Cognitive load-based concentration adaptive intervention method and system

Abstract

The present application belongs to the technical field of concentration intervention, and particularly relates to a concentration adaptive intervention method and system based on cognitive load, which comprises the following steps: S1, acquiring eye movement tracking data and heart rate signal data of a user within a preset time window; S2, extracting a pupil diameter time sequence signal and a heart rate variability time sequence signal from the data, and obtaining a pupil-heart rate phase coupling coefficient representing cognitive load by performing complementary ensemble empirical mode decomposition and phase lock value calculation on the two signals; and based on the eye movement tracking data, calculating saccadic entropy, saccadic path backtracking frequency and pupil diameter microfluctuation standard deviation. The intervention measures generated by the present application directly correspond to the vector of concentration deviation in terms of strength and guidance direction, and guide the user's attention to return in a mild and non-invasive manner, thereby realizing precise closed-loop correction of concentration.

Description

Technical Field

[0001] This invention belongs to the field of attention intervention technology, specifically relating to an adaptive attention intervention method and system based on cognitive load. Background Technology

[0002] In modern society, real-time monitoring and intervention of individual attention is crucial in fields such as education, driving safety, and complex task operations. Measurements based on physiological signals have garnered significant attention due to their objectivity and real-time nature. For example, eye-tracking technology infers the allocation of visual attention by analyzing parameters such as fixation point, saccades, and pupil diameter. Many methods rely on single-modal physiological signals, such as using only eye-tracking data or only electrocardiogram (ECG) data. Such single-dimensional information cannot fully represent the complex, multidimensional cognitive processes of attention. For instance, while changes in pupil diameter are related to cognitive load, they are also affected by physical factors such as changes in illumination. Even when multimodal data fusion is employed, it is usually limited to simple linear weighting or feature splicing, failing to delve into the deep coupling relationships between different physiological systems in cognitive activities.

[0003] Current technologies also have shortcomings in attention intervention. The representation of attention states is not refined enough. Traditional eye-tracking features, such as saccade entropy and path retracement, while reflecting macroscopic patterns of visual search strategies, lack a connection to underlying physiological states. Constructing a unified attention state space that integrates higher-order visual behavior patterns and deep physiological states is a current technological challenge. Regarding intervention strategies, existing methods mostly employ alarm modes with preset fixed thresholds. When one or more indicators exceed the threshold, a uniform, non-specific warning signal is triggered. Therefore, there is an urgent need for a closed-loop control method that can generate personalized, guided intervention signals based on the "vector" of attention deviation. This method should be able to predict the user's future attention focus and precisely apply gentle visual signals to that area, thereby achieving precise and flexible correction of attention. Summary of the Invention

[0004] This invention provides a cognitive load-based adaptive intervention method and system for attention, which solves the technical problem that existing attention monitoring methods are singular or superficial and intervention is crude, and there is an urgent need for precise and personalized flexible control solutions.

[0005] In a first aspect, the present invention provides a focus-adaptive intervention method based on cognitive load, comprising the following steps:

[0006] S1, acquires eye-tracking data and heart rate signal data of the user within a preset time window;

[0007] S2. Extract the pupil diameter time-series signal and heart rate variability time-series signal from the data. By performing complementary ensemble empirical mode decomposition and phase-locking value calculation on the two signals, obtain the pupil-heart rate phase coupling coefficient representing cognitive load. Based on eye-tracking data, calculate saccade entropy, saccade path backtracking number and pupil diameter micro-fluctuation standard deviation, and construct a multidimensional saccade feature vector by nonlinear weighting based on the pupil-heart rate phase coupling coefficient of the previous time window.

[0008] S3, construct an attention state space defined by multidimensional saccade feature vectors and pupil-heart rate phase coupling coefficient, and preset an ideal attention state anchor point in the space; when the Mahalanobis distance between the current attention state point and the ideal attention state anchor point exceeds a threshold related to the pupil-heart rate phase coupling coefficient, determine an attention correction vector pointing from the current attention state point to the ideal attention state anchor point;

[0009] S4 generates an asymmetric lens field as a visual intervention signal in the next potential gaze area predicted by the historical saccade path, based on the direction and magnitude of the attention correction vector.

