Continuous concentration assessment system and method based on multi-dimensional data

Abstract

The application provides a continuous concentration evaluation system and method based on multi-dimensional data, and belongs to the technical field of attention evaluation. The system comprises a multi-dimensional data acquisition module for acquiring multi-dimensional test data; a curve construction module for constructing a baseline and generating an individual curve with the same characteristics as the baseline by using a sliding window technology; an analysis module for performing abnormality judgment when the individual curve is generated; and a control module for extracting the decay eigenvalue of the individual curve by using a symmetric rotation and other convolution technology during the early warning observation period, and starting an evaluation termination mechanism when the decay eigenvalue still shows abnormality at the end of the early warning observation period. Through multi-modal data fusion and individual baseline construction, the application realizes high-precision dynamic identification of the attention decay process, and outputs the continuous concentration time length as an evaluation index based on the evaluation start and end time, thereby significantly improving the accuracy and individual adaptability of the concentration evaluation.

Description

Technical Field

[0001] This invention relates to the field of attention assessment technology, and more specifically to a continuous attention assessment system and method based on multi-dimensional data. Background Technology

[0002] Attention is a key factor influencing a child's learning and quality of life during their development. In the school environment, children who can concentrate absorb knowledge efficiently, exhibiting significantly higher learning efficiency than those who cannot. Children who cannot concentrate are more likely to miss crucial knowledge points, thus impacting their academic performance. In the medical field, accurately assessing a child's attention level is crucial for diagnosing conditions such as Attention Deficit Hyperactivity Disorder (ADHD) and serves as an important basis for developing scientific treatment plans. Educational institutions also need to optimize their teaching content and methods based on students' attention levels to improve teaching effectiveness.

[0003] However, existing technologies have many shortcomings. Internationally used attention testing methods assess attention solely based on the speed and accuracy of answering questions. Children's attention is influenced by various factors such as emotions and environmental distractions, which cannot be reflected in simple question-and-answer data. This makes it difficult for assessment results to comprehensively and accurately reflect children's attention levels. Some studies use average EEG values ​​over a fixed period to judge attention, but EEG signals are extremely weak, requiring high amplification during collection and processing. This amplifies environmental noise and interference signals from the equipment itself, leading to larger errors. Furthermore, existing testing methods generally have fixed test durations. Young children have limited endurance and attention span. Faced with tests lasting 15-30 minutes or even longer, they are easily distracted by fatigue, boredom, or other emotions, leading to incomplete tests or inaccurate results that fail to accurately reflect their attention levels. Summary of the Invention

[0004] This invention provides a continuous attention assessment system and method based on multi-dimensional data to solve the problems of low testing efficiency and poor accuracy in existing technologies.

[0005] To achieve the above objectives, embodiments of the present invention provide a continuous attention assessment system based on multi-dimensional data. The continuous attention assessment system includes: a multi-dimensional data acquisition module, used to acquire multi-dimensional test data according to a preset time interval and sequence, the multi-dimensional test data including response data, EEG activity data, and eye movement data; a curve construction module, used to construct a baseline based on the acquired multi-dimensional test data, including prefrontal cortex EEG signal activity, the proportion of time the eyes are focused on the screen, and parameters related to response data, and to process the acquired multi-dimensional test data using sliding window technology to generate individual curves with the same characteristics as the baseline; an analysis module, used to perform anomaly warning judgment using preset anomaly judgment rules when generating the individual curves; and a control module, used to extract the decay feature value of the individual curves during the warning observation period using symmetrical rotational isovariant convolution technology, and to activate the assessment termination mechanism when the decay feature value still shows an anomaly at the end of the warning observation period, and to use the duration from the start to the end of the test as an evaluation indicator of continuous attention duration.

[0006] Optionally, the construction of the baseline, which includes prefrontal EEG signal activity, the proportion of time spent looking at the screen, and the response parameters, includes: normalizing and correcting the collected multidimensional test data, extracting features of prefrontal EEG signal activity, the proportion of time spent looking at the screen, and the parameters related to the response to the questions to obtain multimodal features; and using a weighted Gaussian mixture model to fuse the multimodal features to obtain the baseline.

[0007] Optionally, the step of using sliding window technology to process the collected multidimensional test data and generate individual curves with the same features as the baseline includes: initializing the sliding window and its parameter configuration according to a preset step size; extracting features of prefrontal EEG signal activity, the proportion of time spent looking at the screen, and parameters related to answer responses within the current sliding window to obtain multimodal features; fusing the multimodal features and performing time axis alignment and missing value processing to generate the individual curve of the current sliding window; and executing the window closing procedure.

