A method and system for attention training that combines EEG and behavioral data

Abstract

This application relates to a method and system for attention training that combines EEG and behavioral data, belonging to the field of brain-computer interface data processing technology. The method simultaneously collects EEG and behavioral data during attention training tasks, divides continuous EEG time windows based on a unified time reference, and maps behavioral data to corresponding windows; preprocesses the EEG data and calculates a reliability index, extracting attention features to obtain candidate attention values; extracts behavioral attention features to generate behavioral representation values, and performs consistency verification based on the deviation between the two; combines the reliability index and verification results to perform gating correction on the candidate attention values, obtaining fused attention values ​​and dividing attention state segments; adaptively adjusts the training load according to the type and duration of attention state segments; after training, generates evaluation results based on the distribution of attention state segments and changes in behavioral data, determining the initial configuration for the next cycle. This application can improve the stability of attention state recognition in children's scenarios.

Description

Technical Field

[0001] This application relates to the field of brain-computer interface data processing, specifically to a method and system for attention training that combines EEG and behavioral data. Background Technology

[0002] Existing brain-computer interface (BCI) attention training systems collect users' electroencephalogram (EEG) signals using non-invasive scalp electrodes. The collected EEG signals are filtered, amplified, and converted from analog to digital to extract EEG feature parameters related to attention states, including sensorimotor rhythms, theta waves, beta waves, alpha waves, and their proportional relationships in the frequency domain. The system calculates an attention score based on the extracted EEG features and maps this score to dynamic parameters in the training task in real time, such as the speed of a game character's movement, jump height, score changes, and the intensity of audiovisual feedback, forming a neurofeedback training loop. During training, the system continuously monitors changes in EEG features and provides feedback on the attention state to the trainee through a visual interface, helping them learn to regulate their own EEG activity to achieve the target attention level. Such systems have been widely used in areas such as children's cognitive training, learning ability improvement, and attention deficit intervention.

[0003] However, in the context of children's attention training, existing technologies rely solely on EEG signals to determine attention status, which cannot effectively address the need for collaborative processing of multi-source data during training, resulting in a lack of reliable correlation between attention status recognition and training feedback. Summary of the Invention

[0004] This application provides a method and system for attention training that combines EEG and behavioral data to solve the aforementioned technical problems.

[0005] In a first aspect, this application provides a focus training method that combines electroencephalogram (EEG) and behavioral data, the method comprising: S1. During the focus training task, EEG data and behavioral data of the training subjects are collected synchronously. The EEG data is divided into continuous EEG time windows based on a unified time reference, and the behavioral data is mapped to the corresponding EEG time windows. S2. The EEG data within each EEG time window is preprocessed to extract EEG focus features. Based on artifact interference, the EEG reliability index for the corresponding EEG time window is calculated to obtain candidate focus values ​​for each EEG time window. S3. Behavioral focus features are extracted from the behavioral data within each EEG time window to generate corresponding behavioral representation values. A consistency verification result is obtained based on the deviation between the behavioral representation values ​​and the candidate focus values. S4. Based on the EEG reliability index and the... S5. Based on the consistency verification results, the candidate EEG focus values ​​are gating and corrected to obtain the fused focus value for the corresponding EEG time window; S6. Based on the fused focus values ​​of multiple consecutive EEG time windows, the focus state during the training process is continuously determined, and focus state segments are divided; S7. Based on the type and duration of the current focus state segment, the training load of the focus training task is adaptively adjusted, wherein the training load includes at least one of task difficulty, stimulus rhythm, and feedback sensitivity; S8. After the training task is completed, a training evaluation result is generated based on the distribution results of each focus state segment during the training process and the changes in the corresponding behavioral data, and the initial training configuration for the next training cycle is determined based on the training evaluation result.

[0006] Optionally, in step S2, the EEG data within each EEG time window is preprocessed, including: sequentially performing bandpass filtering, power frequency suppression, baseline correction, and artifact detection to obtain preprocessed EEG data; and extracting at least two of the frequency band energy features, frequency band ratio features, and time window fluctuation features based on the preprocessed EEG data to generate the candidate EEG focus value; in step S2, when calculating the EEG reliability index, the following steps are included: calculating the proportion of low-frequency interference components within the current EEG time window to characterize the degree of low-frequency drift caused by eye movement or head movement; calculating the proportion of high-frequency interference components within the current EEG time window to characterize the degree of electromyographic interference; statistically analyzing at least one of the proportion of amplitude mutation points, the proportion of saturated sampling points, and the proportion of adjacent sampling fluctuation abnormalities within the current EEG time window to characterize the degree of motion artifacts; and normalizing and weighting the proportion of low-frequency interference components, the proportion of high-frequency interference components, and the motion artifact characterization to obtain the EEG reliability index.

[0007] Optionally, in step S2, the EEG time window is divided into a high-confidence time window, a time window to be corrected, and a low-confidence time window according to the EEG confidence index: wherein, the EEG candidate focus value corresponding to the high-confidence time window directly enters the consistency check, the EEG candidate focus value corresponding to the time window to be corrected enters the gating correction, and the EEG candidate focus value corresponding to the low-confidence time window is suppressed and directly participates in the focus state determination of the current EEG time window.

[0008] Optionally, in step S3, the behavioral representation value is determined based on the behavioral data corresponding to the current EEG time window. The behavioral data includes at least two of the following: reaction time, answer accuracy, missed answer rate, false touch rate, continuous operation stability, and task switching delay. Generating the behavioral representation value includes: converting the behavioral data into corresponding evaluation quantities representing the degree of focus; normalizing the converted corresponding evaluation quantities; and weighting and synthesizing the normalized corresponding evaluation quantities to obtain the behavioral representation value for the current EEG time window.

[0009] Optionally, in step S3, the consistency verification includes: calculating the deviation value between the behavioral representation value and the candidate focus value of the current EEG time window; comparing the deviation value with a preset deviation threshold; when the deviation value exceeds the preset deviation threshold, determining the current EEG time window as an EEG-behavior inconsistency time window; wherein, the preset deviation threshold is set according to the training task type.

[0010] Optionally, in step S4, gating correction is performed for the EEG-behavioral inconsistency time window. The gating correction includes at least two of the following key implementation steps: attenuating the candidate EEG focus value according to the EEG confidence index of the current EEG time window to obtain the corrected EEG focus value; constraining and synthesizing the corrected EEG focus value with the fused focus value of the previous EEG time window to limit the instantaneous jump variable of the current EEG time window; when the current EEG time window is a low confidence time window, using the behavioral representation value to replace the candidate EEG focus value in the generation of the fused focus value of the current EEG time window.

[0011] Optionally, in step S4, generating the fusion focus value includes: determining the gating coefficient corresponding to the current EEG time window based on the EEG reliability index; using the gating coefficient to correct the candidate EEG focus value to obtain the corrected EEG focus value; and combining the corrected EEG focus value, the behavioral representation value of the current EEG time window, and the fusion focus value of the previous EEG time window to obtain the fusion focus value of the current EEG time window.

[0012] Optionally, in step S5, the focus state segment is obtained by performing sliding statistics on the fused focus values ​​of multiple consecutive EEG time windows. The sliding statistics include at least mean statistics, fluctuation amplitude statistics, and duration statistics, and include the following key implementation steps: when the mean of the fused focus values ​​of multiple consecutive EEG time windows is greater than a first state threshold and the fluctuation amplitude is less than a first fluctuation threshold, the corresponding time period is divided into a stable focus state segment; when the mean of the fused focus values ​​of multiple consecutive EEG time windows is less than a second state threshold, the corresponding time period is divided into a deviated focus state segment; when the fused focus values ​​of multiple consecutive EEG time windows are between the first state threshold and the second state threshold, or the fluctuation amplitude is greater than the second fluctuation threshold, the corresponding time period is divided into a fluctuating transition state segment.

[0013] Optionally, in steps S6 and S7, adaptive adjustment of the training load and generation of training evaluation results include: increasing the training load when the current focus state segment is a stable focus state segment and its duration reaches a preset increase duration threshold; decreasing the training load when the current focus state segment is a deviated focus state segment and its duration reaches a preset regression duration threshold; maintaining the current training load and reducing the sensitivity of training feedback to changes in a single EEG time window when the current focus state segment is a fluctuating transition state segment; generating training evaluation results based on the proportion of stable focus state segments, the proportion of deviated focus state segments, the proportion of fluctuating transition state segments, the trend of behavioral representation value changes, and the number of training load adjustments; and determining the initial training configuration for the next training cycle based on the training evaluation results.

[0014] Secondly, this application provides a focus training system that combines electroencephalogram (EEG) and behavioral data, the system comprising: The data synchronization and alignment module is used to synchronously collect EEG and behavioral data of training subjects during the focus training task, and divide the EEG data into continuous EEG time windows based on a unified time reference, mapping the behavioral data to the corresponding EEG time windows; the EEG feature calculation module is used to preprocess the EEG data within each EEG time window, extract EEG focus features, and calculate the EEG reliability index of the corresponding EEG time window based on artifact interference to obtain the EEG candidate focus value for each EEG time window; the consistency verification module is used to extract behavioral focus features from the behavioral data within each EEG time window, generate corresponding behavioral representation values, and obtain the consistency verification result based on the deviation relationship between the behavioral representation values ​​and the EEG candidate focus values; the focus value gating correction module is used to adjust the focus value based on the EEG reliability index. The system uses the consistency verification results to perform gating correction on the candidate EEG focus values, obtaining the fused focus value for the corresponding EEG time window; the focus state segmentation module is used to determine the continuity of focus states during training based on the fused focus values ​​of multiple consecutive EEG time windows, dividing focus state segments; the training load adaptation module is used to adaptively adjust the training load of the focus training task according to the type and duration of the current focus state segment, wherein the training load includes at least one of task difficulty, stimulus rhythm, and feedback sensitivity; the evaluation and configuration determination module is used to generate training evaluation results after the training task ends, based on the distribution results of each focus state segment during training and the changes in corresponding behavioral data, and determine the initial training configuration for the next training cycle based on the training evaluation results. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram illustrating an application scenario provided in one embodiment of this application; Figure 2 A flowchart illustrating a focus training method combining EEG and behavioral data, provided as an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a focus training system that combines EEG and behavioral data, provided as an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0019] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0020] Figure 1 This is a schematic diagram illustrating an application scenario provided by this application. In the process of children's attention training, applying the method provided in this application can improve the stability of attention state recognition in children's scenarios, reduce the misjudgment rate of a single EEG channel, achieve real-time closed-loop control of the training process, and enhance the continuity of training and children's interactive experience.

[0021] Specifically, the method of this application is applied to any server that communicates with the brain-computer interface smart headband and obtains EEG data and behavioral data provided by the brain-computer interface smart headband through the server.

[0022] For specific implementation details, please refer to the following examples.

[0023] Example 1: Figure 2 This is a flowchart illustrating a focus training method combining EEG and behavioral data, provided as an embodiment of this application. The method of this embodiment can be applied to servers in the above-described scenarios. Figure 2 As shown, the method includes: Existing brain-computer interface (BCI) attention training systems collect EEG signals via scalp electrodes, extracting features such as theta waves, alpha waves, beta waves, sensorimotor rhythms (SMR), and their ratios, outputting attention scores, and providing neurofeedback training through animation, scoring, game character movements, or audiovisual feedback. However, in children's attention training scenarios, there are technical problems such as poor stability of EEG data, high risk of misjudgment due to relying solely on EEG to determine attention status, lack of reliable basis for adjusting training parameters, and rapid fluctuations and short durations of children's attention states.