[0010] Furthermore, by performing complementary ensemble empirical mode decomposition and phase-locked value calculation on the two signals, the pupil-heart rate phase coupling coefficient representing cognitive load is obtained, including:

[0011] Step 1: Complementary set empirical mode decomposition is used to decompose the pupil diameter time series signal and the heart rate variability time series signal into a set of intrinsic mode functions respectively;

[0012] Step 2: Select the target intrinsic mode functions with frequencies in the range of 0.08Hz to 0.15Hz from the two sets of intrinsic mode functions;

[0013] Step 3: Apply Hilbert transform to the two selected sets of target intrinsic mode functions to extract the instantaneous phase signal;

[0014] Step 4: Based on the phase difference distribution of the instantaneous phase signal within the time window, calculate the phase lock value and use the phase lock value as the pupil-heart rate phase coupling coefficient.

[0015] Furthermore, a multidimensional saccade feature vector is constructed by nonlinearly weighting the pupil-heart rate phase coupling coefficient from the previous time window, including:

[0016] Define a nonlinear weighting function :

[0017] ;

[0018] Where x is the pupil-heart rate phase coupling coefficient of the previous time window;

[0019] The calculated saccade entropy, saccade path backtracking times, and pupil diameter micro-fluctuation standard deviation are multiplied by the weight values ​​calculated by the nonlinear weighting function to generate weighted saccade features.

[0020] The three weighted saccadic features are used as three dimensions to form a multidimensional saccadic feature vector.

[0021] Furthermore, ideal focus state anchor points are preset within the space, including:

[0022] Before the method is implemented, users are guided to complete a 5-minute Stroop color word test as a high-focus calibration task;

[0023] During the high-attention calibration task, the user’s multidimensional saccade feature vector and pupil-heart rate phase coupling coefficient are continuously calculated to form a series of calibration state points;

[0024] Calculate the geometric center of a series of calibration state points in the attention state space, and set the coordinates of the geometric center as the ideal attention state anchor point.

[0025] Furthermore, when the Mahalanobis distance between the current focus state point and the ideal focus state anchor point exceeds a threshold related to the pupil-heart rate phase coupling coefficient, a focus correction vector is determined pointing from the current focus state point to the ideal focus state anchor point, including:

[0026] The threshold T is determined by the following formula:

[0027] ;

[0028] in, The average Mahalanobis distance determined through the calibration task. This represents the pupil-heart rate phase coupling coefficient for the current time window.

[0029] When the calculated Mahalanobis distance is greater than the threshold T, intervention is triggered, and the vector pointing from the current focus state point to the ideal focus state anchor point is determined as the focus correction vector.

[0030] Furthermore, in the next potential gaze region predicted by historical saccades, an asymmetric lens field is generated as a visual intervention signal, including:

[0031] Obtain vector information of three consecutive saccades preceding the current fixation point, and predict the landing point of the next saccade based on a linear regression model;

[0032] A circular area with a radius of 2° visual angle, centered on the predicted landing point, is designated as the next potential fixation area.

[0033] An asymmetric lens field is generated within the next potential gaze region. The asymmetric lens field is a Gaussian blur field, and the blur intensity is proportional to the magnitude of the attention correction vector.

[0034] The direction of the blur gradient of the Gaussian blur field is set to point from the predicted landing point to the starting point of the previous scan, so that the area with the lowest blur level provides the user with visual guidance to return to the previous scan path.

[0035] Furthermore, the direction and magnitude of the focus correction vector include:

[0036] The direction of the focus correction vector is the direction from the current focus state point to the ideal focus state anchor point in the focus state space;

[0037] The magnitude of the focus correction vector is the Mahalanobis distance between the current focus state point and the ideal focus state anchor point.

[0038] Furthermore, in S1, the eye-tracking data includes the diameter of both pupils and the screen coordinates of the gaze point.