[0008] Optionally, when generating the individual curve, anomaly judgment is performed using preset anomaly judgment rules, including: if the parameter features related to the answer response show that the dynamic average accuracy rate within a first preset time period is less than the corresponding feature in the baseline, and the variance of the answer response time is greater than a multiple of the dynamic variance of the data in that dimension within the first preset time period, and the duration exceeds a preset abnormal time, then the index of the answer response state dimension is abnormal; if the feature of the proportion of time spent looking at the screen shows that the visual concentration is lower than a preset value, then the index of the eye movement state dimension is abnormal; if the feature of the prefrontal cortex EEG signal activity shows that it is lower than the corresponding feature in the baseline by a preset degree, then the index of the EEG activity state dimension is abnormal.

[0009] Optionally, when generating the individual curve, anomaly judgment is performed using preset anomaly judgment rules, including: comparing and analyzing the data of each dimension in the individual curve with the characteristics of the baseline, and judging the anomaly of each dimension of the data in combination with the anomaly judgment rules; when more than two dimensions of indicators are judged as abnormal, attention state decay is determined and marked as an observation period.

[0010] Optionally, the step of extracting the decay feature values ​​of the individual curve using symmetric rotational equivariant convolution technology includes: constructing a model for extracting the decay feature values ​​of the individual curve, wherein a symmetric rotational equivariant convolution layer is configured to extract the basic features of the rotational equivariance of the individual curve, a multi-scale feature pyramid module is configured to capture decay patterns at different time scales, and a decay feature regression module is configured to output the decay feature values; the model is trained based on a training dataset and using a three-stage training method to extract the decay feature values ​​of the individual curve using the trained model.

[0011] Optionally, the control module is further configured to: during the early warning observation period, if the attenuation characteristic value shows a normal fluctuation range returning to the previous attenuation characteristic value, then continue the evaluation process.

[0012] Optionally, the continuous focus assessment system further includes a result presentation module, which is used to: generate a visual chart based on the focus assessment results to display the decay trend and the final determined continuous focus duration value; and generate focus assessment results.

[0013] This invention also provides a continuous attention assessment method, which includes: collecting multidimensional test data according to a preset time interval and sequence, wherein the multidimensional test data includes answer response data, EEG activity data, and eye movement data;

[0014] Based on the collected multidimensional test data, a baseline is constructed, including prefrontal cortex EEG signal activity, the proportion of time spent looking at the screen, and parameters related to answer responses. A sliding window technique is used to process the collected multidimensional test data and generate individual curves with the same characteristics as the baseline. During the generation of these individual curves, a pre-defined anomaly detection rule is used for anomaly warning. After an anomaly warning is triggered, an observation period begins. A symmetric rotational isovariant convolution technique is used to extract the decay characteristic values ​​of the individual curves. If the decay characteristic values ​​still indicate anomalies at the end of the observation period, a test termination mechanism is activated, and the duration from the start to the end of the test is used as an evaluation indicator of continuous focus duration.

[0015] This invention also provides a machine-readable storage medium storing instructions that cause a machine to perform the aforementioned continuous attention assessment method.

[0016] This invention provides a continuous attention assessment system based on multi-dimensional data, comprising: a multi-source data acquisition module for acquiring multi-dimensional test data according to a preset time interval and sequence, the multi-dimensional test data including response data, EEG activity data, and eye movement data; a curve construction module for constructing a baseline based on the acquired multi-dimensional test data, including prefrontal EEG signal activity, the proportion of time spent looking at the screen, and parameters related to response, and processing the acquired multi-dimensional test data using sliding window technology to generate individual curves with the same characteristics as the baseline; an analysis module for performing anomaly judgment using preset anomaly judgment rules when generating the individual curves; and a control module for extracting the decay feature value of the individual curves during the warning observation period using symmetric rotation isovariant convolution technology, and activating an assessment termination mechanism when the assessment termination time still shows an anomaly at the end of the warning observation period, and using the duration from the start to the end of the assessment as an assessment indicator of continuous attention duration. This invention achieves high-precision dynamic extraction of attention decay characteristics through multimodal data fusion (EEG signals, eye tracking, and response data) and sliding window technology. Based on dynamically constructed individual baselines and multi-dimensional anomaly judgment rules, it can monitor attention fluctuations in real time and trigger an early warning mechanism, significantly improving the accuracy, anti-interference, and personalized adaptability of attention assessment. It can be widely applied to the quantitative analysis of attention states in education, medical care, and human-computer interaction scenarios. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying 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 invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0018] Figure 1 This is a schematic diagram of the continuous attention assessment system provided in an embodiment of the present invention;

[0019] Figure 2 This is a schematic diagram of the continuous attention assessment system provided in a preferred embodiment of the present invention;

[0020] Figure 3 This is a flowchart of an example continuous attention assessment system;

[0021] Figure 4 This is a schematic diagram illustrating the process of building a baseline;

[0022] Figure 5 This is a schematic diagram illustrating an example of continuous attention assessment;

[0023] Figure 6 This is a flowchart illustrating the symmetric rotation equivariant convolution technique. Detailed Implementation

[0024] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0025] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0026] In a child's development, attention, as a core cognitive ability, directly impacts learning efficiency, knowledge absorption, and daily behavior. Especially in education, children with good concentration are more likely to achieve learning goals, while those with poor concentration face learning difficulties and academic setbacks. Therefore, developing a more accurate and efficient attention assessment system is crucial.

