ADHD child cognitive training system

Abstract

The invention relates to the technical field of child cognitive training, and discloses an ADHD child cognitive training system. The system comprises a multi-modal sensing module, an attention evaluation module, a dynamic task generation module, a behavior feedback module and a self-adaptive adjustment module. The multi-modal perception module collects multi-dimensional behavior data, the attention evaluation module extracts attention fluctuation features, the dynamic task generation module generates a cognitive training task sequence, the behavior feedback module performs emotion state classification processing to generate feedback signals, and the self-adaptive adjustment module generates personalized training difficulty steps. According to the system and the method, through comprehensive data acquisition, precise attention assessment, dynamic task generation, real-time emotion feedback and difficulty adaptive adjustment, personalized and precise cognitive training of ADHD children is realized, the training effect is effectively improved, and the problems of lack of personalization, insufficient state monitoring, unscientific difficulty adjustment and the like of an existing training method are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of children's cognitive training, and specifically to a cognitive training system for ADHD children. Background Art

[0002] Attention Deficit Hyperactivity Disorder (ADHD) is a common neurodevelopmental disorder with a relatively high incidence rate among children. Children with ADHD usually have obvious deficiencies in aspects such as attention, behavior control, and learning ability. These problems seriously affect their academic performance, social development, and mental health, and also pose great challenges to their future life and career development.

[0003] Currently, the treatment methods for ADHD children mainly include drug treatment and behavioral intervention. Although drug treatment can relieve symptoms to a certain extent, there are risks of side effects, and long-term use may affect children's growth and physical health. For example, it may cause adverse reactions such as loss of appetite, insomnia, and dizziness. Moreover, drug treatment cannot fundamentally solve the problem of cognitive function defects, and the symptoms are likely to recur after stopping the drug.

[0004] Behavioral intervention is an important part of ADHD treatment, and cognitive training is a key link. However, there are many deficiencies in existing cognitive training methods. Most traditional cognitive training programs adopt standardized training content and fixed training difficulties, lacking personalized customization. Since the symptom severity, cognitive ability basis, and learning characteristics of each ADHD child are different, this "one-size-fits-all" training method is difficult to meet their special needs, resulting in a significant reduction in training effects. Some training programs may be too simple for some children to effectively improve their cognitive abilities; while for other children, they may be too difficult, easily causing them to have a sense of frustration and reducing their enthusiasm for participating in training.

[0005] In addition, existing cognitive training systems also have deficiencies in the monitoring and evaluation of the training process for children. They can often only obtain data in a single dimension, such as simple answer correct rates or reaction times, and cannot comprehensively and accurately understand information such as children's attention states and emotional changes during training. Without multi-dimensional data support, it is difficult to scientifically evaluate the training effects, and it is also impossible to adjust training strategies in a timely manner, making the training process lack pertinence and effectiveness. Moreover, the current training systems lack flexibility in task generation, cannot dynamically adjust task difficulties and types according to children's real-time performance, and cannot adapt to the development and changes of children's cognitive abilities, restricting the improvement of cognitive training effects.

[0006] With the continuous development of technology, how to develop a more efficient and personalized cognitive training system for children with ADHD using advanced technical means has become an urgent problem in this field. It needs to be able to accurately collect multi-dimensional behavioral data of children, evaluate the training effect in real time, and dynamically adjust the difficulty of training tasks to improve the cognitive training effect of children with ADHD, help them better improve their symptoms, and enhance the quality of life and learning. Summary of the Invention

[0007] The purpose of the present invention is to provide a cognitive training system for children with ADHD to solve the problems raised in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solution: A cognitive training system for children with ADHD, the system includes:

[0009] Multi-modal perception module: used to collect the multi-dimensional behavioral data set of children; the multi-dimensional behavioral data includes eye movement trajectory data, touch operation sequence, and physiological signal data;

[0010] Attention evaluation module: Based on the eye movement trajectory data, extract the attention fluctuation characteristics through the distraction factor calculation algorithm, and the fluctuation characteristics include the frequency of distraction events, the duration of attention concentration, and the focus transfer rate;

[0011] Dynamic task generation module: used to generate a cognitive training task sequence according to the touch operation sequence through the task complexity adaptive algorithm, and the task sequence includes the graphic matching level, the reaction time threshold, and the interference item density parameter;

[0012] Behavior feedback module: perform real-time emotional state classification processing on the physiological signal data to generate a multi-modal feedback signal; the emotional state classification processing includes heart rate variability analysis and skin conductance peak detection;

[0013] Adaptive adjustment module: fuse the attention fluctuation characteristics, the cognitive training task sequence, and the multi-modal feedback signal through the incremental difficulty control algorithm to generate a personalized training difficulty ladder.

