspond to the lens diopter, optotype size, and rest duration, respectively. Indicates the first i The expert-annotated label vector for each sample, with units and a range of values: The labeling is done by a senior optometrist according to clinical guidelines. This represents the squaring operation, which amplifies the penalty weight for larger errors.

[0072] A regularization term is introduced to prevent overfitting, and a safety penalty term is added. When the difficulty adjustment of the model output exceeds the physiological safety threshold (e.g., a single adjustment exceeds ±1.00D), a high penalty value is applied. ; in, This represents the safety penalty loss value, in units of non-negative real numbers, and the penalty is only applied when the predicted value exceeds the limit. j This indicates the output dimension index, with the unit / value range as follows: (Refractive power) or (Target size), rest duration T rest There is no hard upper limit. The model predicts the first j Dimensional parameter values, units / range: ±0.25~±2.00D; ±5%~±30%, where the absolute value represents the adjustment range. Indicates the first j Safety threshold for dimension parameters 0.50D; 15%, the maximum allowable range for a single adjustment. This represents the rectification function, which only calculates penalties for out-of-limit parts; normal values ​​are not penalized.

[0073] Total loss function: λ is the safety weight coefficient (default value is 10.0), and the larger the value, the stronger the constraint on the safety boundary. This represents the overall optimization objective for model training. An example value is a non-negative real number, which is minimized during backpropagation. This represents the mean squared error loss, ensuring prediction accuracy. This indicates a safety penalty, ensuring that the output remains within physiologically safe limits.

[0074] 1.3.3 Optimization and Hyperparameter Settings: Optimizer: The Adam optimizer is used, with the initial learning rate set to 0.001 and a learning rate decay strategy set.

[0075] Batch size: Set to 64.

[0076] Training epochs: Set to 100, and use early stopping method. Stop training when the validation set loss does not decrease for 5 consecutive epochs.

[0077] Dropout: Setting the Dropout rate to 0.3 after a fully connected layer enhances the model's generalization ability.

[0078] 1.4 Reasoning Applications Combined with Scenarios: In the actual user training process, the logic of combining model inference with the rehabilitation scenario is as follows: Real-time inference: After each training unit is completed (e.g., 10 flips), the terminal will collect the real-time feature vectors. Input the model and get the predicted output. .

[0079] Safety guardrails constraints: The model output is not executed directly and must be verified by the rule engine.

[0080] Rule 1 (Amplitude Limitation): Regardless of the model output, the absolute value of a single lens diopter adjustment must not exceed 0.50D to prevent accommodative spasm.

[0081] Rule 2 (Fatigue Circuit Breaker): If the load management module has already triggered the warning zone state, the model output will be forcibly overwritten, the locking difficulty will not increase, and this rule will be executed first. suggestion.

[0082] Parameter execution: If If the value is >0, the system prompts the user to replace the lens with a higher diopter for reversal imaging. If the value is less than 0, the system replaces the visual target card, increasing the difficulty of resolution. >0, system forcibly locks interface. For a few seconds, guide the user to close their eyes and rest.

[0083] 1.5 Technical Effects: Through the above model construction, the present invention achieves the following technical effects: 1.5.1 Nonlinear Mapping Capability: Compared to traditional threshold judgment, the difficulty adaptive model can capture the nonlinear relationship between the time delay vector and the difficulty parameter (e.g., in a strong light environment, the same...). (Extended duration may indicate more severe fatigue) and improve adjustment precision.

[0084] 1.5.2 Personalized Adaptation: By embedding historical state channels, the model can identify individual differences among users (such as the naturally slower adjustment ability of elderly users), avoiding a one-size-fits-all approach to difficulty assessment.

[0085] 1.5.3 Explainability and Security: By directly mapping the output layer to specific rehabilitation parameters (refractive power, target size) and combining it with the rule engine guardrail, the security and feasibility of AI decision-making in medical rehabilitation scenarios are ensured.

[0086] III. Realization of environmental optical compensation based on equivalent refractive power model.