[0024] Step 1: During the focus training task, collect the EEG data and behavioral data of the training subjects simultaneously, and divide the EEG data into continuous EEG time windows based on a unified time reference, and map the behavioral data to the corresponding EEG time windows. The unified time reference can refer to a timestamp system shared by EEG acquisition devices and behavioral interaction terminals driven by the same high-precision clock source, with a time resolution of not less than 1 millisecond; the EEG time window is a time segment of fixed length, such as 1 second, 2 seconds, or 3 seconds; behavioral data mapping can refer to normalizing the occurrence timestamps of behavioral events (such as key responses, touch operations, target selection, and task switching) to the start time of the corresponding EEG time window, and aggregating all behavioral events within the time window into a structured behavioral record; this mapping process ensures that EEG features and behavioral performance have a strict temporal causal correspondence, providing basic alignment support for subsequent cross-modal fusion.

[0025] This application can, for example, synchronize EEG analog-to-digital conversion and behavioral event tagging based on a unified hardware clock to achieve millisecond-level time alignment; it can also allocate logical timestamps based on a global time service at the software layer and complete the time window assignment matching between EEG segments and behavioral events in the post-processing stage; furthermore, this application can also employ a timestamp interpolation compensation mechanism to dynamically correct the time offset between the EEG sampling start point and the first behavioral event. This application obtains a strict correspondence between EEG and behavioral data in the time dimension based on any of the above methods, thereby ensuring the spatiotemporal consistency of the input data in subsequent fusion determination.

[0026] For example, in a jumping-type focus training task, the trainee wears a brain-computer interface headband and operates a touchscreen to complete a target jump response; the system continuously collects EEG signals at a sampling rate of 1000 Hz, and simultaneously records the timestamp of each touch operation in an event-driven manner; the EEG signals are divided into 2-second windows to form a continuous EEG time window sequence; for each 2-second window, all touch events occurring within its time range (including jump trigger, left and right movement, and target arrival confirmation) are retrieved, and these events are organized into a behavioral data set corresponding to that window; thereby completing the window-by-window mapping of EEG segments and behavioral actions.

[0027] Step 2: Preprocess the EEG data within each EEG time window, extract EEG focus features, and calculate the EEG reliability index for the corresponding EEG time window based on artifact interference to obtain the candidate EEG focus value for each EEG time window. The preprocessing includes at least one of bandpass filtering, power frequency suppression, and baseline correction, used to retain the main frequency band signals related to the focus state and suppress environmental noise; the EEG focus features can be at least one of the following: theta wave power, alpha wave power, beta wave power, theta / β ratio, sensorimotor rhythm (SMR) energy, frequency band energy change rate, or EEG stability index, which characterizes the focus tendency of the training subject's neural activity within the current time window; the EEG reliability index (Qe) can be a quantitative value characterizing the degree of interference from eye movement, electromyography, or motion artifacts in the current EEG time window, with a value range of [0,1], and a higher value indicates that the EEG signal quality within the time window is more reliable; the EEG candidate focus value (Ae) can be a scalar value obtained by normalization and weighted synthesis based on the EEG focus features, used to initially characterize the focus level corresponding to the time window.

[0028] This application, for example, may involve removing DC drift and high-frequency noise through bandpass filtering (e.g., 1–40 Hz), extracting the θ / β ratio and SMR energy as EEG focus features, and calculating the EEG reliability index by combining the proportions of low-frequency and high-frequency interference components. Alternatively, this application may first perform 50 Hz power frequency notch suppression, then perform baseline correction on the filtered signal, and subsequently extract alpha wave power and frequency band energy change rate to generate candidate EEG focus values. Furthermore, this application may employ a sliding window overlap method to segment and preprocess the original EEG signal to improve the temporal continuity of feature extraction. Based on any of the above methods, this application obtains candidate EEG focus values ​​corresponding to each EEG time window, ensuring that they reflect both the neurophysiological focus state and are constrained by the quality of the EEG signal.

[0029] For example, in parkour sustained attention training, the system sequentially performs 0.5–45 Hz bandpass filtering, 50 Hz power frequency suppression, and detrended baseline correction for each 2-second EEG time window; then calculates the power ratio of theta waves (4–8 Hz) to beta waves (13–30 Hz) within the window, and extracts the average energy of SMR rhythm (12–15 Hz) in the central channel C3 / C4; after normalizing the two features to the [0,1] interval, they are weighted and summed to obtain the initial EEG focus score; at the same time, the proportion of low-frequency component energy (0.1–1 Hz, representing eye movement drift), the proportion of high-frequency component energy (30–100 Hz, representing electromyography interference), and the proportion of amplitude abrupt change points (representing motion artifacts) within the window are statistically analyzed, and after normalization and weighted synthesis, the EEG reliability index Qe=0.73 is obtained; finally, the initial score is multiplied by Qe to output the EEG candidate focus value Ae=0.61.

[0030] Step 3: Extract behavioral focus features from the behavioral data within each EEG time window, generate corresponding behavioral representation values, and obtain consistency verification results based on the deviation relationship between the behavioral representation values ​​and the candidate focus values ​​of the EEG. Among them, behavioral focus characteristics can refer to at least one of reaction time, answer accuracy, missed answer rate, false touch rate, continuous operation stability, or task switching delay, which reflects the training subject's explicit behavioral focus performance within the current time window; behavioral representation value (Ab) can refer to the scalar value obtained by converting each behavioral focus characteristic into a unidirectional evaluation quantity, normalizing it, and weighting it, and its numerical range is consistent with the EEG candidate focus value; consistency verification result can refer to the binary or hierarchical judgment result that characterizes whether the EEG candidate focus value and the behavioral representation value synergistically reflect the same focus state, such as consistent, slightly deviating, or seriously inconsistent.

[0031] This application may, for example, generate a co-directional evaluation metric based on the reciprocal of reaction time and accuracy, and then synthesize the behavioral representation value with equal weights after normalization of the two. Alternatively, this application may use negative values ​​for the missed response rate and false touch rate as inverse indicators, transforming them into positive evaluation metrics through linear transformation before participating in the synthesis. Furthermore, this application may introduce a task-type adaptive weighting strategy, such as increasing the weight of accuracy in inhibitory control tasks and increasing the weight of continuous operation stability in sustained attention tasks. This application obtains the behavioral representation value based on any of the above methods and uses the absolute deviation Δ=|Ae-Ab| between it and the candidate EEG attention value as the consistency verification basis.

[0032] For example, in a color sprint training task, the behavioral data corresponding to a 2-second EEG time window includes 3 valid responses with reaction times of 420 ms, 480 ms, and 510 ms, respectively, with an accuracy rate of 100% and no missed answers or false touches. The system takes the average of the reciprocals of the reaction times (approximately 2.38, 2.08, and 1.96 Hz) to 2.14 Hz, normalizes it to [0,1] to obtain 0.82, and directly normalizes the accuracy rate to obtain 1.0. The weighted average of the two results in a behavioral representation value Ab=0.91. The concurrent EEG candidate focus value Ae=0.61, so the deviation Δ=0.30. If the preset deviation threshold is 0.25, then the time window is determined to be an EEG-behavioral inconsistency time window.

[0033] Step 4: Based on the EEG credibility index and consistency verification results, gating correction is performed on the candidate EEG focus values ​​to obtain the fused focus values ​​for the corresponding EEG time windows; Among them, gating correction can refer to applying a decay coefficient to the candidate focus value of EEG based on the EEG credibility index, or constraining the synthesis of the candidate focus value of EEG and the behavioral representation value based on the consistency verification result. Its purpose is to suppress the dominant influence of low-quality EEG segments and enhance the discourse power of behavioral data when the verification is inconsistent. The fusion focus value (Af) can refer to the final focus representation quantity generated after gating correction and used to drive subsequent focus state judgment and training regulation.

[0034] This application, for example, may linearly decay the candidate EEG focus value by multiplying it by Qe when the EEG confidence index Qe is below a first threshold (e.g., 0.6); it may also synthesize the fused focus value from the previous EEG time window and the current behavioral representation value by performing a moving average when the consistency check result is severely inconsistent; furthermore, this application may set a gating coefficient w = Qe × (1 - k × Δ), where k is a bias sensitivity factor, and use this coefficient to weight and correct Ae, and then superimpose Ab for conservative fusion. This application obtains the fused focus value based on any of the above methods, thus possessing both EEG physiological basis and behavioral verification support.

[0035] For example, in the aforementioned color sprint training, Ae=0.61, Ab=0.91, Δ=0.30, Qe=0.73; because Δ>0.25, a consistency check anomaly is triggered; the system sets the gating coefficient w=Qe×(1-2×Δ)=0.73×(1-0.6)=0.292; the corrected EEG focus value=0.61×0.292≈0.18; then it is fused with Ab=0.91 and the previous window focus value Af. ﹣1 We perform a weighted synthesis with a weight of 0.75 (weights of 0.3, 0.5, and 0.2 respectively), resulting in Af = 0.18 × 0.3 + 0.91 × 0.5 + 0.75 × 0.2 = 0.054 + 0.455 + 0.15 = 0.659.

[0036] Step 5: Based on the fusion focus values ​​of multiple consecutive EEG time windows, the continuity of focus during the training process is determined, and focus state segments are divided. Among them, multiple consecutive EEG time windows can refer to N adjacent time windows selected in a sliding window manner (e.g., N=5 or N=10), with a coverage duration of N×T (T is the length of a single time window); the focus state segment can refer to at least one of the stable focus state segment, the deviated focus state segment, or the fluctuating transition state segment, which are divided according to the statistical characteristics of the fused focus value (such as mean, fluctuation amplitude, and duration); the division results are used to characterize the phased evolution law of focus state during training.

[0037] This application may, for example, use a sliding window of length 5 to perform mean and standard deviation statistics on the fused focus value sequence, and identify stable focus state segments based on whether the mean is higher than a first state threshold (e.g., 0.7) and the standard deviation is lower than a first fluctuation threshold (e.g., 0.1). Alternatively, this application may determine a deviation from a focus state segment when the mean of the fused focus value is lower than a second state threshold (e.g., 0.4) for 10 consecutive time windows. Furthermore, this application may combine duration statistics and fluctuation amplitude statistics for joint determination to avoid misjudgments caused by short-term disturbances. This application obtains focus state segment division results based on any of the above methods, providing a state basis for subsequent training load adjustments.

[0038] For example, in a study room setting, the system collects and integrates focus values ​​in a 2-second time window, using 10 windows (i.e., 20 seconds) as the sliding statistical unit. If the mean value of Af is 0.78 and the standard deviation is 0.06 within a consecutive 20 seconds, it is classified as a stable focus state segment. If the mean value of Af drops to 0.35 and the standard deviation rises to 0.18 within a subsequent consecutive 20 seconds, it is classified as a deviated focus state segment. If Af fluctuates frequently between 0.52 and 0.63 and the mean value is 0.58, it is classified as a fluctuating transition state segment.