[0014] Preferably, the distraction factor calculation algorithm includes:

[0015] Detect distraction events in the eye movement trajectory data and mark the segments where the line of sight deviation exceeds the preset threshold;

[0016] Statistically analyze the total duration and interval distribution of distraction events, and calculate the proportion of attention concentration per unit time;

[0017] Analyze the stable time after focus transfer based on the attention recovery rate model to generate a distraction factor score;

[0018] Compare the distraction factor score with a preset fluctuation threshold, and output an attention fluctuation feature vector.

[0019] Preferably, the task complexity adaptive algorithm includes:

[0020] Conduct reaction delay analysis on the touch operation sequence, extract the time window of effective operations, and dynamically adjust the number of graphic matching levels according to the accuracy distribution within the time window;

[0021] Combine the correlation between the interference item density parameter and the reaction time threshold to generate a gradient task sequence;

[0022] Map the task parameters to a preset complexity interval through normalization processing, and output a standardized task sequence.

[0023] Preferably, the emotional state classification process includes:

[0024] Perform sliding window segmentation on the physiological signal data, extract the heart rate variability spectrum features, and identify the stress response segments in the skin conductance signal through a peak detection algorithm;

[0025] Perform feature splicing on the spectrum features and the stress response segments, input them into a pre-trained emotion classification model, and output a multi-modal feedback signal including an anxiety index and a concentration score.

[0026] Preferably, the incremental difficulty adjustment algorithm includes:

[0027] Perform time series difference calculation on the attention fluctuation features to extract the attention change trend;

[0028] Dynamically adjust the interference item density of the task sequence according to the anxiety index in the multi-modal feedback signal;

[0029] Perform weighted fusion on the attention change trend and the interference item density adjustment amount to generate a difficulty ladder parameter;

[0030] Stack the parameters to the historical training record through a ladder update strategy, and output a personalized training difficulty ladder.

[0031] Preferably, the construction method of the attention recovery rate model includes: dividing the attention recovery stage according to historical eye movement data, counting the focus stability duration within each stage, establishing a correlation function between the distraction event duration and the recovery duration through regression analysis, and dynamically updating the calculation weight of the distraction factor score based on the correlation function to optimize the accuracy of fluctuation feature extraction.

[0032] Preferably, the generation steps of the gradient task sequence include:

[0033] Calculate the fitness score of the current task level according to the accuracy rate distribution, and select candidate tasks that match the level from the preset task library through the probability sampling algorithm;

[0034] Generate an interference intensity gradient table based on the interference item density parameter of the candidate task, and generate a multi-dimensional task sequence by combining the gradient table with the reaction time threshold.

[0035] Preferably, the training method of the pre-trained emotion classification model includes:

[0036] Collect a physiological signal dataset with labeled emotion states, and perform time-frequency feature extraction and data augmentation;

[0037] Build a convolutional neural network model, and input the time-frequency feature map and peak detection results;

[0038] Optimize the model parameters through the cross-entropy loss function, and output the emotion classification probability distribution.

[0039] Preferably, the step-by-step update strategy includes: generating a version identifier according to the timestamp of the difficulty step parameter, identifying the deviation interval between the current parameter and the historical record through the difference comparison algorithm, and superimposing and updating the parameters in the deviation interval according to the weight decay coefficient.

[0040] Preferably, the present invention also includes an ADHD children's cognitive training method, and the method includes:

[0041] Collect a multi-dimensional behavior data set of children; the data includes eye movement trajectory data, touch operation sequences, and physiological signal data;

[0042] Extract the attention fluctuation characteristics based on the distraction factor calculation algorithm;

[0043] Generate a cognitive training task sequence through the task complexity adaptive algorithm;

[0044] Perform real-time emotion state classification processing on the physiological signal data to generate multi-modal feedback signals;

[0045] Adopt an incremental difficulty regulation algorithm to fuse the fluctuation characteristics, task sequence, and feedback signals to construct a personalized training difficulty ladder;

[0046] Generate a training task instruction sequence according to the difficulty ladder through the training execution engine, and output it to the interaction terminal.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] In terms of personalized customization of training, the system collects the eye movement trajectory data, touch operation sequences, and physiological signal data of children through a multimodal perception module to comprehensively obtain the behavior information of children during the training process. The attention assessment module extracts attention fluctuation features based on the eye movement trajectory data, the dynamic task generation module generates a cognitive training task sequence according to the touch operation sequences, the behavior feedback module generates multimodal feedback signals based on the physiological signal data, and the adaptive adjustment module integrates this information and uses an incremental difficulty regulation algorithm to generate a personalized training difficulty ladder. This means that the training difficulty can be adjusted in real time according to the specific situation of each child. For example, for children with a shorter attention concentration duration and who are easily distracted, the initial training task difficulty is appropriately reduced, and the difficulty is gradually increased according to the attention change trend and emotional feedback during their training process, avoiding children having a sense of frustration due to overly difficult tasks or being unable to effectively improve their cognitive abilities due to overly easy tasks, greatly improving the pertinence and adaptability of the training.