[0087] 1.1 Parameter Acquisition: Light sensing module: Collects ambient illuminance values ​​(lux).

[0088] Image module: Identifies anchor points on the target card and estimates the physical distance (cm) by combining focal length parameters.

[0089] 1.2 Model Structure Description: The equivalent refractive error model used in this invention is a regression model guided by physical formulas, and its core expression is: ; in: Indicates equivalent diopter (unit: D, diopter); Indicates the actual physical distance between the target card and the user (unit: m); This represents the depth-of-field compensation term, which is related to the pupil diameter. It is calculated using an empirical formula after estimating the pupil size using the Stanley-Davies pupil model. This represents the brightness compensation term, which is positively correlated with the degree to which the ambient illuminance deviates from the standard value. It is used to reflect the impact of changes in light on the human eye's ability to adjust.

[0090] 1.3 Parameter settings are shown in Table 2: Table 2 Parameter Settings The brightness compensation term is calculated as follows: .

[0091] 1.4 Verification of Technical Effectiveness: To verify the effectiveness of this model, the applicant conducted a small-scale controlled experiment. The specific process and data are as follows: 1.4.1 Experimental Design: Experiment type: Randomized controlled trial (single-blind).

[0092] Sample size: 30 people (15 in the experimental group and 15 in the control group), aged 8-30 years, with mild regulatory dysfunction.

[0093] Comparison of options: Experimental group: Difficulty adjustment was achieved using environmental optical compensation based on the equivalent refractive power model of this invention.

[0094] Control group: A fixed difficulty scheme was adopted, using standard ±1.50D reverse shots throughout, with the difficulty parameters remaining unchanged.

[0095] Training cycle: 14 consecutive days, 15 minutes per session.

[0096] 1.4.2 Key evaluation indicators are shown in Table 3: Table 3 Key Assessment Indicators 1.4.3 Exemplary experimental data are shown in Table 4: Table 4 Exemplary Experimental Data 1.4.4 Experimental Conclusions: (1) Effectiveness advantage: The experimental group showed a significant improvement in regulatory flexibility (+3.8 cycles / minute) compared to the control group (+1.5 cycles / minute), proving that dynamic difficulty adaptation is more effective than fixed difficulty in stimulating ciliary muscle regulatory function.

[0097] (2) Efficiency improvement: The effective response time of the experimental group was significantly shortened (-87ms vs -28ms), which verified the training efficiency advantage of the model through targeted regulation of the time delay vector (adjusting only the abnormal dimension).

[0098] (3) Safety assurance: The subjective fatigue score of the experimental group decreased (-1.2 points), while that of the control group increased (+2.3 points), indicating that the elastic load threshold control of the model effectively avoided overtraining caused by fixed difficulty.

[0099] 1.5 Compensation Execution: If the ambient illuminance is >600 lux or <300 lux, the system will prompt the user to adjust the lighting environment via voice. If the ambient illuminance deviates from the standard illuminance (450 lux), the system will slightly adjust the visual training difficulty by fine-tuning the distance between the optotype and the user (i.e., the brightness compensation distance) based on the deviation value. If the estimated distance is insufficient (standard 40cm + / - brightness compensation distance), the system will prompt the user to adjust the position of the optotype. All compensation adjustments are performed within the preset safety range to ensure the safety and comfort of the training process.

[0100] IV. Elastic physiological load threshold control and incentive generation.

[0101] 1.1 Example values ​​for physiological safety threshold and fatigue index: 1.1.1 Flexible safety load threshold ( ): For example, it is set to the mean of the user's average effective training latency over the past 7 days plus twice the standard deviation (μ + 2σ). If the user's average reaction latency is 500ms and the standard deviation is 50ms, then For example, it is set to 600ms.

[0102] 1.1.2 Warning interval threshold: Example setting: 0.85× When the cumulative load reaches 85% of the water level, a soft deprivation intervention is triggered.

[0103] 1.1.3 Consecutive error rate threshold: For example, it is set to 30%. If the number of errors exceeds 3 in 10 consecutive training sessions, it is judged as too difficult or too fatigued, triggering a difficulty reduction or rest suggestion.