[0027] To address the aforementioned issues, this invention provides a continuous attention assessment system and method based on multi-dimensional data. By integrating multi-dimensional data from EEG, eye movement, and behavioral responses, and combining deep learning to construct dynamic decay baselines and individual baselines, it utilizes symmetrical rotational convolution to accurately capture attention decay characteristics, enabling real-time monitoring and early warning across scenarios. Its adaptive test duration adjustment mechanism and anti-interference data processing technology significantly improve assessment accuracy, meeting the precise needs of personalized attention measurement analysis in fields such as educational diagnosis and medical rehabilitation.

[0028] Please refer to Figures 1-3 This invention provides a continuous attention assessment system based on multi-dimensional data. The continuous attention assessment system includes: a multi-dimensional data acquisition module, used to acquire multi-dimensional test data according to a preset time interval and sequence, the multi-dimensional test data including answer response data, EEG activity data, and eye movement data; a curve construction module, used to construct a baseline based on the acquired multi-dimensional test data, including prefrontal EEG signal activity, the proportion of time the eyes are focused on the screen, and parameters related to answer response, and to process the acquired multi-dimensional test data using sliding window technology to generate individual curves with the same characteristics as the baseline; an analysis module, used to perform abnormal warning judgment using preset abnormal judgment rules when generating the individual curves; and a control module, used to enter a warning observation period after an abnormal warning is triggered, and to extract the decay feature value of the individual curves using symmetric rotation isovariant convolution technology. When the warning observation period ends and the decay feature value still shows an abnormality, the assessment termination mechanism is activated, and the duration from the start to the end of the test is used as the assessment index of continuous attention duration.

[0029] Among these parameters, those related to response time may include the accuracy of the response and / or the variability of the response time.

[0030] This invention achieves high-precision dynamic extraction of attention decay characteristics through multimodal data fusion (EEG signals, eye tracking, and response data) and sliding window technology. Based on dynamically constructed individual baselines and multidimensional anomaly judgment rules, it can monitor attention fluctuations in real time and trigger early warning mechanisms, significantly improving the accuracy, anti-interference, and personalized adaptability of attention assessment. It can be widely applied to the quantitative analysis of attention states in education, healthcare, and human-computer interaction scenarios. In this embodiment, an individual baseline and individual curve (including the baseline) are constructed based on the individual's multidimensional test data, and decay characteristic values ​​are calculated for each individual, making the monitoring of the continuous attention assessment system more objective.

[0031] Please refer to Figure 2For example, after collecting multidimensional test data, this embodiment of the invention preferably utilizes an improved deep singular value decomposition (DSD) denoising algorithm to effectively improve the quality of the multidimensional test data. For instance, when collecting multidimensional information such as children's EEG signals and eye-tracking data, noise such as power line interference, electromyography artifacts, and ambient light fluctuations is easily mixed in, making it difficult for traditional methods to accurately separate signals from noise. The DSD denoising algorithm, however, constructs a deep neural network to automatically learn noise distribution characteristics and adaptively separates the signal subspace and noise subspace using a learnable nonlinear transformation layer. This algorithm can completely preserve the subtle features in the EEG signals, such as the power changes of prefrontal cortex alpha, beta, theta, and gamma waves reflecting attentional states, while efficiently removing various noise interferences, significantly improving the signal-to-noise ratio of the EEG signals. Simultaneously, it effectively purifies the interference information in the eye-tracking data, resulting in cleaner multidimensional data. This lays a solid foundation for subsequent accurate extraction of attenuation feature values ​​using symmetric rotational convolution techniques and accurate anomaly detection and attention assessment based on the attenuation curve baseline, effectively ensuring the reliability and accuracy of the continuous attention assessment system results.

[0032] Furthermore, image data from the multidimensional test data is randomly selected and downsampled. The downsampled image data is then stitched together in a grid to generate a hybrid image, thus preserving multi-scale information. All categories of labels in the generated hybrid image are then merged to generate a hybrid image label that retains multi-hot encoding, thereby improving label diversity.