[0049] In terms of the accuracy of training effect evaluation, the multi-dimensional behavior data collected by the system provides strong support for comprehensively evaluating the training effect. Traditional training methods only evaluate based on simple indicators and it is difficult to accurately grasp the training status of children. In this system, the attention assessment module precisely analyzes fluctuation features such as the frequency of distraction events, attention concentration duration, and focus transfer rate through a distraction factor calculation algorithm; the behavior feedback module performs real-time emotional state classification processing on the physiological signal data to obtain information such as anxiety index and concentration score. These multi-dimensional data can more comprehensively and accurately reflect the cognitive and emotional changes of children during the training process, enabling trainers and parents to better understand the training progress of children and providing a scientific basis for adjusting subsequent training strategies.

[0050] In terms of the flexibility of training task generation, the task complexity adaptive algorithm of the dynamic task generation module plays a key role. This algorithm analyzes the reaction delay of the touch operation sequences, dynamically adjusts the number of graphic matching levels according to the correct rate distribution, combines the interference item density parameter and the reaction time threshold to generate a gradient task sequence, and outputs a standardized task sequence through normalization processing. This enables the training tasks to dynamically change according to the real-time operation performance of children. When children perform well at a certain difficulty level, the system automatically increases the task difficulty and provides more challenging training content; when children encounter difficulties, the difficulty is promptly reduced to ensure the coherence and effectiveness of the training and continuously stimulate children's learning motivation and participation.

[0051] In terms of enhancing children's cognitive abilities, the comprehensive training mode of this system has achieved remarkable results. Through continuous personalized cognitive training, it can effectively improve the attention deficit problems of children with ADHD, extend the duration of concentrated attention, and reduce the frequency of distraction events. At the same time, with the dynamic adjustment of the training task difficulty and the training of diverse tasks, children's cognitive skills such as reaction ability and graphic matching ability can also be gradually improved. Long-term use of this training system helps to improve the academic performance of children with ADHD, enhance their self-confidence and adaptability in social activities, promote their mental health development, and lay a solid foundation for their better integration into society in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is the working principle diagram of the ADHD children's cognitive training system described in the present invention;

[0053] Figure 2 is the flowchart of the distraction factor calculation algorithm;

[0054] Figure 3 is the flowchart of the task complexity adaptive algorithm;

[0055] Figure 4 is the flowchart of the incremental difficulty regulation algorithm. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0057] Please refer to Figures 1-4 , the present invention provides an ADHD children's cognitive training system, and its overall implementation solution is as follows:

[0058] This ADHD children's cognitive training system mainly consists of a multi-modal perception module, an attention evaluation module, a dynamic task generation module, a behavior feedback module, and an adaptive adjustment module that work together to achieve personalized cognitive training for children with ADHD.

[0059] The multi-modal perception module is responsible for collecting the multi-dimensional behavior data set of children during the training process, and these data are the basic information source of the entire training system. Among them, the eye movement trajectory data is obtained through a professional eye tracking device, which can accurately record the eye movement trajectory of children when they are looking at the training content; the touch operation sequence is recorded through the touch screen on the training device, recording the operation sequence of children interacting with the training interface; the physiological signal data is collected by physiological sensors worn on children, such as heart rate sensors, skin conductance sensors, etc., for monitoring the changes in the physiological state of children during training.

[0060] Based on the eye movement trajectory data collected by the multi-modal perception module, the attention evaluation module uses the distraction factor calculation algorithm to extract the attention fluctuation characteristics. By analyzing the eye movement trajectory, the segments where the line of sight deviation exceeds the preset threshold are marked to determine the distraction events. Furthermore, the total duration and interval distribution of the distraction events are counted, and the proportion of attention concentration per unit time is calculated. At the same time, according to the attention recovery rate model, the stable time after the focus transfer is analyzed, and these information are synthesized to generate the distraction factor score. By comparing this score with the preset fluctuation threshold, the attention fluctuation characteristics including the frequency of distraction events, the duration of attention concentration, and the focus transfer rate are finally output to evaluate the attention state of children.

[0061] Based on the touch operation sequence collected by the multi-modal perception module, the dynamic task generation module uses the task complexity adaptive algorithm to generate the cognitive training task sequence. First, the reaction delay of the touch operation sequence is analyzed to extract the time window of effective operations, and the number of graphic matching levels is dynamically adjusted according to the correct rate distribution within this time window. Then, combining the correlation between the interference item density parameter and the reaction time threshold, a gradient task sequence is generated. Finally, through normalization processing, the task parameters are mapped to the preset complexity interval, and a standardized cognitive training task sequence is output, so that the difficulty of the training task can be dynamically adjusted according to the operation situation of children.