[0104] 1.1.4 Fatigue Index Range: The fatigue index is normalized to the range [0,1]. For example, when the fatigue index > 0.8, it is determined to be a state of deep fatigue, and training is forcibly paused; when 0.6 < fatigue index ≤ 0.8, it is determined to be mild fatigue, and incentive decay is triggered.

[0105] 1.1.5 Visibility-Motion Adjustment Delay Anomaly Threshold: An example setting is 600ms. If... If the time exceeds 600ms for three consecutive times, it is determined to be ciliary muscle slowness, triggering a reduction in optical difficulty.

[0106] 1.2 Definition and Generation Logic of Virtual Incentive Resources: 1.2.1 Types of Virtual Incentive Resources: The virtual incentive resources include, but are not limited to: Points-based: Training points used to redeem physical prizes or value-added services.

[0107] Decorative items: Virtual badges and theme skins used to personalize the settings interface.

[0108] Unlocking type: Access keys used to unlock higher-level training levels or advanced analysis reports.

[0109] 1.2.2 Rarity Level Setting: Incentive resources are divided into three levels: Common (R), Rare (SR), and Epic (SSR).

[0110] 1.2.3 Correlation between Fall Probability and Performance: Establish a mapping relationship between training performance data and fall probability. An example is set with a base fall probability of... The performance adjustment factor is : Rare resource drop rate: .

[0111] Epic resource drop rate: .

[0112] in, It is positively correlated with the performance score. For example, when the score is greater than 90, =0.5 (probability increased by 50%); when Score < 60, =-0.5 (probability reduced by 50%).

[0113] 1.2.4 Generation Steps: After the training unit is completed, the system generates a random number in the range [0,1]. If the random number is less than the currently calculated drop probability, the corresponding level of virtual incentive resources are generated and displayed to the user.

[0114] 1.3 State Transition and Feedback Regulation: Normal zone (load < 0.85 × ): Normally generate virtual incentive resources and maintain standard interactive feedback.

[0115] Warning zone (0.85× ≤load< The probability of stimulus generation decays linearly, and each interaction confirmation injection has a random delay of 0.4-2.0s.

[0116] Restricted area (load ≥ ): Pause the training entry, display recovery suggestions, and support remote authorization unlocking from the professional end or automatic recovery the next day.

[0117] V. AI-Professional Collaborative Instruction Execution Mechanism Based on Priority Arbitration.

[0118] This addresses the potential conflict between local adaptive AI decision-making and remote professional intervention. By constructing an instruction priority arbitration mechanism and a parameter smoothing transition strategy, it ensures the continuity of the user training experience while safeguarding medical safety and professional authority.

[0119] 1.1 Instruction Priority Hierarchy: The system divides training control commands into three priority levels, with different execution strategies for each level: 1.1.1 First Priority (Emergency Intervention Level): Source: Manually and immediately distributed by doctors.

[0120] Triggering scenarios: abnormal user physiological data, continuous training failures, or forced adjustments before a follow-up visit.

[0121] Execution strategy: Immediate interruption. Regardless of the current state of the local AI, immediately pause the current training process and force a rest, lock the difficulty, or terminate training.

[0122] 1.1.2 Second Priority (Solution Configuration Level): Source: Pre-set training plan or phase adjustment instructions from doctors.

[0123] Triggering scenarios: New user onboarding, and plan updates after a period of follow-up visits.

[0124] Execution strategy: Smooth transition. Instead of immediately interrupting the current training unit, the parameters are gradually adjusted to the target value after the current unit ends.

[0125] 1.1.3 Third Priority (Local Adaptive Level): Source: Real-time decision-making in difficulty-adaptive models.

[0126] Triggering scenario: Micro-level difficulty adjustment after each round of training.

[0127] Execution strategy: Autonomous execution. Without conflicting with higher-priority instructions, the local AI autonomously adjusts its actions based on real-time physiological feedback.