[0039] Step Six: Based on the type and duration of the current focus state, adaptively adjust the training load of the focus training task. The training load includes at least one of the following: task difficulty, stimulus rhythm, and feedback sensitivity. Among them, task difficulty can refer to at least one of the following: target stimulus presentation duration, number of distractors, pass threshold, or character movement speed mapping slope; stimulus rhythm can refer to at least one of the following: target appearance interval, cue audio rate, or screen refresh rate; feedback sensitivity can refer to at least one of the following: visual feedback intensity change rate, voice cue trigger delay, or vibration feedback response threshold; adaptive adjustment can refer to dynamically modifying at least one of the above parameters according to the type and duration of the focus state segment in order to maintain a balance between training challenge and feasibility.

[0040] This application, for example, can shorten the presentation duration of the target stimulus and increase the density of distractors when the duration of the stable focus state reaches a preset threshold for enhancement duration (e.g., 60 seconds); it can also extend the target appearance interval and reduce the intensity of visual feedback when the duration of the deviance from the focus state reaches a preset threshold for regression duration (e.g., 45 seconds); further, it can maintain the current task difficulty unchanged while keeping the feedback sensitivity from real-time response to a hysteresis 2-window smooth response when the fluctuating transition state persists. This application achieves closed-loop control of training load based on any of the above methods, improving training adaptability and sustainability.

[0041] For example, in focused jumping training, the system identifies a 72-second period of stable focus; then automatically increases the target jump height mapping slope from 1.0 to 1.3, while increasing the frequency of interference obstacles from once every 15 seconds to once every 10 seconds; and in the subsequent 50-second period of defocus, the system extends the character's landing buffer time from 200 ms to 400 ms, and disables non-critical voice prompts, retaining only basic vibration feedback.

[0042] Step 7: After the training task is completed, generate training evaluation results based on the distribution of each focus state segment during the training process and the corresponding changes in behavioral data, and determine the initial training configuration for the next training cycle based on the training evaluation results.

[0043] The distribution results of the focus state segments can refer to the proportion of stable focus state segments, the proportion of deviating focus state segments, the proportion of fluctuating transition state segments, and their time series distribution characteristics; the change results of behavioral data can refer to at least one of the following: reaction time change trend, accuracy change curve, false touch rate decrease rate, or task switching latency reduction magnitude; the training evaluation results can refer to the ability analysis report generated by comprehensively considering the above multi-dimensional indicators, the content of which includes focus stability rating, behavioral response maturity rating, and training adaptability rating; the initial training configuration for the next training cycle can refer to at least one of the following: task mode, initial difficulty level, stimulus rhythm baseline value, or feedback sensitivity default value.

[0044] This application can, for example, generate an assessment conclusion of good focus stability and mature behavioral control based on the result that the proportion of stable focus states is ≥60% and the standard deviation of reaction time decreases by ≥30%, and set the initial difficulty of the next cycle to the current level +1. Alternatively, when the proportion of deviations from a focus state exceeds 40% and the accidental touch rate shows an upward trend, this application can generate an assessment conclusion that basic attention maintenance ability needs to be strengthened, and recommend using a basic training mode with smaller movement amplitude and slower feedback rhythm. Furthermore, this application can also construct a user profile based on historical training data, update the current assessment results into the profile, and use this to support personalized recommendations for course training, phased programs, or self-study room scenarios. This application completes the training loop based on any of the above methods, realizing the continuous evolution of individualized training paths.

[0045] Example 2: In one embodiment, this application also provides step S2 for preprocessing EEG data within each EEG time window and calculating EEG reliability indicators, including: Step 1: Preprocess the EEG data within each EEG time window, including sequentially performing bandpass filtering, power line suppression, baseline correction, and artifact detection to obtain preprocessed EEG data; Bandpass filtering can refer to a linear filtering operation that retains signal components within a set frequency range while attenuating components outside the range. In this embodiment, bandpass filtering can restrict EEG data to the 1 Hz–50 Hz frequency band to retain the main frequency bands related to cognitive states, such as theta waves (4–7 Hz), alpha waves (8–13 Hz), beta waves (14–30 Hz), and some gamma waves (30–50 Hz), while filtering out ultra-low frequency drift (<1 Hz) and high frequency noise (>50 Hz). Power frequency suppression can refer to eliminating 50 Hz or 60 Hz noise introduced by the AC power supply system. Operations involving Hz and its harmonic interference; in this embodiment, power frequency suppression can be an adaptive notch filter or a discrete Fourier transform domain frequency point zeroing method, used to reduce the energy amplitude of the power frequency and its adjacent frequency bands; baseline correction can refer to the operation of eliminating slowly changing DC offset or trend terms; in this embodiment, baseline correction can be a polynomial fitting removal method or a wavelet decomposition low-frequency component elimination method, used to suppress long-period drift caused by unstable electrode contact or slow changes in scalp impedance; artifact detection can refer to the operation of identifying and marking EEG segments interfered with by non-neurogenic physiological activities; in this embodiment, artifact detection can be a detection method based on amplitude threshold, differential gradient, variance mutation, or independent component analysis (ICA) residual energy, used to locate abnormal sampling intervals caused by eye movement, electromyography, motion, or electrode loosening; This application, for example, may remove ultra-low frequency and ultra-high frequency noise using bandpass filtering, eliminate 50 Hz power interference using power frequency suppression, eliminate slow drift trends using baseline correction, and finally identify and label abnormal EEG segments using artifact detection. Alternatively, this application may employ a cascaded digital filter structure to simultaneously perform bandpass filtering and power frequency suppression, then achieve baseline correction through sliding window local mean estimation, and combine amplitude-gradient dual threshold joint criteria to complete artifact detection. Furthermore, this application may dynamically adjust the baseline correction intensity based on real-time updated interchannel impedance monitoring results, and adaptively adjust the current artifact detection threshold based on the historical artifact distribution of the preceding time window. This application obtains preprocessed EEG data that meets the requirements for subsequent feature extraction based on any of the above methods.

[0046] Step 2: Based on the preprocessed EEG data, extract at least two of the following: frequency band energy features, frequency band ratio features, and time window fluctuation features to generate candidate EEG focus values; The frequency band energy characteristic can refer to the integral value of the signal power spectral density within a unit time window calculated within a preset frequency band; for example, it can be theta wave energy (total energy in the 4–7 Hz band), alpha wave energy (total energy in the 8–13 Hz band), beta wave energy (total energy in the 14–30 Hz band), or SMR rhythm energy (12–15 Hz). The total energy of the Hz band is used in this embodiment to characterize the intensity of neural activity in the corresponding frequency band, serving as the basic input for decoding the state of focus. The frequency band ratio feature can refer to the ratio relationship between two different frequency band energies; for example, it can be the θ / β ratio (the ratio of θ wave energy to β wave energy), the α / β ratio (the ratio of α wave energy to β wave energy), or the β / α ratio (the ratio of β wave energy to α wave energy). In this embodiment, it is used to weaken the influence of individual baseline differences, enhance cross-subject comparability, and reflect the cortical excitation / inhibition balance. The time window fluctuation feature can refer to the degree of dispersion of signal amplitude, power, or frequency band energy over time within the current EEG time window; for example, it can be the standard deviation of α wave energy, the coefficient of variation of β wave energy, the absolute mean of the first-order difference of the sliding window of the θ / β ratio, or the time series entropy value of the variance of the full channel amplitude; in this embodiment, it is used to characterize the stability of EEG activity and assist in identifying short-term attention fluctuations or micro-arousal states. This application, for example, can generate candidate EEG focus values ​​by jointly modeling frequency band energy characteristics and frequency band ratio characteristics, where the θ / β ratio serves as the core feature and alpha wave energy as a supplementary modulatory factor. Alternatively, this application can generate candidate EEG focus values ​​by jointly modeling frequency band ratio characteristics and time window fluctuation characteristics, where the β / α ratio reflects the arousal level and its first-order difference absolute mean reflects instantaneous stability. Further, this application can generate candidate EEG focus values ​​by jointly modeling frequency band energy characteristics and time window fluctuation characteristics, where β wave energy characterizes the intensity of attentional resource input, and its standard deviation characterizes input stability. This application obtains candidate EEG focus values ​​representing the current EEG focus tendency within the time window based on any of the above methods.

[0047] Step 3: Calculate the proportion of low-frequency interference components within the current EEG time window to characterize the degree of low-frequency drift caused by eye movements or head movements; The proportion of low-frequency interference components can refer to the proportion of signal energy in the 0.1 Hz to 1 Hz frequency band to the total energy of the entire frequency band (e.g., 1 Hz to 50 Hz); this proportion can be the ratio of the power spectrum integral calculated in the low-frequency sub-band after bandpass filtering to the full-band integral; in this embodiment, it is used to quantify the degree of influence of slow displacement artifacts (such as head shaking, electrode slippage, breathing coupling) on ​​the signal quality of the current time window; This application may calculate the energy proportion in the 0.1–1 Hz frequency band using a power spectral density estimation method; it may also estimate the energy proportion based on the lowest frequency sub-band after wavelet packet decomposition; further, it may perform dynamic weighted correction based on the rate of change of low-frequency energy in adjacent time windows. This application obtains the proportion of low-frequency interference components characterizing the degree of low-frequency drift based on any of the above methods.

[0048] Step 4: Calculate the proportion of high-frequency interference components within the current EEG time window to characterize the degree of electromyographic interference; The proportion of high-frequency interference components can refer to the proportion of signal energy in the 30 Hz to 50 Hz frequency band to the total energy of the entire frequency band (e.g., 1 Hz to 50 Hz); this proportion can be the ratio of the power spectrum integral calculated in the high-frequency sub-band after bandpass filtering to the full-band integral; in this embodiment, it is used to quantify the degree of influence of electromyographic activity generated by the contraction of facial, neck or jaw muscles on the signal quality of the current time window; This application may calculate the energy proportion of 30–50 Hz based on the sum of squared amplitudes of the spectrum after Fast Fourier Transform (FFT); it may also estimate it based on the energy proportion of high-frequency intrinsic mode functions (IMFs) in the Hilbert-Huang Transform (HHT); further, it may combine the spatial distribution characteristics of the channels to weighted amplify the high-frequency energy of the frontalis and temporalis muscle regions before participating in the proportion calculation. This application obtains the proportion of high-frequency interference components characterizing the degree of electromyographic interference based on any of the above methods.

[0049] Step 5: Calculate at least one of the following within the current EEG time window: the proportion of amplitude abrupt change points, the proportion of saturation sampling points, and the proportion of abnormal fluctuations in adjacent sampling points, to characterize the degree of motion artifacts; The amplitude abrupt change ratio can be the ratio of the number of sampling points whose amplitude jump exceeds a preset multiple threshold within a unit time window to the total number of sampling points; this threshold can be 5 times the amplitude standard deviation within the time window; in this embodiment, it is used to reflect instantaneous voltage jumps caused by sudden limb movements or strong blinking; the saturation sampling point ratio can be the ratio of the number of sampling points that reach the maximum / minimum quantization value of the analog-to-digital converter within a unit time window to the total number of sampling points; in this embodiment, it is used to reflect clipping distortion caused by poor electrode contact or excessive signal amplification gain; the adjacent sampling fluctuation abnormal ratio can be the ratio of the number of point pairs whose absolute difference value of adjacent sampling points is greater than a preset fluctuation threshold within a unit time window to the total number of point pairs; this threshold can be 3 times the median of the absolute difference value within the time window; in this embodiment, it is used to reflect irregular fluctuations caused by high-frequency jitter, equipment micro-vibration, or electrode micro-motion; This application may, for example, calculate the degree of motion artifacts by jointly considering the proportion of amplitude abrupt change points and the proportion of saturated sampling points; it may also calculate the degree of motion artifacts by jointly considering the proportion of amplitude abrupt change points and the proportion of abnormal fluctuations in adjacent sampling points; further, it may calculate the maximum value of the three indicators separately and use it as the motion artifact characterization quantity. This application obtains the motion artifact characterization quantity representing the degree of motion artifacts based on any of the above methods.