[0062] The behavior feedback module performs real-time emotion state classification processing on the physiological signal data collected by the multi-modal perception module to generate multi-modal feedback signals. First, the physiological signal data is segmented by a sliding window to extract the heart rate variability spectrum characteristics, and at the same time, the stress response segments in the skin conductance signal are identified through the peak detection algorithm. Then, the spectrum characteristics and the stress response segments are feature spliced and input into the pre-trained emotion classification model to output multi-modal feedback signals including the anxiety index and the concentration score, so as to timely feedback the emotion state of children during the training process.

[0063] The adaptive adjustment module generates a personalized training difficulty ladder by integrating attention fluctuation features, cognitive training task sequences, and multimodal feedback signals through an incremental difficulty regulation algorithm. First, perform time series differencing calculation on the attention fluctuation features to extract the attention change trend. Then, dynamically adjust the interference item density of the task sequence according to the anxiety index in the multimodal feedback signal. Next, perform weighted fusion on the attention change trend and the adjusted amount of interference item density to generate difficulty ladder parameters. Finally, use the ladder update strategy to superimpose the parameters on the historical training records and output a personalized training difficulty ladder suitable for each child, enabling the training difficulty to be adjusted in real time according to the child's training progress and status.

[0064] The following further illustrates the implementation of the present invention in combination with Examples 1 to 6.

[0065] Example 1:

[0066] This example elaborates in detail the specific implementation method of the distraction factor calculation algorithm in the attention assessment module.

[0067] In the attention assessment module, detecting distraction events in the eye movement trajectory data is a key step. Set a preset threshold, which is determined based on a large amount of experimental data and professional experience in the study of the attention of ADHD children, and is used to measure whether the child's line of sight has a significant deviation. When a certain segment of the line of sight deviation in the eye movement trajectory data exceeds this preset threshold, mark this segment as a distraction event segment.

[0068] Statistical the total duration of distraction events, that is, add up the time lengths of all marked distraction event segments. At the same time, record the interval time between each distraction event to form an interval distribution. Quantify the attention concentration degree by calculating the proportion of the duration of attention concentration within a unit time to the total time. For example, within a 10-minute training time, the total duration of distraction events is 2 minutes, then the proportion of attention concentration within a unit time is (10 - 2) ÷ 10 = 80%.

[0069] Analyze the stabilization time after focus transfer based on the attention recovery rate model. The attention recovery rate model is established based on historical eye movement data. By dividing the attention recovery stage, statistically the focus stabilization duration within each stage, and using regression analysis to establish the correlation function between the distraction event duration and the recovery duration. Assume the correlation function is y = ax + b, where y represents the recovery duration, x represents the distraction event duration, and a and b are coefficients obtained through regression analysis. According to this correlation function, dynamically update the calculation weight of the distraction factor score. For example, if the duration of a distraction event is x1 and the recovery duration calculated through the correlation function is y1, then when calculating the distraction factor score, the relevant weight will be adjusted according to the size of y1, so that the distraction factor score can more accurately reflect the attention fluctuation situation.

[0070] When generating the distraction factor score, factors such as the total duration of distraction events, interval distribution, the proportion of time with concentrated attention per unit time, and the stable time after focus transfer are comprehensively considered. The distraction factor score is compared with a preset fluctuation threshold, which is also set based on a large amount of experimental data and professional research. If the distraction factor score is higher than the preset fluctuation threshold, it indicates that the child's attention fluctuates greatly; otherwise, it indicates that the attention is relatively stable. Finally, an attention fluctuation feature vector is output, which contains information such as the frequency of distraction events, the duration of concentrated attention, and the focus transfer rate, providing an important basis for subsequent training difficulty adjustment and personalized training.

[0071] Example 2:

[0072] This example details the specific operation process of the task complexity adaptive algorithm in the dynamic task generation module.

[0073] In the dynamic task generation module, reaction delay analysis is performed on the touch operation sequence. First, the time window of valid operations is determined, which can be determined by analyzing the touch operation data of a large number of normal children and ADHD children to determine a reasonable time range. For example, through data analysis, it is found that the operations made by most children within 0.5 seconds to 3 seconds after seeing the training task are valid operations, so this time range is set as the time window of valid operations.

[0074] Within the time window of valid operations, the correct rate distribution of the children's operations is counted. Suppose there are 10 questions in a graphic matching task during a training session, and within the time window of valid operations, the child answers 6 questions correctly, then the correct rate is 60%. The number of graphic matching levels is dynamically adjusted according to the correct rate distribution. If the correct rate is relatively high, for example, more than 80%, it indicates that the current graphic matching level may be too simple for the child, and the number of graphic matching levels can be appropriately increased to increase the task difficulty; if the correct rate is relatively low, such as less than 40%, the number of graphic matching levels is appropriately reduced to lower the task difficulty.