[0033] For image data in the multidimensional test data, the system further optimizes it using data augmentation techniques. By randomly selecting images and downsampling them, the downsampled images are then stitched together in a grid to generate a hybrid image. Simultaneously, labels are merged to form multi-hot encoded hybrid image labels. This series of operations essentially creates rich and diverse "learning materials" for the model. On one hand, the hybrid image retains multi-scale information, enabling the model to learn features at different levels of detail, enhancing its understanding of complex image structures. On the other hand, the multi-hot encoded labels enrich label diversity, better meeting the needs of multi-label image classification training, allowing the model to access more comprehensive sample information during training, thereby improving the model's generalization ability and robustness. These data processing methods work closely together, from unified integration of data over time to data quality purification and enhancement, and then to the enrichment and enhancement of data features, comprehensively optimizing the data processing flow. This effectively reduces data errors and interference, improves data usability, and provides solid and reliable data support for the subsequent system to extract data features and accurately evaluate continuous focus using symmetric rotation and equal-variable convolution techniques, significantly improving the accuracy and stability of the entire evaluation system.

[0034] In a preferred embodiment of the present invention, the construction of a baseline including prefrontal EEG signal activity, the proportion of time spent looking at the screen, and the variability of response time can include: normalizing and correcting some of the collected multidimensional test data, extracting features of prefrontal EEG signal activity, the proportion of time spent looking at the screen, and the variability of response time to obtain multimodal features; and using a weighted Gaussian mixture model to fuse the multimodal features to obtain the baseline.

[0035] Some of the multidimensional test data can be collected from the test subjects within a preset time period. After extracting features such as prefrontal EEG signal activity, the proportion of time spent looking at the screen, and the variability of response time, these features can be expanded. For example, alpha wave envelope slope (i.e., time-varying feature of prefrontal EEG signal) can be added to the prefrontal EEG signal activity feature; fixation hotspot transfer entropy (i.e., eye movement feature) can be added to the proportion of time spent looking at the screen; and reaction acceleration ratio (rate of change in reaction time in later trials, i.e., reaction pattern) can be added to the variability of response time.

[0036] In this embodiment of the invention, a weighted Gaussian mixture model is used to fuse the multimodal features. The weighted Gaussian mixture model can be constructed using the following formula:

[0037]

[0038] Where X represents the normalized and corrected portion of the multidimensional test data, K is the preset number of clusters, and w k The weights can be set based on expert experience, u k Let Σ be the mean vector of the k-th class features. k It is a diagonal covariance matrix.

[0039] When using a weighted Gaussian mixture model to fuse the multimodal features to obtain a baseline, drift detection can also be performed. When significant drift is detected, a learning engine can be added to the weighted Gaussian mixture model, and credibility assessment can be performed. When the credibility exceeds a threshold, the baseline is corrected. For example, the test taker can be prompted to close their eyes for three minutes before preparing for the test.

[0040] Please refer to Figure 4 For example, embodiments of the present invention capture the nonlinear relationships between features through WGMM to achieve multimodal joint probabilistic modeling; utilize static statistical baselines combined with dynamic probabilistic baselines to achieve dual-track baselines; and based on the combination of drift detection and incremental learning, provide real-time adaptive updates and accurate baselines.

[0041] In a preferred embodiment of the present invention, the step of using sliding window technology to process the collected multidimensional test data and generate individual curves with the same features as the baseline includes: initializing the sliding window and its parameter configuration according to a preset step size; extracting features of prefrontal EEG signal activity, the proportion of time spent looking at the screen, and the variability of response time within the current sliding window to obtain multimodal features; fusing the multimodal features and performing time axis alignment and missing value processing to generate the individual curve of the current sliding window; and executing the window closing process.

[0042] In this embodiment of the invention, the sliding window and its parameter configuration are initialized according to a preset step size (detection interval, for example, data is collected once every 1 second after the test begins). For example, parameters such as time-driven window, window length, sliding step size, minimum percentage of valid data, and time-triggered window are defined. Within the current sliding window, the aforementioned weighted Gaussian mixture model can be reused to extract features of prefrontal cortex EEG signal activity, the percentage of time the eyes are focused on the screen, and the variability of response time, to obtain multimodal features. After time axis alignment and missing value processing, individual curves for the current sliding window are generated. In a preferred embodiment of the invention, a multi-level window architecture can be used, which may include a time-driven window (preset step size) and an event-driven window (events such as task start, task end, and curve presentation). Furthermore, a window closing process can be executed when the end time of the window is reached. For example, invalid window marking and buffer memory release can be performed for resource reclamation.

[0043] In a preferred embodiment of the present invention, the window parameters can also be dynamically adjusted according to the data quality. For example, when the quality of EEG activity data is poor, the window time can be extended; when the eye movement count loss rate is high, an event-driven window can be switched; and during high-frequency response periods, the window time can be shortened, etc.