[0050] Step 6: Normalize and weight the proportions of low-frequency interference components, high-frequency interference components, and motion artifact representations to obtain the EEG reliability index. Normalization can refer to mapping various interference indicators to a linear or nonlinear transformation within the [0, 1] interval. For example, it can be performing min-max normalization on each indicator, mapping the minimum value to 0 and the maximum value to 1; or it can be using a sigmoid function to compress it to an open interval of (0, 1). In this embodiment, it is used to unify interference indicators of different dimensions and orders of magnitude to ensure the rationality of weighted synthesis. Weighted synthesis can refer to assigning different weights to the three normalized indicators and then calculating a weighted sum. For example, it can be assigning a weight of 0.4 to the low-frequency interference component, a weight of 0.3 to the high-frequency interference component, and a weight of 0.3 to the action artifact representation; or it can be dynamically adjusting the weights according to the age group of the training subjects, such as increasing the weight of action artifacts to 0.5 for children. In this embodiment, it is used to generate a single scalar form of EEG credibility index, which serves as the quantitative basis for subsequent gating correction and fusion judgment. This application may generate an EEG reliability index based on a min-max normalization and fixed-weight weighted synthesis method; alternatively, it may generate an EEG reliability index based on a sigmoid normalization and task-type adaptive weight allocation method; furthermore, it may dynamically update the normalization upper and lower limits based on the historical window interference distribution and select a differentiated weight vector by combining the current time window artifact type combination. This application obtains an EEG reliability index that comprehensively represents the quality level of the current EEG time window based on any of the above methods.

[0051] For example, in a focus-jumping training session, the subject is an 8-year-old child wearing a brain-computer interface (BCI) headband to collect EEG signals. When the child makes a slight head-shaking motion, the energy proportion in the 0.1–1 Hz frequency band rises to 18% of the total frequency band, triggering an increase in the proportion of low-frequency interference components. When the child frowns or chews, the energy proportion in the 30–50 Hz frequency band rises to 12%, triggering an increase in the proportion of high-frequency interference components. When the child quickly taps the screen, causing a micro-vibration of the head, the abnormal fluctuation ratio of adjacent samples reaches 9.3%. After min-max normalization, the three indicators are 0.82, 0.71, and 0.65, respectively. They are weighted by weights of 0.4, 0.3, and 0.3 to obtain the EEG reliability index Qe = 0.75. This value falls within the time window interval to be corrected as defined in Example 3, and the corresponding candidate focus value of the EEG enters the gating correction process to avoid direct participation in the consistency verification.

[0052] Example 3: In one possible implementation, this application also provides a method for dividing the EEG time window into a high-confidence time window, a time window to be corrected, and a low-confidence time window based on the EEG confidence index. Specifically, candidate focus values ​​corresponding to the high-confidence time window directly enter the consistency check, candidate focus values ​​corresponding to the time window to be corrected enter the gating correction, and candidate focus values ​​corresponding to the low-confidence time window are suppressed and directly participate in the focus state determination of the current EEG time window. Step 1: Divide the EEG time window into a high-confidence time window, a time window to be corrected, and a low-confidence time window based on the EEG confidence index; Among them, the EEG reliability index is a dimensionless comprehensive score that characterizes the degree of interference from artifacts in the current EEG time window, and its value ranges from 0 to 1; the EEG reliability index can refer to the quantitative result obtained by normalization and weighted synthesis based on the proportion of low-frequency interference components, the proportion of high-frequency interference components, and the amount of motion artifact representation. In this embodiment, the EEG reliability index is used to make quality grading decisions for each EEG time window. Its role is to provide a basis for subsequent differentiated data processing paths: when the EEG reliability index is higher than the preset high reliability threshold, it indicates that the EEG signal quality within the time window is good, the artifact interference is controllable, and it has a reliable basis for directly participating in the focus state fusion determination; when the EEG reliability index is between the preset high reliability threshold and the preset low reliability threshold, it indicates that the time window has a moderate degree of artifact interference and needs to be constrained by a gating mechanism before it can participate in fusion; when the EEG reliability index is lower than the preset low reliability threshold, it indicates that the time window is strongly contaminated by artifacts, its candidate focus values ​​are unreliable, and it should be excluded from the current focus state determination process.

[0053] This application, for example, classifies the current EEG time window based on a dual-threshold interval division rule consisting of an EEG reliability index and preset high-reliability and low-reliability thresholds. It also dynamically adjusts the threshold boundaries to adapt to individualized artifact features based on the changing trend of the EEG reliability index across multiple consecutive time windows, combined with the interval in which the current instantaneous value falls. Furthermore, this application inputs the EEG reliability index into a preset classification function, outputting three categories: high reliability, needing correction, or low reliability. This application obtains the reliability level classification result for the EEG time window based on any of the above methods, thereby supporting data of different quality levels to enter the corresponding processing channels.

[0054] For example, this application sets a preset high confidence threshold of 0.75 and a preset low confidence threshold of 0.35; when the EEG confidence index of a certain EEG time window is 0.82, it is determined to be a high confidence time window; when its value is 0.53, it is determined to be a time window to be corrected; when its value is 0.21, it is determined to be a low confidence time window; the determination result directly determines whether the candidate EEG focus value in the time window participates in the consistency verification, whether it enters the gating correction process, or whether it is suppressed and does not participate in the current focus state determination.

[0055] Step 2: The EEG candidate focus values ​​corresponding to the high-confidence time window directly enter the consistency check; Among them, the high-confidence time window can refer to the EEG time window where the EEG confidence index is higher than the preset high-confidence threshold; the high-confidence time window is an effective segment retained after bandpass filtering, power frequency suppression, baseline correction and artifact detection, and its EEG candidate focus value is obtained by weighted synthesis of at least two of the frequency band energy characteristics, frequency band ratio characteristics and time window fluctuation characteristics. In this embodiment, the EEG candidate focus value corresponding to the high-confidence time window is directly used as one of the inputs of the consistency verification module without additional correction or delay. Its function is to ensure that high-quality EEG information can participate in the EEG-behavior joint judgment in a timely and complete manner, thereby improving the response speed and accuracy of focus state recognition.

[0056] Step 3: The candidate attention values ​​of the EEG corresponding to the time window to be corrected are entered into the gating correction; Among them, the time window to be corrected can refer to the EEG time window where the EEG confidence index is between the preset high confidence threshold and the preset low confidence threshold; the time window to be corrected is a time segment with mild eye movement drift or local electromyographic interference but not saturation level. Although its EEG candidate focus value has a certain risk of deviation, it still has reference value. In this embodiment, the EEG candidate focus value corresponding to the time window to be corrected does not directly participate in the consistency verification, but is sent to the gate control correction module to undergo at least two gate control operations, such as attenuation processing based on EEG credibility index, constraint synthesis of focus value fused with the previous time window, or replacement with behavioral representation value in low credibility cases. Its function is to apply controllable intervention to medium quality data, suppress abnormal jumps while retaining potential information, and prevent them from disrupting the continuity judgment of focus state.

[0057] Step 4: The candidate focus values ​​of EEG corresponding to the low confidence time window are suppressed and directly participate in the focus state determination of the current EEG time window; Among them, the low confidence time window can refer to the EEG confidence index below the preset low confidence threshold; the low confidence time window is a severely contaminated segment containing significant amplitude mutation points, high proportion of saturated sampling points or adjacent sampling fluctuation abnormal proportion exceeding the limit, and its EEG candidate focus value can no longer reflect the true neural activity state. In this embodiment, the candidate focus value of EEG corresponding to the low confidence time window is marked as invalid and removed from the focus state determination process of the current EEG time window. It does not participate in consistency verification, gating correction, or fusion focus value generation. Its function is to completely block the pollution path of strong artifact data on system judgment, avoid the cascading effect of misjudgment caused by a single inferior segment, and thus enhance the robustness and stability of the system in children's high interference training scenarios.

[0058] Example 4: In one embodiment, this application further provides that in step S3, the behavioral representation value is determined based on the behavioral data corresponding to the current EEG time window, and the behavioral data includes at least two of the following: reaction time, answer accuracy rate, missed answer rate, false touch rate, continuous operation stability, and task switching latency; generating the behavioral representation value includes: Step 1: Behavioral data should include at least two of the following: reaction time, accuracy of answer, missed answer rate, accidental touch rate, continuous operation stability, and task switching latency; Among these, reaction time refers to the time interval between the presentation of a stimulus and the completion of an effective response, used to characterize the speed of attention arousal and the state of response readiness; the correct response rate is the proportion of correct responses made by the user within a specified time to the total number of effective responses, used to characterize the accuracy of attention selection; the missed response rate is the proportion of responses that are not triggered in the response event, used to characterize the lack of attention maintenance ability; the false touch rate is the proportion of incorrect responses (such as premature response, incorrect target selection) to the total number of responses, used to characterize the ability of inhibitory control; continuous operation stability is a measure of the stability of operation parameters (such as the standard deviation of click position offset, the coefficient of variation of key interval, and the rate of change of curvature of sliding trajectory) within a unit time window, used to characterize the persistence of attention and the coordination of motor control; and the task switching latency is the time increment required for the user to switch responses between different subtasks, used to characterize the flexibility of attention and the efficiency of cognitive conversion.

[0059] In this embodiment, the above six types of behavioral data are all independent technical features, and their selection and combination do not constitute a limitation on the scope of protection; when any two or more are used at the same time, the limitation of the technical feature is satisfied.

[0060] Among them, reaction time, missed answer rate, and accidental touch rate are negative indicators, and their increased values ​​reflect a decrease in concentration; answer accuracy, continuous operation stability, and task switching delay (after directional correction) are positive indicators, and their increased values ​​reflect an increase in concentration; the differences in these directions are unified through subsequent conversion of the same-direction evaluation quantities.

[0061] Step 2: Convert the behavioral data into symmetrical evaluation metrics that represent the degree of focus; Among them, the same-direction evaluation quantity is a dimensionless value that maps the original behavioral data to the same monotonic direction (i.e., the larger the value, the higher the level of focus); this transformation does not change the original semantics carried by each behavioral data, but only standardizes and adjusts the relationship between its numerical direction and scale.

[0062] This application, for example, generates a unidirectional evaluation quantity based on the reciprocal relationship of the reaction time, i.e. Formal expression of the focus-promoting effect; this application may also set a threshold range based on the empirical distribution of reaction time, mapping the reaction time to the [0,1] range, where the shorter the reaction time, the higher the mapping value; further, this application may also convert the current reaction time into a percentile rank value as a co-evaluation quantity based on the historical training group reaction time quantile.