[0075] A gradient task sequence is generated by combining the relevance between the distraction item density parameter and the reaction time threshold. The distraction item density parameter represents the proportion of the number of distraction elements in the training task, and the reaction time threshold refers to the maximum time allowed for the child to complete the task. Generally, the higher the distraction item density, the longer the reaction time threshold will be accordingly. For example, when the distraction item density is 20%, the reaction time threshold is set to 3 seconds; when the distraction item density increases to 40%, the reaction time threshold is extended to 5 seconds. Through this relevance, according to the current task requirements and the actual situation of the child, a task sequence with different difficulty gradients is generated.

[0076] The task parameters are mapped to a preset complexity interval through normalization. Assume the preset complexity interval is [0, 1], and the task parameters include the number of levels of graphic matching, the interference item density parameter, the reaction time threshold, etc. The normalization formula is used:

[0077]

[0078] where x is the original task parameter, x min and x max are respectively the minimum and maximum values of this parameter among all possible values, and x norm is the normalized parameter. After normalization, all task parameters are mapped into the interval [0, 1], thereby outputting a standardized task sequence, ensuring the comparability of the difficulty between different tasks, and facilitating the provision of training tasks with appropriate difficulty for children.

[0079] Example 3:

[0080] This example focuses on introducing the specific implementation details of emotional state classification processing in the behavior feedback module.

[0081] In the behavior feedback module, the physiological signal data is segmented by a sliding window. The size and step length of the sliding window are determined according to the characteristics of the physiological signal and relevant research. For example, for the heart rate signal, a 1-second sliding window is selected with a step length of 0.1 second; for the skin conductance signal, a 2-second sliding window is selected with a step length of 0.2 second. Through sliding window segmentation, the continuous physiological signal data can be divided into multiple small segments, facilitating subsequent feature extraction.

[0082] Extract the spectral features of heart rate variability. Heart rate variability refers to the minute differences between successive inter-beat intervals, reflecting the activity of the cardiac autonomic nervous system. Methods such as Fourier transform are used to process the heart rate data within the sliding window to obtain the spectral features of heart rate variability. These features contain information on the energy distribution of the heart rate at different frequency components. For example, the low-frequency component reflects the combined action of the sympathetic and vagus nerves, and the high-frequency component mainly reflects the activity of the vagus nerve.

[0083] Identify the stress response segments in the skin conductance signal through a peak detection algorithm. The skin conductance signal fluctuates with the emotional changes of the human body. When a person is in a stress state, the skin conductance signal will show peaks. A suitable peak detection algorithm is used, such as a threshold-based peak detection algorithm, and a reasonable threshold is set. When the value of the skin conductance signal exceeds the threshold and shows a trend of rising first and then falling within a certain time range, this segment is identified as a stress response segment.

[0084] Perform feature splicing on the spectral features and stress response segments. Connect the extracted heart rate variability spectral features and the identified skin conductance stress response segments in a certain order to form a comprehensive feature vector. For example, first arrange the values of each frequency component of the heart rate variability spectral features in sequence, and then follow with the relevant features of the skin conductance stress response segments, such as peak size, occurrence time, etc.

[0085] Input the comprehensive feature vector into a pre-trained emotion classification model. The pre-trained emotion classification model can adopt deep learning models such as convolutional neural networks (CNNs). When training this model, collect a large physiological signal dataset with labeled emotion states, perform time-frequency feature extraction and data augmentation. For example, by performing operations such as translation and scaling on the original data, increase the diversity of the data and improve the generalization ability of the model. Input the time-frequency feature maps and peak detection results into the constructed CNN model, and optimize the model parameters through the cross-entropy loss function, so that the model can accurately output multi-modal feedback signals including anxiety index and concentration score. For example, the range of the anxiety index output by the model is [0,1], where 0 means not anxious at all and 1 means extremely anxious; the range of the concentration score is [0,100], and the higher the score, the higher the concentration, thus providing a basis for real-time feedback during the training process.

[0086] Example 4:

[0087] This example details the specific execution process of the incremental difficulty regulation algorithm in the adaptive adjustment module.

[0088] In the adaptive adjustment module, perform time series difference calculation on the attention fluctuation features. Assume that the attention fluctuation features are time series data. For example, in consecutive multiple training time periods, the attention concentration durations t1, t2, t3,... are recorded respectively. Adopt the difference calculation method to calculate the differences in attention concentration durations between adjacent time periods, such as Δt1 = t2 - t1, Δt2 = t3 - t2, etc. Through these differences, extract the attention change trend. If the value of Δt is continuously positive, it indicates that the attention concentration duration is gradually increasing and the attention is on an upward trend; conversely, if the value of Δt is continuously negative, it indicates that the attention is on a downward trend.