[0128] 1.2 The arbitration and enforcement process for instructions includes the following steps: 1.2.1 Command Reception and Parsing: During training breaks or background heartbeat cycles, the terminal client queries the cloud for commands to be executed. If a command is received, the digital signature is first verified, the command content is parsed, and the command type, target parameter value, and priority identifier are extracted.

[0129] 1.2.2 Priority Arbitration: The arbitration module inside the terminal compares the priority of the received remote command with the priority of the currently executing local AI decision. If the remote command has the highest priority, it is determined to have the highest authority, and an interrupt control signal is generated directly. Furthermore, the priority arbitration also involves conflict handling with the safety load threshold: if the remote intervention command is at the emergency intervention level (P0), its authority is higher than the safety load threshold, and the pause state can be forcibly lifted or a forced rest can be executed; if the remote command has the second priority, conflict detection is initiated. If no remote command is received, or the priority of the remote command is lower than the local decision, the local AI decision is maintained.

[0130] 1.2.3 Conflict Detection: When a second-priority remote command exists, the arbitration module calculates the difference between the parameters the local AI intends to adjust and the target parameters of the remote command. Conflicting directions: For example, the local AI suggests reducing the difficulty, while the professionals instruct that the difficulty should be increased.

[0131] Threshold conflict: For example, the safety load threshold set by the local AI is higher than the safety limit set by the doctor.

[0132] If a conflict is detected, the system is marked as requiring arbitration, with the remote command target value as the final target; if no conflict is detected, the commands are merged and executed.

[0133] 1.2.4 Smooth Transition Execution: To avoid user discomfort caused by sudden parameter changes, a smooth transition strategy is adopted for second-priority instructions. Parameter Classification: Training parameters are divided into key safety parameters (such as mandatory rest signs and maximum training duration) and general experience parameters (such as lens diopter fine-tuning and target difficulty).

[0134] Immediate effect: Critical safety parameters take effect immediately after the instruction is verified, ensuring medical safety boundaries.

[0135] Gradual adjustment: Instead of jumping directly to the target value, the normal experience parameter is given a transition period (e.g., 3 training units or 60 seconds). In each period, the current parameter value is moved towards the target value by a certain percentage (e.g., 30% of the difference each time) until the target value is reached.

[0136] Dynamic monitoring: If abnormal fluctuations in the user's physiological indicators are detected during the transition (such as a sudden and significant increase in response time), the transition will be paused and the system will revert to the safe parameter values.

[0137] 1.2.5 Execution Log and Feedback: After the instruction is executed, the terminal generates a structured execution log, including the instruction source, arbitration result, actual execution parameters, transition time, and user's subsequent training performance. This log is encrypted and uploaded to the cloud for doctors to view and for subsequent model optimization.

[0138] 1.3 Key Indicator Definitions and Transmission Protocols: Key Metric Fields List: Key metrics for real-time synchronization should include at least the following: Identity identifier: Hash value of user ID (de-identified).

[0139] Core latency: Visual-motor accommodation latency Cognitive judgment delay .

[0140] Training results: accuracy rate and number of loops completed in this training session.

[0141] Safety status: Current cumulative load percentage, whether circuit breaker protection has been triggered.

[0142] Environmental parameters: ambient illuminance during training, estimated target distance.

[0143] Upload method: Encrypted transmission using HTTPS protocol. After training is complete, the terminal encapsulates the above key indicators into a JSON format data packet and sends it to the designated interface of the cloud collaboration platform via a POST request.

[0144] Definition of immediacy: Immediate upload means initiating an upload request within 5 seconds after the training session ends; if the network is unavailable, the data is temporarily stored in a local encrypted database and uploaded in a priority queue after the network is restored.

[0145] 1.4 Priority Policy Update Mechanism: Policy configuration and distribution: The professional management interface provides a priority policy configuration panel, allowing professionals to modify the weight parameters of each level of command.

[0146] Version management: Each policy modification generates a new policy version number, and policy change logs are recorded in the cloud.

[0147] Terminal synchronization: Each time the terminal client starts up or performs a heartbeat synchronization, it checks the cloud policy version number. If the version number has changed, it downloads the latest priority policy configuration file and overrides the locally cached policy.