[0063] This application may, for example, directly use the original proportion of the correct answer rate as the same-direction evaluation metric; this application may also use the complement of the missed answer rate (i.e., 1 – missed answer rate) as the same-direction evaluation metric; this application may also use the complement of the false touch rate as the same-direction evaluation metric; furthermore, this application may use the continuous operation stability transformed by the inverse of variance or the stability exponential function as the same-direction evaluation metric; this application may also use the task switching delay after reverse mapping (such as the maximum allowable delay minus the actual delay) as the same-direction evaluation metric.

[0064] This application obtains a co-directional evaluation quantity representing the degree of focus based on any of the above methods, so that behavioral indicators with different physical meanings, different dimensions, and different directions can be compared and added at the semantic level of focus.

[0065] Step 3: Normalize the converted evaluation quantities. Normalization is an operation that maps all unidirectional evaluation quantities to a uniform numerical range (e.g., [0,1] or [–1,1]) to eliminate systematic biases caused by acquisition conditions, device response delays, or individual baseline differences.

[0066] This application, for example, employs a range normalization method, that is, for each unidirectional evaluation quantity... calculate ,in and These are the minimum and maximum values ​​of the metric within the current training session or the historical sliding window, respectively; this application may also employ the Z-score standardization method, i.e., calculating... ,in and These are the mean and standard deviation of the indicator over the reference period, respectively. Furthermore, this application may also use the quantile normalization method to map each unidirectional evaluation quantity to its cumulative probability value in the historical distribution.

[0067] This application obtains the normalized same-direction evaluation quantity based on any of the above methods, ensuring that each indicator has a comparable weight basis in the subsequent weighted synthesis.

[0068] Step 4: Weight the normalized positive evaluation values ​​to obtain the behavioral representation value of the current EEG time window; The weighted synthesis process involves dynamically configuring weight coefficients based on task type, training stage, or individual user characteristics, and performing a linear weighted summation operation on the normalized evaluation quantities.

[0069] This application, for example, sets weight allocation based on click-based training tasks: reaction time weight is 0.35, answer accuracy weight is 0.30, false touch rate weight is 0.20, and missed answer rate weight is 0.15, synthesizing the behavioral representation value; this application can also set weight allocation based on continuous control tasks: continuous operation stability weight is 0.40, reaction time weight is 0.25, answer accuracy weight is 0.20, and task switching latency weight is 0.15; furthermore, this application can also dynamically adjust the weights based on the user's historical performance, for example, gradually increasing the weight coefficient of users whose continuous operation stability performance is consistently better than the group average.

[0070] For example, in focused jumping training, the behavioral data corresponding to the current EEG time window includes reaction time (280 ms) and answer accuracy (92%). After inverse mapping and complement transformation, the same-direction evaluation values ​​are 3.57 (unit: 1 / s) and 0.92, respectively. After range normalization (the range of the same-direction evaluation value of reaction time within the reference window is [2.0, 5.0], and the range of the same-direction evaluation value of accuracy is [0.75, 0.98]), the normalized values ​​are 0.63 and 0.76, respectively. The behavioral representation value of the current EEG time window is obtained by weighting and synthesizing with weights of 0.45 and 0.55.

[0071] For example, in parkour continuous attention training, the behavioral data corresponding to the current EEG time window includes continuous operation stability (trajectory deviation standard deviation of 12.3 pixels) and task switching latency (average of 310 ms). After stability index transformation and inverse mapping, the same-direction evaluation values ​​are 0.81 and 0.69, respectively. After Z-score standardization (the mean and standard deviation of the reference time period are known), the normalized values ​​are 0.74 and 0.62, respectively. The behavioral representation value of the current EEG time window is obtained by weighted synthesis with weights of 0.60 and 0.40, respectively.

[0072] Example 5: In an optional embodiment, this application further provides a consistency check in step S3, including: Step 1: Calculate the deviation between the behavioral representation value and the candidate focus value of the current EEG time window; The deviation value can refer to the degree of difference between the behavioral representation value and the candidate focus value of the EEG in the numerical domain. It can be any one of the absolute difference, normalized difference, or weighted difference. Semantically, the deviation value represents the degree of consistency between the focus level obtained by EEG decoding within the current EEG time window and the user's actual behavioral performance. In this embodiment, the deviation value is used to quantify the degree of deviation between the EEG channel output and the behavioral channel output in the determination of focus state. The larger the value, the more inconsistent the two are in their assessment of the focus level in the same time period. This is more likely to indicate that the EEG data is affected by artifacts, the behavioral data has abnormal responses, or there is a phased mismatch in the cognitive load mapping relationship between the two.

[0073] Step 2: Compare the deviation value with the preset deviation threshold; The preset deviation threshold can be a critical value used to determine whether EEG-behavioral consistency is valid. It can be a fixed scalar, an interval boundary value, or a dynamic benchmark value. In the relevant technical field, this threshold is usually used as a general criterion for judging the reliable matching relationship between modalities in a multi-source signal fusion system. Its purpose is to provide an acceptable fault tolerance range for dual-channel output. In this embodiment, the preset deviation threshold is configured to be strongly correlated with the type of training task currently being run. For example, in a bouncing-type reaction task, the threshold is set to 0.15 because the behavioral response has strong timeliness and high certainty; in a study room-type sustained attention task, the threshold is set to 0.35 because behavioral indicators (such as page dwell time) have natural lag and ambiguity. This differentiated setting enables the system to adapt to the inherent uncertainty of the focus-behavior mapping relationship of different tasks, ensuring that the consistency check is neither too lenient, leading to missed detections, nor too strict, leading to misjudgments.

[0074] Step 3: When the deviation value exceeds the preset deviation threshold, the current EEG time window is determined to be the EEG-behavioral inconsistency time window; Among them, the EEG-behavior inconsistency time window can refer to a technical status indicator that, within the current EEG time window, the deviation between the behavioral representation value and the candidate EEG focus value has exceeded the reasonable fluctuation range allowed by the task type, thus indicating that the focus state judgment result within the time window cannot be directly accepted; this indicator serves as the trigger condition for the subsequent gating correction module in terms of function, and in terms of information semantics, it carries the control instructions to reduce the weight, delay, or replace the candidate EEG focus value. In this embodiment, the determination result does not directly negate the validity of the data in any channel, but instead activates a collaborative verification mechanism: if the inconsistent state continues to appear in multiple consecutive EEG time windows, the artifact review process (such as re-evaluating the EEG reliability index of the current time window) or behavioral data quality re-examination (such as checking whether the reaction time falls into the tail of the abnormal distribution) is further triggered; this mechanism ensures that the system maintains the data traceability capability while identifying anomalies, and avoids the interruption of the overall fusion logic due to misjudgment caused by a single deviation.

[0075] This application, for example, can determine whether the current EEG time window meets the EEG-behavioral consistency constraint condition by comparing the deviation value with a preset deviation threshold. Alternatively, it can generate inconsistency confidence levels (e.g., mild, moderate, severe inconsistency) based on the extent to which the deviation value exceeds the preset deviation threshold, and select a corresponding gating correction strategy accordingly. Furthermore, this application can dynamically update the preset deviation threshold value by combining the cognitive load characteristics of the current training task type, so that the threshold adaptively adjusts with changes in task difficulty gradient. Based on any of the above methods, this application obtains reliable identification of the EEG-behavioral consistency state, providing a clear and operable judgment basis for subsequent gating correction.

[0076] For example, in a parkour sustained attention training task, the system extracts the behavioral representation value of the current EEG time window as 0.62 (normalized to a maximum of 1.0), corresponding to an EEG candidate focus value of 0.85; the deviation between the two is |0.85 - 0.62| = 0.23; the preset deviation threshold for this task type is 0.20; since 0.23 > 0.20, the system determines that the current EEG time window is an EEG-behavior inconsistency time window and outputs a control signal to the gating correction module to attenuate the EEG candidate focus value; subsequently, the system calls the previously generated EEG credibility index (Qe = 0.41, belonging to the time window to be corrected), performs a weighted correction with an attenuation coefficient of 0.6, and obtains a corrected EEG focus value of 0.85 × 0.6 = 0.51, which then participates in the subsequent fusion focus value generation.

[0077] Example 6: In one optional embodiment, this application also provides gating correction for EEG-behavioral inconsistency time windows, including at least two of the following key implementation steps: Step 1: Attenuate the candidate EEG focus values ​​based on the EEG reliability index of the current EEG time window to obtain the corrected EEG focus values; Among them, the EEG reliability index is a quantitative indicator that characterizes the degree of interference from eye movement, electromyography, motion artifacts and poor electrode contact in the current EEG time window; the EEG reliability index can refer to a dimensionless value obtained by normalization and weighted synthesis based on the proportion of low-frequency interference components, the proportion of high-frequency interference components and the representation of motion artifacts, and its value range is 0 to 1. In this embodiment, the EEG reliability index is used to dynamically adjust the attenuation intensity of the EEG candidate focus value: when Qe is high, the attenuation coefficient approaches 1, and the EEG candidate focus value is basically preserved; when Qe decreases, the attenuation coefficient decreases according to a preset monotonically decreasing function relationship, so that the EEG candidate focus value is systematically compressed, thereby reducing its dominant role in subsequent fusion calculation. For example, this application determines the attenuation coefficient by looking up the EEG reliability index and the preset attenuation mapping table, and then multiplies the attenuation coefficient by the candidate EEG focus value to obtain the corrected EEG focus value. For example, this application calculates the attenuation coefficient by linear interpolation based on the EEG confidence index, and then performs scaling operations on the candidate EEG focus values ​​to obtain the corrected EEG focus values; furthermore, this application constructs a Sigmoid-type attenuation function based on the EEG confidence index to achieve gentle suppression of the low confidence interval and strong fidelity output of the high confidence interval, thereby obtaining the corrected EEG focus values. This application obtains a controllable decay of EEG candidate attention values ​​based on any of the above methods, so that it can still retain some effective information under the premise of EEG-behavior inconsistency, and avoid a cliff drop in fusion attention values ​​due to complete discarding.

[0078] Step 2: Combine the corrected EEG focus value with the fused focus value of the previous EEG time window to constrain the instantaneous jump variable of the current EEG time window; Among them, constraint synthesis can refer to the introduction of time continuity protection mechanisms such as weighted moving average, slope limitation or change threshold truncation between the corrected EEG focus value and the previous fused focus value; constraint synthesis is a smoothing operation to prevent sudden changes in focus state value. Its technical purpose is to avoid drastic fluctuations in controlled parameters such as character movement speed, jump height, and feedback intensity in the training interface, thereby improving the stability of children's user interaction and task immersion. In this embodiment, the constraint synthesis uses the corrected EEG focus value as the current input and the fused focus value Af(t-1) of the previous EEG time window as the historical reference benchmark. By setting the maximum allowable rate of change Δmax, the difference between the two is limited, and the limited result is superimposed on Af(t-1) to generate an intermediate fused value. For example, this application determines whether to enable amplitude limiting logic based on whether the absolute value of the difference between the modified EEG focus value and Af(t-1) exceeds Δmax: if it does not exceed the limit, the modified EEG focus value is directly used to participate in subsequent fusion; if it exceeds the limit, Af(t-1) and Δmax are combined in the direction to generate a constrained intermediate value. For example, this application performs a weighted average of the modified EEG focus value and Af(t-1) with variable weights, where the weights increase with Qe to enhance responsiveness under high confidence conditions and strengthen historical dependence under low confidence conditions. Furthermore, this application inputs the modified EEG focus value into a first-order inertial filter, whose time constant is dynamically adjusted with Qe, so that the output has adaptive response delay characteristics, thereby balancing stability and real-time performance. Based on any of the above methods, this application obtains the transient suppression ability for the output of the current EEG time window focus value, ensuring that the fused focus value shows a continuous and progressive change trend in the time dimension.