[0089] Dynamically adjust the interference item density of the task sequence according to the anxiety index in the multimodal feedback signal. The anxiety index is output by the behavioral feedback module and reflects the anxiety level of the child during the training process. When the anxiety index is relatively high, it indicates that the child may have anxiety due to excessive task difficulty or other reasons. At this time, appropriately reduce the interference item density of the task sequence to relieve the child's cognitive burden. For example, if the original interference item density is 40%, when the anxiety index exceeds 0.6, reduce the interference item density to 30%. On the contrary, when the anxiety index is relatively low and the attention concentration duration is stable, the interference item density can be appropriately increased to improve the task difficulty.

[0090] Weightedly fuse the attention change trend and the interference item density adjustment amount to generate a difficulty ladder parameter. Set two weight coefficients w1 and w2. w1 is used to measure the importance of the attention change trend, and w2 is used to measure the importance of the interference item density adjustment amount, and w1 + w2 = 1. Assume that the value of the attention change trend is A (A being a positive number indicates an increase in attention, and a negative number indicates a decrease in attention), and the interference item density adjustment amount is B (B being a positive number indicates an increase in the interference item density, and a negative number indicates a decrease in the interference item density). The difficulty ladder parameter D = w1A + w2B. By reasonably adjusting the weight coefficients w1 and w2, the difficulty ladder parameter can more accurately reflect the child's training situation and needs.

[0091] Use the ladder update strategy to superimpose the parameter onto the historical training record and output a personalized training difficulty ladder. The ladder update strategy generates a version identifier based on the timestamp of the difficulty ladder parameter. For example, each time a difficulty ladder parameter is generated, record the generated time and use the time as part of the version identifier. Use the difference comparison algorithm to identify the deviation interval between the current parameter and the historical record. If there is a difference between the current difficulty ladder parameter and the parameter in the historical record, update the parameters in the deviation interval by weighted attenuation. Assume that the weighted attenuation coefficient is k (0 < k < 1), and the difficulty ladder parameter in the historical record is D old , the currently calculated difficulty ladder parameter is D new , the updated difficulty ladder parameter D final = D old + k(D new - D old ). Through continuous updating, output a personalized training difficulty ladder suitable for each child, enabling the training difficulty to be adjusted in real time according to the child's training progress and status.

[0092] Example 5:

[0093] This example details the construction method of the attention recovery rate model and its application in the distraction factor calculation algorithm.

[0094] First, collect a large amount of historical eye movement data, which are the eye movement records of different ADHD children in various cognitive training scenarios. Analyze these historical eye movement data to divide the attention restoration stage. In the eye movement data, when a distraction event (the line of sight deviates beyond the preset threshold) occurs, the period from the end of the distraction event to the moment when the eyes fixate on the target stably again is defined as the attention restoration stage.

[0095] Statistically analyze the focus stability duration within each attention restoration stage. For example, in one training session, 5 distraction events occur. Respectively, statistically analyze the focus stability durations t1, t2, t3, t4, t5 after each distraction event. Meanwhile, record the durations T1, T2, T3, T4, T5 of each distraction event.

[0096] Establish an association function between the distraction event duration and the restoration duration through regression analysis. Adopt linear regression or non-linear regression methods to find the mathematical relationship between the distraction event duration T and the restoration duration t. Assume that the association function obtained through regression analysis is t = aT + b, where a and b are regression coefficients. These coefficients are obtained by fitting a large amount of historical data and reflect the internal relationship between the distraction event duration and the restoration duration.

[0097] Dynamically update the calculation weights of the distraction factor scores based on the association function. In the distraction factor calculation algorithm, the distraction event duration and the restoration duration are important influencing factors. According to the association function t = aT + b, when the distraction event duration T changes, the corresponding restoration duration t can be predicted. For example, if the duration of a distraction event is T0, the restoration duration is calculated as t0 through the association function. When calculating the distraction factor score, for the case of a longer restoration duration, appropriately reduce the influence weight of this distraction event on the distraction factor score; for the case of a shorter restoration duration, appropriately increase the influence weight. This can more accurately reflect the attention fluctuation characteristics, optimize the accuracy of fluctuation feature extraction, make the distraction factor score more accurately evaluate the attention state of children, and provide a more reliable basis for subsequent training adjustment.

[0098] Example 6:

[0099] This example elaborates in detail the generation steps of the gradient task sequence and the training method of the pre-trained emotion classification model and the specific operation of the step-by-step update strategy.

[0100] When generating the gradient task sequence, calculate the fitness score of the current task level according to the correct rate distribution. Assume that in a graphic matching task training, a child completed n questions at a certain task level and answered m questions correctly, and the correct rate is Set the fitness scoring function as S = f(p). For example, S = 2p - 0.5 (the function here is only an example, and it can be designed according to specific situations in practice). Calculate the fitness score of the current task level through this function. The higher the score, the better the child's fitness for this task level.