[0044] The embodiments of the present invention achieve the generation of high-quality individual curves through a finely designed sliding window mechanism and multimodal feature calculation, providing reliable data support for continuous attention assessment.

[0045] In a preferred embodiment of the present invention, when generating the individual curve, anomaly judgment is performed using preset anomaly judgment rules, including: if the parameter features related to the answer response show that the dynamic average accuracy rate within a first preset time period is less than the corresponding feature in the baseline, and the variance of the answer response time is greater than a multiple of the dynamic variance of the data in that dimension within the first preset time period, and the duration exceeds a preset abnormal time, then the index of the answer response state dimension is abnormal; if the feature of the proportion of time the eyes are focused on the screen shows that the visual concentration is lower than a preset value, then the index of the eye movement state dimension is abnormal; if the feature of the prefrontal cortex EEG signal activity shows that it is lower than the corresponding feature in the baseline by a preset degree, then the index of the EEG activity state dimension is abnormal.

[0046] In a preferred embodiment of the invention, for each dimension of the test data, the extracted feature values ​​are compared and analyzed in real time with the data of the corresponding time and dimension on the pre-constructed individual baseline. Through meticulous comparison, subtle changes in the data can be keenly captured. Combined with preset anomaly judgment rules, this module can accurately judge the anomalies in the data of each dimension. When more than one dimension of data is judged as abnormal, and there is no reverse trend in the remaining dimensions, the system will determine that the attention state has significantly declined. This judgment method, which integrates multi-dimensional data and considers the trend relationship between data, avoids misjudgment caused by single-dimensional anomalies, is more in line with the complex reality of children's attention changes, and significantly improves the accuracy and reliability of the judgment. Once the attention state is judged to have significantly declined, or the test time reaches the test time corresponding to the test subject, the system will immediately stop the multi-source data acquisition module from collecting data, promptly terminating the invalid or already achieved data collection process, and improving the evaluation efficiency. Based on the anomalies in the data of each dimension and the judgment results, combined with the individual baseline, a comprehensive and integrated evaluation of the child's continuous attention is conducted, and the duration of continuous attention is determined. This assessment method is no longer limited to simple data comparison, but fully considers the changing trends, anomalies and baseline standards of the data. It can more realistically and accurately reflect the children's concentration during the test, providing valuable reference for education, medical care and other fields related to children's attention, and helping to develop more scientific and reasonable intervention measures and teaching programs.

[0047] In a preferred embodiment of the present invention, when generating the individual curve, anomaly judgment is performed using a preset anomaly judgment rule, including: comparing and analyzing the data of each dimension in the individual curve with the features of the baseline, and judging the anomaly of each dimension of the data in combination with the anomaly judgment rule; when more than two dimensions of indicators are judged as abnormal, attention state decay is determined and marked as an observation period.

[0048] In a preferred embodiment of the present invention, a scientific and flexible early warning mechanism is added to the continuous attention assessment system, which significantly improves the safety and effectiveness of the assessment process. When the system detects that two dimensions of data are deemed abnormal, it immediately initiates a high-frequency acquisition and analysis mode to track and monitor the abnormal dimension data in real time and intensively. During the observation period, if the abnormal dimension data recovers to a state consistent with the test subject's previous normal fluctuation range, it indicates that the test subject may only have experienced a temporary fluctuation in attention or been subjected to momentary interference. The system will automatically resume the normal testing process to avoid unnecessary test interruptions and ensure the continuity of the assessment process. However, if the abnormal data still fails to recover to normal after a preset time period, it indicates that the test subject's attention state has a persistent problem. The system will immediately trigger a stop test command to terminate the test process in a timely manner. This not only avoids the generation of invalid data but also prevents the test subject from affecting the accuracy of the test results due to excessive fatigue or poor condition, providing dynamic and reliable risk management protection for the entire assessment process.

[0049] In this embodiment of the invention, accurate and targeted anomaly judgment rules are set for data in each dimension: For answer response data, by comparing the dynamic average accuracy rate within a first preset time period with the corresponding value in the individual baseline, and combining the relationship between the total average, dynamic average, and standard deviation, while also considering the duration of the abnormal state, random fluctuations can be effectively eliminated, and abnormal answer responses caused by decreased attention can be accurately identified; eye movement data uses a preset value for visual concentration as the measurement standard, and once it falls below this value, it is determined that the child has inattention; brain activity data is compared with the individual baseline by the ratio of the difference, and when the brain activity data is significantly lower than the brain activity data recorded in the individual baseline, abnormal brain neural activity can be detected, and an abnormal attention state can be judged. These rules, starting from multiple dimensions such as cognitive behavior, visual attention, and brain neural activity, comprehensively cover data related to children's attention, providing a reliable basis for accurately judging children's attention abnormalities, and significantly improving the accuracy and credibility of the assessment results.