[0079] Step 3: When the current EEG time window is a low-confidence time window, use the behavioral characterization value to replace the EEG candidate focus value and participate in the generation of the fused focus value for the current EEG time window.

[0080] Among them, the low-confidence time window may refer to a time window in which the EEG confidence index Qe is lower than the first preset threshold. This first preset threshold is used to identify that the current EEG segment has been severely interfered by artifacts and does not have the effectiveness of independently decoding the focus state; the low-confidence time window state comes from the three-level division result defined in Embodiment 3, and its determination basis has been explained in the foregoing implementation manner, and here it is only naturally cited as an existing technical feature; The behavioral characterization value may refer to a co-directional evaluation quantity representing the user's focus degree, which is converted and weighted and synthesized according to at least two of the reaction time, answer accuracy rate, omission rate, mis-touch rate, continuous operation stability, and task switching delay; in this embodiment, the behavioral characterization value assumes a fallback channel function under the low-confidence time window: when the EEG signal quality is untrustworthy and the consistency check fails (that is, it has been determined as an EEG-behavior inconsistency time window), the system automatically switches to the behavioral data main drive mode, and directly uses the behavioral characterization value to replace the original EEG candidate focus value and enter the fusion calculation process; For example, in this application, when Qe < Qe_low_threshold and the consistency check result is inconsistent, the behavioral characterization value is assigned to the position of the EEG candidate focus value of the current EEG time window, and the subsequent fusion steps are executed according to the normal process; For example, in this application, when the above two conditions are met, the system activates the backup fusion path, skips the EEG attenuation and historical constraint links, and directly uses the behavioral characterization value as the only input source to participate in the generation of the fused focus value; further, in this application, while meeting the two conditions, the sensitivity of the subsequent training feedback is synchronously reduced, and a wearing state prompt is triggered to form a coordinated response of data degradation + interaction slow change + user reminder; Based on any of the above methods, this application obtains the ability to maintain the focus state in extreme interference scenarios, ensuring that the system can still output a reasonable focus determination result through the behavioral channel when the EEG channel fails.

[0081] For example, in a focus-jumping training exercise, when a child rapidly turns their head, causing Qe to plummet below 0.25, and their behavioral representation values ​​show prolonged reaction time and increased false touch rate, and the consistency check determines that the EEG-behavior is inconsistent, the system immediately initiates gating correction. First, based on Qe=0.25, the attenuation coefficient is found to be 0.3, and the original EEG candidate focus value of 0.85 is attenuated to 0.255. Then, it is constrained and synthesized with the previous fused focus value of 0.62. Since the difference between the two is 0.365, exceeding the preset Δmax=0.2, the output amplitude is limited to 0.82. Finally, because Qe is below the 0.3 threshold, the system abandons the amplitude limit result and directly uses the current behavioral representation value of 0.41 as the fusion input to generate the fused focus value of 0.41 for the current EEG time window, driving the character's jump height back to the low-to-medium range, while reducing the intensity of visual feedback to avoid operational loss of control caused by EEG overestimation.

[0082] Example 7: In one optional implementation, this application also provides a method for generating a fusion focus value, including: Step 1: Determine the gating coefficient corresponding to the current EEG time window based on the EEG reliability index; Among them, the gating coefficient is a dimensionless modulating factor that characterizes the contribution weight of the current EEG time window data quality to the fusion focus value; the gating coefficient can be a value generated by a monotonic mapping function based on the EEG credibility index, and its value range is [0,1]. In this embodiment, the gating coefficient is used to dynamically adjust the participation intensity of the EEG candidate focus value during the fusion process: when the EEG confidence index is in the high confidence range, the gating coefficient approaches 1, so that the EEG candidate focus value dominates the fusion; when the EEG confidence index is in the uncorrected or low confidence range, the gating coefficient decreases as the confidence decreases, so as to reduce the influence of the EEG candidate focus value on the current fused focus value. This application can determine the gating coefficient by linear mapping of the EEG confidence index, i.e., setting a first confidence threshold and a second confidence threshold, and linearly mapping the EEG confidence index interval between the two thresholds to [0.3, 0.9]. Alternatively, this application can determine the gating coefficient by looking up a table of segmented thresholds for the EEG confidence index, i.e., presetting three confidence intervals corresponding to gating coefficients of 0.2, 0.6, and 0.95 respectively. Furthermore, this application can determine the gating coefficient by transforming the EEG confidence index using a Sigmoid-type nonlinear function to enhance sensitivity to low and medium confidence intervals. Based on any of the above methods, this application obtains the ability to controllably attenuate candidate EEG focus values, thereby achieving adaptive adjustment of the EEG channel contribution.

[0083] Step 2: Correct the candidate EEG focus values ​​using gating coefficients to obtain the corrected EEG focus values; Among them, the corrected EEG focus value is an intermediate result of the EEG candidate focus value weighted by the gating coefficient; the corrected EEG focus value can be the product of the gating coefficient and the EEG candidate focus value; In this embodiment, the correction action achieves a credibility response to the original EEG decoding result: when the gating coefficient is low, even if the EEG candidate focus value is high, its corrected output is suppressed to avoid false high focus judgment caused by artifact interference; when the gating coefficient is high, the corrected EEG focus value is close to the original EEG candidate focus value, ensuring the information integrity of high-quality EEG segments. This application can, for example, obtain the corrected EEG focus value by multiplying the gating coefficient by the candidate EEG focus value; it can also obtain the corrected EEG focus value by truncating and scaling the candidate EEG focus value based on the gating coefficient, i.e., when the gating coefficient is lower than a preset lower limit, the corrected EEG focus value is set to a fixed low value; furthermore, this application can also obtain the corrected EEG focus value by applying an exponential decay factor to the candidate EEG focus value based on the gating coefficient, so as to enhance the suppression effect on abnormal jumps. This application obtains the reliability adaptation correction result of the candidate EEG focus value based on any of the above methods, providing EEG input items with controllable quality for subsequent multi-source fusion.

[0084] Step 3: Combine the corrected EEG focus value, the behavioral representation value of the current EEG time window, and the fusion focus value of the previous EEG time window to obtain the fusion focus value of the current EEG time window; Among them, the fusion focus value is a quantitative indicator of the focus state output at the end of the current EEG time window; the fusion focus value can be a weighted sum of the corrected EEG focus value, the behavioral representation value, and the fusion focus value of the previous EEG time window; In this embodiment, the combined action introduces the immediate verification capability of behavioral modality and the temporal inertia constraint of historical state: the behavioral representation value provides evidence of the level of focus directly related to the current task performance, and the fused focus value of the previous EEG time window serves as a state continuity anchor point to suppress output jitter caused by EEG mutations or instantaneous behavioral disturbances in a single time window. This application can, for example, obtain a fusion focus value based on a linear weighted sum of three inputs, where the weights of each input are dynamically adjusted jointly based on the EEG reliability index and the consistency check result. Alternatively, it can obtain the fusion focus value based on a moving average structure of the three inputs, where the corrected EEG focus value and the behavioral representation value are first weighted and synthesized, and then updated with a first-order low-pass filter compared to the previous fusion focus value. Furthermore, this application can obtain the fusion focus value based on a conditional combination of the three inputs, where when the consistency check result is inconsistent, the weight of the behavioral representation value is increased and the weight of the corrected EEG focus value is decreased, while a smoothing constraint coefficient from the previous fusion focus value is introduced. This application obtains a fusion focus value with quality responsiveness, behavioral interpretability, and temporal stability based on any of the above methods.

[0085] For example, in a specific training session involving focused jumping, the EEG reliability index for the current EEG time window is 0.42 (within the correction range), corresponding to a gating coefficient of 0.58; the candidate focus value is 0.81, so the corrected focus value is 0.47; the behavioral representation value for the current time window is 0.63 (derived from a combination of reduced reaction time and improved accuracy); the fusion focus value for the previous EEG time window is 0.59; the system uses a linear weighted combination method, assigning a weight of 0.4 to the corrected focus value, a weight of 0.45 to the behavioral representation value, and a weight of 0.15 to the previous fusion focus value, to calculate the current fusion focus value as 0.47×0.4 + 0.63×0.45 + 0.59×0.15 = 0.56; this result neither fully adopts the EEG values ​​affected by interference nor ignores some of their effective information, while taking into account both behavioral trends and historical stability, consistent with the physiological characteristics of children's focus states fluctuating rapidly but changing continuously.

[0086] Example 8: In one embodiment, this application further provides that the focus state segment in step S5 is obtained by sliding statistics on the fused focus values ​​of multiple consecutive EEG time windows, and the sliding statistics include at least mean statistics, fluctuation amplitude statistics, and duration statistics; and includes the following key implementation steps: Step 1: The focus state segment is obtained by sliding statistics of the fused focus values ​​of multiple consecutive EEG time windows. The sliding statistics include at least the mean statistics, fluctuation amplitude statistics, and duration statistics. Sliding statistics can refer to the process of moving along the time axis point by point or window by window on a continuous fusion focus value sequence and repeatedly calculating statistics, using a fixed-length time window as the unit; mean statistics calculate the arithmetic mean of all fusion focus values ​​within the current sliding window, used to characterize the overall focus level within that time period; fluctuation amplitude statistics calculate the difference between the maximum and minimum values, or the standard deviation, of the fusion focus values ​​within the current sliding window, used to characterize the stability of the focus state within that time period; duration statistics record the number of consecutive occurrences of sliding windows that meet specific statistical conditions on the time axis and convert them into actual time lengths, used to eliminate interference from instantaneous abnormal fluctuations; In this embodiment, multiple consecutive EEG time windows refer to N EEG time windows that are adjacent in time and without interval, where N is a positive integer and its value is determined based on the training task rhythm and the typical cycle of changes in the child's attention state; the window length used for sliding statistics is consistent with the EEG time window length, and the sliding step is a single EEG time window to ensure that the statistical results have temporal continuity and comparability; the mean statistics, fluctuation amplitude statistics, and duration statistics work together to form the basic criteria for dividing attention state segments.

[0087] This application determines whether a stable focus state segment has been formed, for example, based on whether the average of the fused focus values ​​within the sliding window exceeds a first state threshold and whether the fluctuation amplitude is less than a first fluctuation threshold; it determines whether a deviating focus state segment has been formed, for example, based on whether the average of the fused focus values ​​within the sliding window is lower than a second state threshold; further, it determines whether a fluctuating transition state segment has been formed based on whether the fused focus values ​​within the sliding window are between the first and second state thresholds, or whether their fluctuation amplitude is greater than the second fluctuation threshold. This application obtains semantic segmentation results of the fused focus value time series based on any of the above methods, supporting the differentiated execution of subsequent training load adjustment strategies.