[0101] Select candidate tasks at the matching level from the preset task library through the probability sampling algorithm. A large number of training tasks with different difficulty levels and types are stored in the preset task library. Set different sampling probabilities according to the fitness score. For task levels with a high fitness score, the probability of selecting higher-difficulty tasks from the preset task library is greater; for task levels with a low fitness score, the probability of selecting lower-difficulty tasks is greater. For example, when the fitness score is greater than 0.8, the sampling probability from high-difficulty tasks is 0.8; when the fitness score is less than 0.4, the sampling probability from low-difficulty tasks is 0.9.

[0102] Generate an interference intensity gradient table based on the interference item density parameter of the candidate task. Assume that the interference item density parameter of the candidate task is d. Based on d, generate an interference intensity gradient table according to certain rules. For example, when d = 0.3, the generated interference intensity gradient table may be [0.25, 0.3, 0.35], indicating that in subsequent tasks, tasks with different interference intensities can be selected according to needs.

[0103] Generate a multi-dimensional task sequence by combining the gradient table and the reaction time threshold. Generate a multi-dimensional task sequence containing different interference intensities and reaction time requirements according to the correlation between the interference intensity gradient table and the reaction time threshold. For example, when the interference intensity is 0.25, the reaction time threshold is set to 4 seconds; when the interference intensity is 0.35, the reaction time threshold is set to 3 seconds. In this way, generate a task sequence with different difficulty gradients to meet the training needs of children at different stages.

[0104] When training the pre-trained emotion classification model, collect a physiological signal data set with labeled emotion states. Through professional physiological signal acquisition equipment, collect a large amount of physiological signal data of ADHD children in different emotion states, such as heart rate, skin conductance, etc. At the same time, label these data with emotion states, such as labeled as anxious, focused, relaxed, etc.

[0105] Extract time-frequency features and perform data augmentation on the collected data. Use methods such as Fourier transform and wavelet transform to extract the time-frequency features of physiological signals, convert the time-domain signals to the frequency domain, and obtain richer feature information. At the same time, through data augmentation techniques, such as performing operations such as translation, rotation, and scaling on the original data, increase the diversity of the data and improve the generalization ability of the model.

[0106] Build a convolutional neural network model and input the time-frequency feature map and peak detection results. The convolutional neural network model has powerful feature extraction capabilities and can automatically learn the features in the data. Take the extracted time-frequency feature map and the results obtained through the peak detection algorithm as the input of the model. After a series of network layers such as the convolutional layer, pooling layer, and fully connected layer, perform feature extraction and classification on the input data. During the training process, optimize the model parameters through the cross-entropy loss function. The formula for the cross-entropy loss function is:

[0107]

[0108] where L represents the loss value, C is the number of categories (in this case, the number of emotion categories such as anxiety, concentration, relaxation, etc.), y i represents the value of the true label of the sample in the i-th category (if the sample belongs to the i-th category, then y i = 1, otherwise y i = 0), and p i is the probability that the model predicts the sample belongs to the i-th category. By continuously adjusting the model parameters, minimize the loss value, so that the model can accurately output the emotion classification probability distribution, that is, obtain a multi-modal feedback signal including the anxiety index, concentration score, etc.

[0109] When implementing the step update strategy, first generate a version identifier based on the timestamp of the difficulty step parameter. Each time the difficulty step parameter is generated, the system records the specific time of generation and uses this time as the key part of the version identifier, such as the timestamp accurate to the second "20241010153020". In this way, each version of the difficulty step parameter has a unique identifier, which is convenient for subsequent management and traceability.

[0110] Next, use the difference comparison algorithm to identify the deviation interval between the current parameter and the historical record. Compare the currently calculated difficulty step parameter with the parameters of each version in the historical record to determine the changed part and its range of the parameter. For example, compare the interference term density parameter in the current difficulty step parameter with the corresponding parameter in the historical version and find that the current interference term density has increased from 0.3 to 0.35, which is a deviation interval.

[0111] Finally, superimpose and update the parameters in the deviation interval according to the weight decay coefficient. Let the weight decay coefficient be λ (0 < λ < 1), assume that the difficulty step parameter vector in the historical record is the currently calculated parameter vector is the updated parameter vector is calculated as follows: for parameter i in the deviation interval, n i = h i + λ(c i - hi ) For the parameter j, n without deviation j = h j . For example, if λ = 0.8, the historical interference term density parameter h1 = 0.3, and the current interference term density parameter c1 = 0.35, then the updated interference term density parameter n1 = 0.3 + 0.8×(0.35 - 0.3) = 0.34. In this way, the updated parameters are superimposed on the historical training records, so as to output a personalized training difficulty ladder that better conforms to the child's current training situation, and promote the continuous optimization of the training process.