[0039] Furthermore, in S4, based on the vector information of the user's five most recent gaze points, a Kalman filter is used to predict the location of the next gaze.

[0040] Secondly, the present invention provides a cognitive load-based adaptive attention intervention system, including a memory and a processor. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned cognitive load-based adaptive attention intervention method is implemented.

[0041] The beneficial effects are: it deeply reveals the profound coupling relationship between the visual system and the autonomic nervous system in cognitive activities, overcoming the limitations of being easily interfered with by environmental factors. This invention uses the cognitive load index to weight saccadic behavior characteristics, constructing a multi-dimensional state space that more comprehensively and precisely presents the essence of focus. Furthermore, the intervention trigger conditions are correlated with the cognitive load level, making the judgment of focus deviation more reliable. The generated intervention measures, with their intensity and guidance direction directly corresponding to the vector of focus deviation, guide the user's attention back in a gentle, non-invasive manner, achieving precise closed-loop correction of focus. Attached Figure Description

[0042] Figure 1 This is a flowchart of a cognitive load-based attention-adaptive intervention method. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] An embodiment of the attention-adaptive intervention method based on cognitive load provided by this invention:

[0045] like Figure 1 As shown, the attention-adaptive intervention method based on cognitive load includes the following steps:

[0046] S1, acquires the user's eye-tracking data and heart rate signal data within a preset time window.

[0047] Specifically, a head-mounted eye-tracking device continuously collects eye-tracking data such as the user's pupil diameter and fixation point screen coordinates at a sampling rate of 120Hz. Simultaneously, a wearable heart rate wristband or ECG patch acquires the user's heart rate signal, such as photoplethysmography (PPG) wave signals. A preset time window is set to 30 seconds, meaning the collected data is processed and analyzed every 30 seconds.

[0048] S2. Extract the pupil diameter time-series signal and heart rate variability time-series signal from the data. By performing complementary ensemble empirical mode decomposition and phase-locked value calculation on the two signals, obtain the pupil-heart rate phase coupling coefficient representing cognitive load. Based on eye-tracking data, calculate saccade entropy, saccade path backtracking number and pupil diameter micro-fluctuation standard deviation, and construct a multidimensional saccade feature vector by nonlinear weighting according to the pupil-heart rate phase coupling coefficient of the previous time window.

[0049] Specifically, the raw pupil diameter data is denoised and interpolated to eliminate artifacts such as blinking, resulting in a smooth pupil diameter time-series signal. Peak detection is performed on the heart rate signal to obtain the RR interval sequence, which is then resampled into an equally time-interval heart rate variability time-series signal using cubic spline interpolation. Complementary ensemble empirical mode decomposition is performed on both the processed pupil diameter and heart rate variability time-series signals, decomposing each signal into a set of intrinsic mode functions (IMFs). Specific frequency bands related to cognitive activity are selected, such as IMF components in the 0.01-0.15 Hz range, and Hilbert transform is applied to extract the instantaneous phase. The time average of the instantaneous phase difference between the two signals is calculated, which is the pupil-heart rate phase coupling coefficient. This coefficient ranges from 0 to 1; a higher value indicates stronger synchronicity between the two physiological systems and a higher level of cognitive load.

[0050] The user's screen field of view is divided into an 8×6 grid. The transition probability matrix is ​​calculated based on the transition frequency of the saccade path between different grids, and the saccade entropy is calculated accordingly. By analyzing the saccade vector sequence, saccades opposite to the overall saccade direction are defined as backtracking, and the number of backtracking occurrences within a time window is counted. A high-pass filter with a cutoff frequency of 2Hz is applied to the pupil diameter time-series signal to filter out low-frequency variations caused by factors such as illumination. The standard deviation of the filtered signal is then calculated as the standard deviation of the micro-fluctuations in pupil diameter. Using the pupil-heart rate phase coupling coefficient C from the previous time window as an adjustment factor, the weights of the three currently calculated saccade features are non-linearly adjusted using an exponential function, such as a weight equal to e raised to the power of C, to construct a three-dimensional weighted saccade feature vector.