[0088] For example, in focused jumping training, the system divides the fusion focus value sequence into continuous EEG time windows in 2-second units; it uses a sliding window with a length of 5 time windows (i.e., 10 seconds) for statistical analysis; when the average fusion focus value in a certain sliding window is 0.78 (scale range 0–1), which is higher than the first state threshold of 0.75, and the fluctuation amplitude is 0.06, which is less than the first fluctuation threshold of 0.08, then the 10-second time period is divided into a stable focus state segment; when the average value in another sliding window is 0.42, which is lower than the second state threshold of 0.50, then it is divided into a deviated focus state segment; when the average value in a third sliding window is 0.62 (between 0.50 and 0.75), or the fluctuation amplitude reaches 0.15 (greater than the second fluctuation threshold of 0.12), then it is divided into a fluctuating transition state segment.

[0089] Step 2: When the average fusion focus value of multiple consecutive EEG time windows is greater than the first state threshold and the fluctuation amplitude is less than the first fluctuation threshold, the corresponding time period is divided into a stable focus state segment. The first state threshold is a preset value used to distinguish high levels of focus. Its value is set according to the age of the training subject, baseline focus ability and task type. For example, it is a fixed value or an adaptive update value in the range of 0.70 to 0.80. The first fluctuation threshold is the upper limit tolerance used to constrain the stability of focus. Its value reflects the natural fluctuation range of the child's focus. For example, it is a fixed value in the range of 0.05 to 0.10. In this embodiment, the technical function of this step is to identify the behavioral-physiological synergistic state interval with high focus intensity and high time stability as a sufficient condition for triggering the increase of training load; stable focus is only determined when both the mean and fluctuation dimensions meet the requirements, so as to avoid misjudgment caused by a single peak or short-term stability.

[0090] This application, for example, determines whether a stable focus state segment has been entered by comparing the average fused focus value with a first state threshold and the fluctuation amplitude with a first fluctuation threshold. Alternatively, this application inputs the fused focus value sequence within a sliding window into a dual-condition logic gate, outputting a stable focus flag only when both comparison outputs are true. Furthermore, this application sets a minimum duration constraint, requiring that the state satisfying the above dual conditions exist continuously for at least three sliding windows (i.e., at least 30 seconds) before it can be formally marked as a stable focus state segment. This application obtains reliable identification results of stable focus state segments based on any of the above methods, ensuring the reasonable timing of increasing training difficulty.

[0091] For example, in parkour continuous attention training, if the system detects that the average fusion focus value is always higher than 0.76 within 6 consecutive sliding windows (60 seconds in total), and the fluctuation range of each window does not exceed 0.07, then the 60 seconds are marked as a stable focus state segment, and a signal to increase the difficulty is sent to the training control module.

[0092] Step 3: When the average fusion focus value of multiple consecutive EEG time windows is less than the second state threshold, the corresponding time period is divided into a defocused state segment; The second state threshold is a preset value used to distinguish low focus levels. Its value is lower than the first state threshold, for example, a fixed value or a dynamic calibration value in the range of 0.40 to 0.55. In this embodiment, the technical function of this step is to quickly identify time intervals with significantly insufficient focus resources, serving as a direct basis for triggering a reduction in training load; a single mean criterion can be used to trigger deviation judgment, taking into account both response sensitivity and judgment robustness; it forms an asymmetric judgment structure with the above-mentioned one—stable focus requires dual constraints, while deviation from focus only requires intensity constraints, which is in line with the cognitive characteristics of children's attention being easily distracted and difficult to maintain.

[0093] This application, for example, independently determines the deviation from the focus state segment based on whether the average fused focus value is lower than a second state threshold. This application also introduces a decay timer mechanism: once the average value falls below the second state threshold, a countdown begins; if it does not recover to above the threshold within a preset fallback time threshold (e.g., 20 seconds), it is officially marked as a deviation from the focus state segment. Furthermore, this application superimposes behavioral data for verification; when the average value is lower than the second state threshold and the behavioral representation value also decreases by more than a preset proportion, the deviation determination is accelerated. This application obtains timely identification results of deviation from the focus state segment based on any of the above methods, preventing the training load from continuously exceeding the user's current capability boundaries.

[0094] For example, in color sprint / inhibition control training, if the system detects that the mean fusion focus value is below 0.45 for four consecutive sliding windows (40 seconds), and the reaction time increases by 15% and the false response rate increases by 22% during this period, it is determined to be a deviation from the focus state segment, and the density of interference items is automatically reduced from high to medium, while the frequency of voice prompts is increased.

[0095] Step 4: When the fusion focus value of multiple consecutive EEG time windows is between the first state threshold and the second state threshold, or the fluctuation amplitude is greater than the second fluctuation threshold, the corresponding time period is divided into a fluctuation transition state segment. The second fluctuation threshold is a preset value used to define the boundary of moderate fluctuations. Its value is higher than the first fluctuation threshold, for example, a fixed value in the range of 0.10 to 0.15. In this embodiment, the technical function of this step is to capture the intermediate state where the focus state has not yet stabilized or has already shown obvious disturbances, as the basis for maintaining the current training load and reducing feedback sensitivity; or the relationship indicates that either of the two situations can trigger the fluctuation transition judgment, reflecting the inclusive design for uncertain states; this state segment does not drive the training parameters to increase or decrease, but affects the feedback response mechanism, thereby improving the system interaction stability.

[0096] This application determines whether a state is a fluctuating transition state based on whether the fusion focus value falls within the range between a first state threshold and a second state threshold. Alternatively, it independently determines a fluctuating transition state based on whether the fluctuation amplitude exceeds a second fluctuation threshold. Furthermore, this application sets a fluctuation persistence constraint, requiring at least two consecutive sliding windows of fluctuation amplitude exceeding the limit before formally marking it as a fluctuating transition state segment. This application obtains reasonable identification results for fluctuating transition state segments based on any of the above methods, providing triggering conditions for the training feedback smoothing mechanism.

[0097] For example, in a study room scenario, the system detects that the average fusion focus value within a certain sliding window is 0.61 (between 0.50 and 0.75). Although it does not reach the stable focus standard, it does not fall into the deviation range. At the same time, its fluctuation amplitude is 0.13, which exceeds the second fluctuation threshold of 0.12. Based on this, the system marks it as a fluctuation transition state segment and automatically changes the intensity of visual feedback from emphasized to medium, while temporarily suspending the adjustment of the character speed mapping slope to maintain the current training rhythm unchanged.

[0098] Example 9: In another optional embodiment, this application also provides steps S6 and S7 for adaptively adjusting the training load and generating training evaluation results, including: Step 1: When the current focus state is a stable focus state and the duration reaches the preset improvement duration threshold, increase the training load; Among them, the stable focus state segment can refer to the time period in which the average fusion focus value of multiple consecutive EEG time windows is greater than the first state threshold and the fluctuation amplitude is less than the first fluctuation threshold; the preset improvement duration threshold is a fixed duration set according to the training task type, which is 15 seconds, 30 seconds or 60 seconds. The role of this technical feature in this embodiment is to avoid sudden changes in training load due to a momentary increase in attention value, and to ensure that the increase in training difficulty is based on continuous and reliable attention performance, thereby maintaining the stability of training rhythm and the continuity of children's task participation.

[0099] Step 2: When the current focus state segment is a deviated focus state segment and the duration reaches the preset backoff duration threshold, reduce the training load; Among them, the deviation from the focus state segment can refer to the time period when the average fusion focus value of multiple consecutive EEG time windows is less than the second state threshold; the second state threshold is lower than the first state threshold; the preset regression time threshold is a fixed duration set according to the training task type, which is 20 seconds, 45 seconds or 90 seconds. The function of this technical feature in this embodiment is to trigger a reduction in difficulty only when the level of focus is consistently low, to prevent misjudgment and downgrading due to accidental factors such as single reaction delay or brief distraction, and to ensure a balance between training challenge and user sense of progress.

[0100] Step 3: When the current focus state is in a fluctuating transition state, maintain the current training load and reduce the sensitivity of training feedback to changes in a single EEG time window; Among them, the fluctuating transition state segment can refer to the time period in which the fusion focus value of multiple consecutive EEG time windows is between the first state threshold and the second state threshold, or the fluctuation amplitude is greater than the time period corresponding to the second fluctuation threshold; reducing the sensitivity of training feedback to changes in a single EEG time window can refer to: applying a moving average filter to the fusion focus value used to drive interface feedback, setting an upper limit on the slope of change, introducing a hysteresis response mechanism, or limiting the number of times feedback parameters are updated per unit time. The role of this technical feature in this embodiment is to suppress feedback jitter caused by EEG noise, occasional behavioral deviations, or short-term attention drift during the transitional stage when the state of focus has not yet been stabilized, so that interactive outputs such as character movement, visual cues, and sound feedback are smoother and more predictable, which is in line with children's cognitive rhythms and operating habits.

[0101] Step 4: Generate training evaluation results based on the percentage of stable focus state segments, the percentage of defocused state segments, the percentage of fluctuating transition state segments, the trend of behavioral representation value changes, and the number of training load adjustments, and determine the initial training configuration for the next training cycle based on the training evaluation results. Among them, the proportion of stable focus states can refer to the total time spent in stable focus states during training, which is the proportion of the total training time; the proportion of defocused states can refer to the total time spent in defocused states during training, which is the proportion of the total training time; the proportion of fluctuating transitional states can refer to the total time spent in fluctuating transitional states during training, which is the proportion of the total training time; the trend of behavioral representation values ​​can refer to the rising, falling, or stable change pattern of behavioral representation values ​​corresponding to each EEG time window during training, which is obtained through linear fitting slope, segmental mean comparison, or moving window trend determination methods; the number of training load adjustments can refer to the total number of times the above, the above, or any of the above load adjustment actions are performed during this training process. The role of this technical feature in this embodiment is to: structure and integrate multi-dimensional process indicators into interpretable and reusable evaluation dimensions to support cross-cycle strategy evolution; wherein, the proportion of the three types of state segments jointly characterizes the distribution characteristics of focus stability, the trend of behavioral representation value changes reflects the co-evolutionary relationship between explicit behavior and internal focus, and the number of training load adjustments reflects the frequency of system intervention and user adaptation load, which together constitute a comprehensive measure of training effectiveness and individual adaptability.

[0102] Based on the conditions that the proportion of stable focus state segments is higher than the preset stable proportion threshold, the behavioral representation value shows a positive growth trend, and the number of training load adjustments does not exceed the preset upper limit, this application determines that the training effect is good, and accordingly sets the initial task difficulty of the next training cycle to the current highest achievement difficulty. Based on the conditions that the proportion of the defocused state segment is consistently higher than the preset defocused state segment threshold, the behavioral representation value shows a negative decay trend, and the training load has been reduced two or more times, this application determines that the user is currently in a capability bottleneck period. Accordingly, the initial training configuration for the next training cycle is set to reduce the starting difficulty, extend the stimulus presentation duration, and increase the density of guiding voice prompts. Furthermore, based on the conditions that the proportion of the fluctuating transitional state segment is significantly increased, the behavioral representation value shows a flattening trend but no significant decrease, and the number of training load adjustments is concentrated on the type of reducing feedback sensitivity, this application determines that the user is in a focus strategy adjustment phase. Accordingly, the initial training configuration for the next training cycle is set to maintain a medium difficulty, enable a stability-first training mode, and strengthen the continuous operation reward mechanism.