[0112] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0113] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An ADHD children's cognitive training system, characterized in that, Comprising: A multi-modal perception module: used to collect a multi-dimensional behavior data set of children; The multi-dimensional behavior data includes eye movement trajectory data, touch operation sequences, and physiological signal data; An attention evaluation module: based on the eye movement trajectory data, extracting attention fluctuation features through a distraction factor calculation algorithm, where the fluctuation features include the frequency of distraction events, the duration of attention concentration, and the focus transfer rate; A dynamic task generation module: used to generate a cognitive training task sequence through a task complexity adaptive algorithm according to the touch operation sequence, where the task sequence includes a graphic matching level, a reaction time threshold, and a distraction item density parameter; A behavior feedback module: performing real-time emotional state classification processing on the physiological signal data to generate a multi-modal feedback signal; the emotional state classification processing includes heart rate variability analysis and skin conductance peak detection; An adaptive adjustment module: fusing the attention fluctuation features, the cognitive training task sequence, and the multi-modal feedback signal through an incremental difficulty regulation algorithm to generate a personalized training difficulty ladder.

2. The ADHD children's cognitive training system according to claim 1, characterized in that, The distraction factor calculation algorithm includes: Performing distraction event detection on the eye movement trajectory data, and marking segments where the line of sight deviation exceeds a preset threshold; Statistical total duration and interval distribution of distraction events, and calculating the proportion of attention concentration per unit time; Analyzing the stable time after focus transfer based on an attention recovery rate model to generate a distraction factor score; Comparing the distraction factor score with a preset fluctuation threshold, and outputting an attention fluctuation feature vector.

3. The ADHD children's cognitive training system according to claim 1, characterized in that, The task complexity adaptive algorithm includes: Performing reaction delay analysis on the touch operation sequence, extracting the time window of effective operations, and dynamically adjusting the number of graphic matching levels according to the correct rate distribution within the time window; Combining the relevance between the distraction item density parameter and the reaction time threshold to generate a gradient task sequence; Mapping task parameters to a preset complexity interval through normalization processing, and outputting a standardized task sequence.

4. The ADHD children's cognitive training system according to claim 1, characterized in that, The emotional state classification processing includes: Performing sliding window segmentation on the physiological signal data, extracting heart rate variability spectrum features, and identifying stress response segments in the skin conductance signal through a peak detection algorithm; Performing feature splicing on the spectrum features and the stress response segments, inputting them into a pre-trained emotion classification model, and outputting a multi-modal feedback signal including an anxiety index and a concentration score.

5. The ADHD children's cognitive training system according to claim 1, wherein The incremental difficulty regulation algorithm includes: Performing time series difference calculation on the attention fluctuation features to extract the attention change trend; Dynamically adjusting the distraction item density of the task sequence according to the anxiety index in the multi-modal feedback signal; Performing weighted fusion on the attention change trend and the adjustment amount of the distraction item density to generate a difficulty ladder parameter; Overlaying the parameters to the historical training record through a ladder update strategy, and outputting a personalized training difficulty ladder.

6. The ADHD children's cognitive training system according to claim 2, characterized in that The construction method of the attention recovery rate model includes: dividing the attention recovery stage according to historical eye movement data, statistically counting the focus stable duration within each stage, establishing a correlation function between the distraction event duration and the recovery duration through regression analysis, and dynamically updating the calculation weight of the distraction factor score based on the correlation function to optimize the accuracy of fluctuation feature extraction.

7. The ADHD children's cognitive training system according to claim 3, wherein, The generation steps of the gradient task sequence include: Calculate the fitness score of the current task level according to the accuracy rate distribution, and select candidate tasks of the matching level from the preset task library through the probability sampling algorithm; Generate an interference intensity gradient table based on the interference item density parameter of the candidate task, and generate a multi-dimensional task sequence by combining the gradient table with the reaction time threshold.

8. The ADHD children's cognitive training system according to claim 4, wherein The training method of the pre-trained emotion classification model includes: Collect a physiological signal dataset with labeled emotion states, and perform time-frequency feature extraction and data augmentation; Build a convolutional neural network model, and input the time-frequency feature map and peak detection results; Optimize the model parameters through the cross-entropy loss function, and output the emotion classification probability distribution.

9. The ADHD children's cognitive training system according to claim 5, wherein The step update strategy includes: generating a version identifier according to the timestamp of the difficulty step parameter, identifying the deviation interval between the current parameter and the historical record through the difference comparison algorithm, and superimposing and updating the parameters in the deviation interval according to the weight decay coefficient.

10. A cognitive training method for children with ADHD, characterized in that, Include: Collect the multi-dimensional behavior data set of children; the data includes eye movement trajectory data, touch operation sequence and physiological signal data; Extract the attention fluctuation characteristics based on the distraction factor calculation algorithm; Generate a cognitive training task sequence through the task complexity adaptive algorithm; Perform real-time emotion state classification processing on the physiological signal data to generate multi-modal feedback signals; Adopt an incremental difficulty regulation algorithm to fuse the fluctuation characteristics, task sequence and feedback signals to construct a personalized training difficulty ladder; Generate a training task instruction sequence according to the difficulty ladder through the training execution engine, and output it to the interactive terminal.

Images