[0148] Feedback-based automatic optimization: The system records user training performance after each policy conflict. If, under a certain type of conflict, the completion rate of users following professional instructions is significantly higher than that of users following AI instructions, the system will automatically mark the scenario and suggest increasing the default weight of professional instructions in subsequent version updates.

[0149] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores static and dynamic information data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0150] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0151] In addition, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0152] In addition, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0153] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0154] This invention is not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this invention is limited only by the appended claims.

Claims

1. A visual training adaptive management method, characterized in that, include: Output audio guidance commands to guide the user's gaze away from the display screen and focus on the physical target medium, acquire the user's eye image, and verify whether the user has completed the real physical space focus switch based on the relative position change of the corneal reflective spot and the pupil center combined with micro-eye movement feature analysis. After the physical focus switching verification is passed, the total response time from the end of the audio command to the user's interactive input confirmation time is recorded. The total response time is decomposed into multiple sub-time period vectors, and a delay feature vector containing multiple sub-time periods is constructed. Input the time delay feature vector into the preset difficulty adaptive control model, identify the time delay anomaly dimension, and dynamically adjust the suggested values ​​of optical parameters of the physical target medium, physical distance guidance instructions or target complexity according to the anomaly dimension; Collect ambient illuminance data and the actual distance from the user to the physical target medium, and calculate the effective pupil aperture and equivalent depth of field in the current environment based on the equivalent refractive power model. When it is detected that the training difficulty is artificially reduced due to excessively bright environment or too close distance, corresponding compensation and voice guidance are performed. Based on the reaction latency variance of users' historical training data, the system dynamically calculates the personalized safety load threshold for the day, accumulates the effective training load in real time, initiates progressive feedback adjustment when the threshold warning range is reached, and pauses the training entry and generates a recovery suggestion when the accumulated load exceeds the safety limit. It also supports remote authorization unlocking. The system encrypts and uploads latency feature vectors, environmental optical compensation data, and effective training load data to the cloud, and remotely overwrites the terminal training parameters. It also supports bidirectional synchronization and version management of training schemes, difficulty curves, and security thresholds.

2. The visual training adaptive management method according to claim 1, characterized in that, The output audio guidance command guides the user's gaze away from the display screen and focuses on the physical target medium, acquires the user's eye image, and verifies whether the user has completed a true physical spatial focus switch based on the relative position change of the corneal reflected light spot and the pupil center combined with micro-eye movement feature analysis. Start the client and load the personalized training configuration, and initialize the image acquisition, light sensing and audio-related components; It outputs audio commands containing background noise to guide the user's gaze away from the display screen and focus on the paper visual markers; The system acquires user eye images at a frame rate no lower than the preset frame rate, and identifies micro-tremor and micro-eye movement characteristics based on the relative displacement sequence of corneal reflected light spots and pupil center, combined with frequency domain analysis, to confirm the real physical focusing behavior. The system presets the coordinate range of the physical target area and the coordinate range of the display screen area, continuously monitors the position of the corneal reflective spot, and determines that the user's gaze has moved back to the display screen when the spot position moves from the coordinate range of the physical target area to the coordinate range of the display screen area.

3. The visual training adaptive management method according to claim 1, characterized in that, After the physical focus switching verification is passed, the total response time from the end of the audio command to the user's interactive input confirmation time is recorded. The total response time is decomposed into multiple sub-time period vectors, and a delay feature vector containing multiple sub-time periods is constructed, including: After the user completes the visual target recognition, a blind touch feedback operation is performed on the low-brightness interface. Record key time nodes, including: the start time of the audio instruction, the end time of the audio instruction, the completion time of eye-tracking focusing features, the completion time of cognitive decision-making, and the interaction confirmation time. Based on the recorded key time points, the auditory encoding delay, visual-motor accommodation delay, cognitive judgment delay, and interaction execution delay are calculated respectively. Calculate the total response delay based on the end time of the audio command and the time of interaction confirmation; The latency feature vector is constructed by combining auditory encoding latency, visual-motor accommodation latency, cognitive judgment latency, and interactive execution latency.