[0050] Please refer to Figure 5For example, suppose a continuous attention assessment is being conducted on a 7-year-old participant. After the test begins, the multi-dimensional data acquisition module works synchronously: in the visual question-and-answer section, the participant needs to identify and click on continuously changing graphics on the computer screen, and the system records their response data in real time; the EEG headband collects EEG signals from the Fp1 and Fp2 positions of the prefrontal cortex; and the 3D camera simultaneously acquires the participant's eye movement trajectory. At the 4-minute mark of the test, the system detects the following abnormal signals: in the response data, the dynamic accuracy rate of the past minute (corresponding to the first preset time period) is 76%, significantly lower than the average of 90% for the corresponding time period in the individual's baseline; simultaneously, the response variance for this period is more than 2.5 times the baseline variance, and this state has lasted for 31 seconds, exceeding the set abnormal trigger threshold (20 seconds), meeting the criteria for abnormal response data; eye movement data analysis shows that the participant's time spent focusing on the central area of ​​the screen in the past minute accounts for 50%, lower than the preset lower limit, and is judged as an eye movement abnormality; EEG activity data shows that the participant's signal strength in the target band is only 8.5 μV. 2 A value at / Hz significantly lower than the corresponding value in the individual's baseline is also considered abnormal. If two or more consecutive data dimensions are deemed abnormal, the system automatically enters a warning observation period. During this period, the system continuously monitors whether various indicators show a recovery trend. If the overall attenuation value still does not meet the recovery criteria by the end of the warning observation period, the system automatically terminates the assessment. Ultimately, the system assesses the participant's effective continuous attention span at approximately 4 minutes and 30 seconds and generates visualization charts and individualized assessment reports based on multi-dimensional data, providing quantitative evidence for subsequent educational or medical interventions.

[0051] In a preferred embodiment of the present invention, the continuous focus assessment system further includes a result presentation module, which is used to: generate a visualization chart based on the focus assessment results to display the decay trend and the final determined continuous focus duration value; and generate and display the focus assessment results.

[0052] The charts visually demonstrate the decline trend and changes in focus duration, while the report presents the assessment results in detail and provides targeted corrective suggestions, making the assessment results easier to understand and apply, and enhancing the practical value of the system.

[0053] In this embodiment of the invention, the baseline construction module is also used to compare the individual baseline with the attenuation curve baseline. When the difference between the data recorded on the individual baseline and the data recorded on the attenuation curve baseline is too large, the test is terminated.

[0054] In a preferred embodiment of the present invention, the step of extracting the decay feature values ​​of the individual curve using symmetric rotational equivariant convolution technology includes: constructing a model for extracting the decay feature values ​​of the individual curve. This model includes configuring a symmetric rotational equivariant convolution layer to extract the basic features of the rotational equivariance of the individual curve; configuring a multi-scale feature pyramid module to capture decay patterns at different time scales; and configuring a decay feature regression module to output the decay feature values. Based on a training dataset and using a three-stage training method, the model is trained to extract the decay feature values ​​of the individual curve using the trained model.

[0055] Symmetric rotation equivariant convolution, as a convolutional neural network technique, maintains equivariance under symmetric transformations such as rotation and reflection. For extracting decay features from time series data, it offers advantages such as insensitivity to curve phase shifts, automatic capture of multi-scale decay patterns, and reduced need for data augmentation. Please refer to [reference needed]. Figure 6 For example, a model for extracting decay feature values ​​of the individual curves may include an input layer (with 3 channels, for example); a symmetric rotation equivariant convolutional layer may be configured with 12 rotations and a kernel size of 15; a multi-scale feature pyramid module may be configured with 3 scales and an expansion rate of [1, 3, 5]; and a decay feature regression module may include a decay extractor that may include branch channels (e.g., 32) and an output layer with a feature dimension of 3, for example.

[0056] The symmetric rotational convolutional layer can be configured using the following formula:

[0057]

[0058] Where K is the rotational symmetry order, x represents the input curve, k represents the rotation index, rot() represents the rotation transformation operator, and w k Represents the convolution kernel of the science department. Indicates the rotation angle.

[0059] Furthermore, the multi-objective loss function can be expressed by the following formula:

[0060] L=λ1L rate +λ2L stability +λ3L shape

[0061] Where λ1, λ2, and λ3 represent weights, and L rate L represents the decay rate loss. stability L represents the stability loss. shape This represents the loss of morphological similarity.

[0062] In a preferred embodiment of the present invention, the model is trained based on a training dataset using a three-stage training method to extract the decay feature values ​​of the individual curves. For example, the first stage (pre-training stage) can synthesize decay curves using a physical model; the second stage (fine-tuning stage) can fine-tune the individual curves, progressively unfreezing them from the output layer to the input layer; the third stage (adaptive stage) uses domain adversarial training based on unlabeled individual curves. This three-stage training method can improve the accuracy of the model.