[0103] This application obtains a quantitative evaluation result of the multidimensional performance of the training process based on any of the above methods, and realizes the personalized generation of the initial training configuration for the next training cycle accordingly.

[0104] For example, in an 8-minute continuous attention training session for parkour, the system identified a stable focus state of 137 seconds (28.5%), a defocused state of 92 seconds (19.2%), and a fluctuating transition state of 251 seconds (52.3%). The behavioral representation value showed a slow increase in the first 3 minutes and a flattening trend in the last 5 minutes. The training load was adjusted 3 times, all of which were the aforementioned operations to reduce feedback sensitivity. The training evaluation results generated were: focus started quickly but lacked sustainability, and the behavioral response was approaching saturation. It was recommended that the next cycle extend the single-segment focus maintenance target and increase the proportion of training with phased rhythm changes. Based on this, the system set the initial training configuration for the next training cycle as follows: the starting difficulty remained at the current level, but the threshold for continuous stable control duration was increased from 8 seconds to 10 seconds, and a rhythm disturbance event was inserted in the 4th minute to enhance the anti-interference training effect.

[0105] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0106] Figure 3 A schematic diagram of the structure of a focus training system combining EEG and behavioral data provided in an embodiment of this application is shown below. Figure 3 As shown, the attention training system 300 combining EEG and behavioral data in this embodiment includes: a data synchronization and alignment module 301, an EEG feature calculation module 302, a consistency verification module 303, an attention value gating correction module 304, an attention state division module 305, a training load adaptive module 306, and an evaluation and configuration determination module 307.

[0107] The data synchronization and alignment module 301 is used to synchronously collect the EEG data and behavioral data of the training subjects during the operation of the attention training task, and divide the EEG data into continuous EEG time windows based on a unified time reference, and map the behavioral data to the corresponding EEG time windows. The EEG feature calculation module 302 is used to preprocess the EEG data in each EEG time window, extract the EEG focus features, and calculate the EEG confidence index of the corresponding EEG time window according to the artifact interference, so as to obtain the EEG candidate focus value of each EEG time window. The consistency verification module 303 is used to extract behavioral focus features from behavioral data within each EEG time window, generate corresponding behavioral representation values, and obtain consistency verification results based on the deviation relationship between the behavioral representation values ​​and the candidate EEG focus values. The focus value gating correction module 304 is used to perform gating correction on the candidate focus values ​​of the EEG based on the EEG credibility index and the consistency verification result, so as to obtain the fused focus value of the corresponding EEG time window. The focus state segmentation module 305 is used to determine the continuity of focus state during training based on the fused focus value of multiple consecutive EEG time windows and to segment focus state segments. The training load adaptive module 306 is used to adaptively adjust the training load of the focus training task according to the type and duration of the current focus state segment, wherein the training load includes at least one of task difficulty, stimulus rhythm and feedback sensitivity. The evaluation and configuration determination module 307 is used to generate training evaluation results based on the distribution results of each focus state segment during training and the changes in corresponding behavioral data after the training task is completed, and to determine the initial training configuration for the next training cycle based on the training evaluation results.

[0108] The system in this embodiment can be used to execute the methods of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

Claims

1. A concentration training method combining electroencephalogram and behavior data, characterized in that, include: S1. During the focus training task, the EEG data and behavioral data of the training subjects are collected synchronously, and the EEG data is divided into continuous EEG time windows based on a unified time reference, and the behavioral data is mapped to the corresponding EEG time windows. S2. Preprocess the EEG data within each EEG time window, extract EEG focus features, and calculate the EEG reliability index for the corresponding EEG time window based on artifact interference to obtain the candidate EEG focus value for each EEG time window. S3. Extract behavioral focus features from the behavioral data within each EEG time window, generate corresponding behavioral representation values, and obtain consistency verification results based on the deviation relationship between the behavioral representation values ​​and the candidate EEG focus values. S4. Based on the EEG credibility index and the consistency verification result, the EEG candidate focus value is gating and corrected to obtain the fusion focus value of the corresponding EEG time window; S5. Based on the fusion focus value of multiple consecutive EEG time windows, the continuity of focus state during the training process is determined, and focus state segments are divided. S6. Based on the type and duration of the current focus state segment, adaptively adjust the training load of the focus training task, wherein the training load includes at least one of task difficulty, stimulus rhythm and feedback sensitivity. S7. After the training task is completed, a training evaluation result is generated based on the distribution results of each focus state segment during the training process and the corresponding changes in behavioral data, and the initial training configuration for the next training cycle is determined based on the training evaluation result.

2. The concentration training method of claim 1, wherein, In step S2, the EEG data within each EEG time window are preprocessed, including: Bandpass filtering, power frequency suppression, baseline correction, and artifact detection are performed sequentially to obtain preprocessed EEG data; Based on the preprocessed EEG data, at least two of the following are extracted: frequency band energy features, frequency band ratio features, and time window fluctuation features, to generate the candidate EEG focus values; In step S2, calculating the EEG reliability index includes: Calculate the proportion of low-frequency interference components within the current EEG time window to characterize the degree of low-frequency drift caused by eye movements or head movements; Calculate the proportion of high-frequency interference components within the current EEG time window to characterize the degree of electromyographic interference; The degree of motion artifacts is characterized by statistically analyzing at least one of the following within the current EEG time window: the proportion of amplitude abrupt change points, the proportion of saturation sampling points, and the proportion of abnormal fluctuations in adjacent sampling points. The proportions of low-frequency interference components, high-frequency interference components, and action artifact representations are normalized and weighted to obtain the EEG reliability index. 3.The focus training method of combining electroencephalogram and behavior data according to claim 2, characterized in that, In step S2, the EEG time window is divided into a high-confidence time window, a time window to be corrected, and a low-confidence time window according to the EEG confidence index: Among them, the EEG candidate focus values ​​corresponding to the high confidence time window directly enter the consistency check, the EEG candidate focus values ​​corresponding to the time window to be corrected enter the gating correction, and the EEG candidate focus values ​​corresponding to the low confidence time window are suppressed and directly participate in the focus state determination of the current EEG time window. 4.The focus training method of combining electroencephalogram and behavior data according to claim 1, wherein, In step S3, the behavioral representation value is determined based on the behavioral data corresponding to the current EEG time window. The behavioral data includes at least two of the following: reaction time, answer accuracy, missed answer rate, accidental touch rate, continuous operation stability, and task switching delay. When generating the behavioral representation value, the following steps are included: The behavioral data were converted into symmetrical evaluation metrics representing the degree of focus. The converted symmetrical evaluation quantities are normalized. The normalized positive evaluation values ​​are weighted and synthesized to obtain the behavioral representation value of the current EEG time window.

5. The concentration training method of claim 4, wherein the brainwave data and the behavior data are combined. In step S3, the consistency verification includes: Calculate the deviation between the behavioral representation value and the candidate focus value of the current EEG time window; The deviation value is compared with a preset deviation threshold. When the deviation value exceeds the preset deviation threshold, the current EEG time window is determined to be an EEG-behavioral inconsistency time window; The preset deviation threshold is set according to the type of training task.

6. The concentration training method of claim 5, wherein the brainwave data and the behavior data are combined. In step S4, gating correction is performed for the EEG-behavioral inconsistency time window, and the gating correction includes at least two of the following key implementation steps: The candidate EEG focus values ​​are attenuated based on the EEG reliability index of the current EEG time window to obtain the corrected EEG focus values. The corrected EEG focus value is combined with the fused focus value of the previous EEG time window to constrain the instantaneous jump variable of the current EEG time window. When the current EEG time window is a low-confidence time window, the behavioral representation value is used to replace the candidate EEG focus value in the generation of the fusion focus value for the current EEG time window.

7. The concentration training method of claim 6, wherein the brainwave data and the behavior data are combined. In step S4, generating the fusion focus value includes: Determine the gating coefficient corresponding to the current EEG time window based on the EEG reliability index; The candidate focus values ​​of the EEG are corrected using the gating coefficient to obtain the corrected focus values ​​of the EEG. The corrected EEG focus value, the behavioral representation value of the current EEG time window, and the fusion focus value of the previous EEG time window are combined to obtain the fusion focus value of the current EEG time window. 8.The focus training method of claim 1, wherein, In step S5, the focus state segment is obtained by performing sliding statistics on the fused focus values ​​of multiple consecutive EEG time windows. The sliding statistics include at least mean statistics, fluctuation amplitude statistics, and duration statistics, and include the following key implementation steps: When the average fusion focus value of multiple consecutive EEG time windows is greater than the first state threshold and the fluctuation amplitude is less than the first fluctuation threshold, the corresponding time period is divided into a stable focus state segment. When the average fusion focus value of multiple consecutive EEG time windows is less than the second state threshold, the corresponding time period is divided into a defocused state segment. When the fusion focus value of multiple consecutive EEG time windows is between the first state threshold and the second state threshold, or the fluctuation amplitude is greater than the second fluctuation threshold, the corresponding time period is divided into a fluctuation transition state segment. 9.The focus training method of claim 8, wherein, In steps S6 and S7, the training load is adaptively adjusted and training evaluation results are generated, including: When the current focus state is a stable focus state and the duration reaches the preset improvement duration threshold, increase the training load; When the current focus state segment is a deviated focus state segment and the duration reaches the preset backoff duration threshold, reduce the training load; When the current focus state is in a fluctuating transition state, maintain the current training load and reduce the sensitivity of training feedback to changes in a single EEG time window. Training evaluation results are generated based on the percentage of stable focus states, the percentage of defocus states, the percentage of fluctuating transition states, the trend of behavioral representation value changes, and the number of training load adjustments. The initial training configuration for the next training cycle is then determined based on the training evaluation results.

10. A concentration training system combining electroencephalogram and behavior data, characterized in that, The method applied to any one of claims 1-9 includes: The data synchronization and alignment module is used to synchronously collect the EEG data and behavioral data of the training subjects during the operation of the attention training task, and divide the EEG data into continuous EEG time windows based on a unified time reference, and map the behavioral data to the corresponding EEG time windows. The EEG feature calculation module is used to preprocess the EEG data within each EEG time window, extract EEG focus features, and calculate the EEG confidence index of the corresponding EEG time window based on the artifact interference, so as to obtain the EEG candidate focus value for each EEG time window. The consistency verification module is used to extract behavioral focus features from behavioral data within each EEG time window, generate corresponding behavioral representation values, and obtain consistency verification results based on the deviation relationship between the behavioral representation values ​​and the candidate EEG focus values. The focus value gating correction module is used to gating and correct the candidate focus values ​​of the EEG based on the EEG credibility index and the consistency verification result, so as to obtain the fused focus value of the corresponding EEG time window. The focus state segmentation module is used to determine the continuity of focus state during training based on the fused focus value of multiple consecutive EEG time windows, and to segment focus state segments. The training load adaptive module is used to adaptively adjust the training load of the focus training task according to the type and duration of the current focus state segment, wherein the training load includes at least one of task difficulty, stimulus rhythm and feedback sensitivity. The evaluation and configuration determination module is used to generate training evaluation results based on the distribution results of each focus state segment during training and the changes in corresponding behavioral data after the training task is completed, and to determine the initial training configuration for the next training cycle based on the training evaluation results.

Images