4. The visual training adaptive management method according to claim 1, characterized in that, The step of inputting the time delay feature vector into a preset difficulty adaptive adjustment model, identifying time delay anomaly dimensions, and dynamically adjusting the suggested values ​​of optical parameters, physical distance guidance commands, or target complexity of the physical target medium based on the anomaly dimensions includes: The model incorporates time-delay feature vectors, eye-tracking verification confidence, user history training moving average features, and user profiles into a vector input difficulty adaptive adjustment model. The difficulty adaptive adjustment model performs multi-layer feature extraction and high-order feature abstraction on the input time delay feature vector, eye-tracking verification confidence, user history training moving average features, and user profile embedding vector. Based on the processed results, dimensional analysis is performed on the delay feature vector to identify abnormal delay dimensions in the delay feature vector; When the visual motion accommodation delay is detected to exceed the preset range, a suggested value for the optical parameters of the physical target medium or a physical distance guidance command is generated. When a persistently high cognitive judgment delay is detected, a visual target complexity adjustment instruction is generated to simplify the visual target complexity or reduce interfering elements. The generated parameter adjustment content is validated by safety rules to constrain the magnitude of a single parameter adjustment within a preset safety range; The recommended values ​​of optical parameters, physical distance guidance instructions, or visual target complexity adjustment instructions generated by abnormal time delays are verified by safety rules, and the magnitude of a single parameter adjustment is constrained within a preset safety range. Based on the verified instructions, output the corresponding suggested values ​​for optical parameters, physical distance guidance instructions, or target complexity adjustment instructions.

5. The visual training adaptive management method according to claim 1, characterized in that, The system collects ambient illuminance data and the actual distance between the user and the physical target medium, and calculates the effective pupil aperture and equivalent depth of field in the current environment based on an equivalent refractive power model. When it detects that the training difficulty has been artificially reduced due to excessively bright ambient light or excessively close distance, corresponding compensation and voice guidance are performed, including: Collect ambient illuminance values, identify the anchor point position of the physical target medium, and estimate the actual physical distance between the user and the physical target medium by combining the focal length parameters; Based on the pupil adjustment model, the pupil size is estimated according to the current environmental conditions and the depth compensation term is calculated. At the same time, the brightness compensation term is calculated based on the collected ambient illuminance value, standard illuminance value, standard training distance and preset brightness compensation coefficient. The actual physical distance, depth of field compensation term, and brightness compensation term are input into the equivalent refractive power model to calculate the equivalent refractive power, and the effective pupil aperture and equivalent depth of field corresponding to the current environment are inferred based on the equivalent refractive power. Determine whether the ambient illuminance exceeds the preset standard illuminance range, and whether the actual physical distance deviates from the sum of the standard training distance and the compensation distance, in order to detect whether the training difficulty is artificially reduced due to excessively bright environment or excessively close distance; When the ambient illuminance exceeds the preset range, a voice prompt is output to guide the user to adjust the lighting environment. When the actual physical distance is less than the sum of the standard training distance and the compensation distance, a voice prompt is output to guide the user to adjust the position of the physical target medium. The distance between the user and the physical target medium is finely adjusted based on the degree of deviation between the ambient illuminance and the actual physical distance to achieve training difficulty compensation adjustment, and all compensation adjustment operations are constrained to be performed within a preset safety range.