[0063] In a preferred embodiment of the present invention, the attenuation characteristic value of each dimension may include multiple indicators. For example, the attenuation characteristic value of prefrontal cortex EEG signal activity may include α, which indicates fatigue warning, β, which indicates attentional stability, and γ, which indicates morphological abnormalities.

[0064] This invention utilizes symmetric rotational isovariant convolution technology to significantly enhance the feature extraction capability of multidimensional data. Based on the invariance of data rotation and translation transformations, it effectively addresses changes in the EEG signal acquisition position caused by slight head movements during testing, stably extracting neural activity features and reducing environmental interference. By designing rotational isovariant filters of different scales, it achieves the fusion and capture of multi-scale features such as micro-fixation patterns and macro-saccade paths in eye-tracking data, comprehensively reflecting visual attention allocation. Through the construction of a multi-channel network architecture, it automatically learns the intrinsic correlations between different dimensions such as answer responses, EEG signals, and eye-tracking data, enabling collaborative analysis of multi-dimensional data. Simultaneously, by leveraging the symmetry prior of the data, it reduces the number of model parameters by more than 40% while maintaining assessment accuracy, significantly improving real-time processing capabilities and meeting dynamic assessment needs. This technology accurately captures attention decay characteristics, adapts to individual differences, and improves assessment efficiency. It is a key technology for achieving high-precision, robust continuous attention assessment, effectively ensuring the accuracy and reliability of the system's continuous attention assessment for children.

[0065] In a preferred embodiment of the present invention, the analysis module is further configured to: compare and analyze the extracted decay feature values ​​with the features of the baseline for each dimension of the data in the individual curve, and determine the abnormality of the data in each dimension in combination with the anomaly judgment rules; when the data in two or more dimensions are judged to be abnormal, and there is no reverse trend in the remaining dimensions, the attention state is judged to be decaying; when the attention state is judged to be decaying or the test time reaches the test time corresponding to the subject, the data collection is stopped.

[0066] This invention also provides a continuous attention assessment method, which may include steps S210-S240:

[0067] Step S210: Collect multidimensional test data according to a preset time interval and sequence. The multidimensional test data includes answer response data, EEG activity data, and eye movement data.

[0068] Step S220: Based on the collected partial multidimensional test data, construct a baseline including prefrontal cortex EEG signal activity, the proportion of time spent looking at the screen, and parameters related to answer responses. Then, use the sliding window technique to process the collected multidimensional test data and generate individual curves with the same characteristics as the baseline.

[0069] Step S230: When generating the individual curve, use the preset anomaly judgment rules to perform anomaly warning judgment.

[0070] Step S240: After the abnormal warning is triggered, the warning observation period is entered. The decay feature value of the individual curve is extracted by the symmetric rotation isovariant convolution technique. When the warning observation period ends and the decay feature value still shows an abnormality, the evaluation termination mechanism is started, and the duration from the start to the end of the test is used as the evaluation index of continuous focus duration.

[0071] This invention also provides a machine-readable storage medium storing instructions that cause a machine to perform the aforementioned continuous attention assessment method.

[0072] Furthermore, the terms "system" and "network" are often used interchangeably in this paper. The term "and / or" in this paper merely describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " in this paper generally indicates that the preceding and following related objects have an "or" relationship.

[0073] It should be understood that, in the embodiments of the present invention, "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0074] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0075] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0076] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, apparatuses, or units, or they may be electrical, mechanical, or other forms of connection.

[0077] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0078] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0079] From the above description of the embodiments, those skilled in the art will clearly understand that the present invention can be implemented in hardware, firmware, or a combination thereof. When implemented in software, the above-described functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media can be any available medium accessible to a computer. For example, but not limited to, computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible to a computer. Furthermore, any connection can suitably be a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the scope of the medium. As used in this invention, disks and discs include CDs, laser discs, optical discs, DVDs, floppy disks, and Blu-ray discs, wherein disks typically magnetically copy data, while discs optically copy data using lasers. The combinations described above should also be included within the scope of protection for computer-readable media.

[0080] In summary, the above description is merely a preferred embodiment of the technical solution of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A continuous attention assessment system based on multi-dimensional data, characterized in that, The continuous attention assessment system includes: The multi-dimensional data acquisition module is used to collect multi-dimensional test data according to a preset time interval and sequence. The multi-dimensional test data includes answer response data, brain activity data, and eye movement data. The curve construction module is used to construct a baseline based on the collected partial multidimensional test data, including the activity of prefrontal EEG signals, the proportion of time spent looking at the screen, and parameters related to the response to questions. It also uses sliding window technology to process the collected multidimensional test data and generate individual curves with the same characteristics as the baseline. The analysis module is used to perform anomaly warning judgment using preset anomaly judgment rules when generating the individual curves; The control module is used to enter the warning observation period after the abnormal warning is triggered. It extracts the decay feature value of the individual curve through the symmetric rotation isovariant convolution technique. When the warning observation period ends and the decay feature value still shows an abnormality, the evaluation termination mechanism is activated, and the duration from the start to the end of the test is used as the evaluation index of continuous focus time.