6. The visual training adaptive management method according to claim 1, characterized in that, The system dynamically calculates the personalized safety load threshold for the day based on the reaction latency variance of the user's historical training data, accumulates the effective training load in real time, initiates progressive feedback adjustment when the threshold warning range is reached, and pauses the training entry and generates a recovery suggestion when the accumulated load exceeds the safety limit. It also supports remote authorization unlocking, including: Obtain the user's most recent effective training reaction latency data, calculate the latency mean and latency standard deviation, and dynamically calculate the personalized elastic safety load threshold for the day based on the mean and standard deviation; Set corresponding early warning interval thresholds based on the elastic safety load threshold, and accumulate the user's current effective training load in real time during the training process; When the cumulative effective training load is in the normal range below the warning range, virtual stimulus resources are generated according to standard rules and regular interactive feedback is maintained. When the cumulative effective training load enters the warning range but does not exceed the elastic safety load threshold, a gradual feedback adjustment is initiated to reduce the probability of generating incentive resources and increase the interaction delay duration. When the cumulative effective training load reaches or exceeds the elastic safety load threshold, the training entry will be directly suspended and corresponding eye recovery suggestions will be generated. After the training access is suspended, training permissions can be unlocked remotely via a professional terminal, or you can wait for the training access to be automatically restored the next day.

7. The visual training adaptive management method according to claim 1, characterized in that, The process of encrypting and uploading latency feature vectors, environmental optical compensation data, and effective training load data to the cloud, and remotely overwriting terminal training parameters, supporting bidirectional synchronization and version management of training schemes, difficulty curves, and security thresholds includes: Acquire time delay feature vectors, environmental optical compensation related data, and effective training load related data, and perform desensitization and encryption processing on the time delay feature vectors, environmental optical compensation related data, and effective training load related data; After the training session ends, the encrypted latency feature vector, environmental optical compensation data, and effective training load data are packaged into a data packet and uploaded to the cloud collaboration platform via encrypted transmission. During startup or timed heartbeat synchronization, the terminal queries the cloud for the latest priority policy version and remote configuration instructions. When a new version of the policy or a remote configuration command is available in the cloud, the terminal downloads and updates the priority policy and training parameter configuration cached locally. The training scheme, difficulty curve, and security threshold configurations distributed remotely are digitally verified, and the parameters are overwritten after the verification is successful. For ordinary training parameters, a gradual transition is used to overwrite the remote configuration, while for critical security parameters, the configuration takes effect directly after verification. During parameter transition and overwrite, the user's training status is monitored in real time. If an abnormality occurs, parameter adjustment is paused and the parameters are rolled back to safe parameters. Record the remote command arbitration results, configuration overwrite process and execution logs, and encrypt and upload the execution logs to the cloud to achieve bidirectional synchronization and version management of training schemes, difficulty curves and security thresholds.

8. A visual training adaptive management system, characterized in that, include: The physical focus verification module is used to output audio guidance commands to guide the user's gaze away from the display screen and focus on the physical target medium, acquire the user's eye image, and verify whether the user has completed the real physical space focus switch based on the relative position change of the corneal reflective spot and the pupil center combined with micro-eye movement feature analysis. The latency feature construction module is used to record the total response time from the end of the audio command to the user's interactive input confirmation time after the physical focus switching verification is passed. The total response time is decomposed into multiple sub-time period vectors, and a latency feature vector containing multiple sub-time periods is constructed. The difficulty adaptive adjustment module is used to input the time delay feature vector into the preset difficulty adaptive adjustment model, identify the time delay anomaly dimension, and dynamically adjust the optical parameter suggestion value, physical distance guidance command or visual object complexity of the physical visual object medium according to the anomaly dimension. The environmental optics compensation module is used to collect environmental illuminance data and the actual distance between the user and the physical target medium, and calculate the effective pupil aperture and equivalent depth of field in the current environment based on the equivalent refractive power model. When it is detected that the training difficulty is artificially reduced due to excessively bright environment or too close distance, the corresponding compensation and voice guidance are executed. The physiological load safety management module is used to dynamically calculate the personalized safety load threshold for the day based on the reaction delay variance of the user's historical training data, accumulate the effective training load in real time, start progressive feedback adjustment when the threshold warning range is reached, suspend the training entry and generate recovery suggestions when the accumulated load exceeds the safety limit, and support remote authorization unlocking. The cloud synchronization and remote configuration module is used to encrypt and upload latency feature vectors, environmental optical compensation data, and effective training load data to the cloud, and to remotely overwrite the terminal training parameters. It supports bidirectional synchronization and version management of training schemes, difficulty curves, and security thresholds.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images