2. The continuous attention assessment system according to claim 1, characterized in that, The construction includes baselines for prefrontal cortex EEG signal activity, the percentage of time spent looking at the screen, and response parameters, including: After normalizing and correcting the collected multidimensional test data, features were extracted from prefrontal cortex EEG activity, the proportion of time spent looking at the screen, and parameters related to answer responses to obtain multimodal features; and The baseline is obtained by fusing the multimodal features using a weighted Gaussian mixture model.

3. The continuous attention assessment system according to claim 1, characterized in that, The process of using sliding window technology to process the collected multidimensional test data and generate individual curves with the same characteristics as the baseline includes: Initialize the sliding window and its parameter configuration according to the preset step size; Within the current sliding window, features of prefrontal cortex EEG signal activity, the proportion of time spent looking at the screen, and parameters related to answer responses are extracted to obtain multimodal features; After fusing the multimodal features and performing time axis alignment and missing value processing, the individual curve of the current sliding window is generated; Perform the window closing procedure.

4. The continuous attention assessment system according to claim 1, characterized in that, When generating the individual curve, anomaly detection is performed using preset anomaly detection rules, including: If the parameter features related to the answer response show that the dynamic average accuracy rate within a first preset time period is less than the corresponding feature in the baseline, and the variance of the answer response time is greater than a multiple of the dynamic variance of the data in that dimension within a first preset time period, and the duration exceeds a preset abnormal time, then the indicator of the answer response status dimension is abnormal. If the percentage of time the eyes spend looking at the screen indicates that visual concentration is lower than a preset value, then the indicators of the eye movement state dimension are abnormal. If the prefrontal cortex EEG signal activity features show a lower than preset level than the corresponding features in the baseline, then the index of the EEG activity state dimension is abnormal.

5. The continuous attention assessment system according to claim 1, characterized in that, When generating the individual curve, anomaly detection is performed using preset anomaly detection rules, including: The data of each dimension in the individual curve is compared and analyzed with the characteristics of the baseline, and the anomaly judgment rules are combined to determine the anomaly of the data in each dimension. When more than two dimensions of indicators are judged as abnormal, it is determined that the attention state is declining and marked as a warning observation period.

6. The continuous attention assessment system according to claim 1, characterized in that, The step of extracting the decay feature values ​​of the individual curves using symmetric rotational isovariant convolution technology includes: A model is constructed to extract the decay feature values ​​of the individual curves. Specifically, a symmetric rotational isovariant convolutional layer is configured to extract the fundamental features of the rotational isovariance of the individual curves. Configure a multi-scale feature pyramid module to capture decay patterns at different time scales. Configure the attenuation feature regression module to output the attenuation feature values; The model is trained using a three-stage training method based on the training dataset, and the trained model is used to extract the decay feature values ​​of the individual curves.

7. The continuous attention assessment system according to claim 1, characterized in that, The control module is also used for: If, during the warning observation period, the attenuation characteristic value shows a return to the normal fluctuation range of the previous attenuation characteristic value, the evaluation process continues.

8. The continuous attention assessment system according to claim 1, characterized in that, The continuous attention assessment system also includes a result presentation module, which is used for: Based on the focus assessment results, a visual chart is generated to show the trend of decay and the final value of continuous focus duration. Generate focus assessment results.

9. A method for assessing continuous attention span, characterized in that, The continuous attention assessment method includes: Multidimensional test data is collected according to a preset time interval and sequence. The multidimensional test data includes answer response data, brain activity data, and eye movement data. Based on the collected multidimensional test data, a baseline was constructed including the activity of prefrontal cortex EEG signals, the proportion of time spent looking at the screen, and parameters related to the response to questions. The collected multidimensional test data was then processed using a sliding window technique to generate individual curves with the same characteristics as the baseline. When generating the individual curves, anomaly warnings are made using preset anomaly judgment rules. After an anomaly warning is triggered, an early warning observation period is entered. The decay feature value of the individual curve is extracted using the symmetric rotation isovariant convolution technique. When the early warning observation period ends and the decay feature value still shows an anomaly, the evaluation termination mechanism is activated, and the duration from the start to the end of the test is used as the evaluation index of continuous focus duration.

10. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions that cause the machine to perform the continuous attention assessment method according to claim 9.

Images