[0051] In an optional embodiment, the pupil-heart rate phase coupling coefficient representing cognitive load is obtained by performing complementary ensemble empirical mode decomposition and phase-locked value calculation on the two signals, including:

[0052] Step 1: Complementary set empirical mode decomposition is used to decompose the pupil diameter time series signal and the heart rate variability time series signal into a set of intrinsic mode functions respectively;

[0053] Step 2: Select the target intrinsic mode functions with frequencies in the range of 0.08Hz to 0.15Hz from the two sets of intrinsic mode functions;

[0054] Step 3: Apply Hilbert transform to the two selected sets of target intrinsic mode functions to extract the instantaneous phase signal;

[0055] Step 4: Based on the phase difference distribution of the instantaneous phase signal within the time window, calculate the phase lock value and use the phase lock value as the pupil-heart rate phase coupling coefficient.

[0056] Specifically, pupil diameter data is collected within a 60-second time window, for example, a series of values ​​such as 3.1mm, 3.2mm, and 3.15mm. Simultaneously, heart rate variability (RR) interval data is collected, for example, 800ms, 810ms, and 795ms. The two sets of time-series data are then input into a complementary set empirical mode decomposition algorithm. This algorithm decomposes each complex signal into multiple simpler intrinsic mode function (EMF) components. For example, the pupil diameter time-series signal is decomposed into five components, and the heart rate variability time-series signal into six components. The frequency characteristics of each component are analyzed. If the dominant frequency of the third component of the pupil diameter time-series signal is found to be 0.12Hz, and the dominant frequency of the fourth component of the heart rate variability time-series signal is found to be 0.10Hz, since both frequencies fall within the target range of 0.08 to 0.15Hz, these two components are selected as the target EMFs.

[0057] Hilbert transforms are applied to the selected pupil and heart rate target components. This transform extracts the time-varying instantaneous phase information from the one-dimensional signal, resulting in two new instantaneous phase sequences representing the phase changes of the pupil and heart rate in the target frequency band, respectively. The phase difference between the two instantaneous phase sequences at each time point is calculated, and a phase lock value is calculated based on this set of phase difference data. For example, if the calculated phase lock value is 0.75, the resulting value is used as the pupil-heart rate phase coupling coefficient within that time window. A value close to 1 indicates a high degree of synchronization between the pupil and heart rate rhythms, reflecting a high cognitive load.

[0058] In an optional embodiment, a multidimensional saccade feature vector is constructed by nonlinearly weighting the pupil-heart rate phase coupling coefficients from the previous time window, including:

[0059] Define a nonlinear weighting function :

[0060] ;

[0061] Where x is the pupil-heart rate phase coupling coefficient of the previous time window;

[0062] The calculated saccade entropy, saccade path backtracking times, and pupil diameter micro-fluctuation standard deviation are multiplied by the weight values ​​calculated by the nonlinear weighting function to generate weighted saccade features.

[0063] The three weighted saccadic features are used as three dimensions to form a multidimensional saccadic feature vector.

[0064] Specifically, assuming the pupil-heart rate phase coupling coefficient x calculated in the previous time window is 0.8, substituting this value into a nonlinear weighting function yields a weight W of approximately 0.82. This weight increases with cognitive load, thus amplifying the magnitude of eye movement characteristics under distracted conditions. Within the current time window, the user's raw eye movement data is measured, resulting in a saccade entropy of 2.8, 5 saccade path backtrackings, and a standard deviation of 0.06 mm for the microscopic fluctuations in pupil diameter.

[0065] The weight value of 0.82 calculated in the previous step is multiplied by each of the three original eye movement features. The weighted saccade entropy is 2.296. The weighted saccade path backtracking count is 4.1. The weighted pupil diameter fluctuation standard deviation is 0.0492. The three weighted values ​​of 2.296, 4.1, and 0.0492 are combined into a three-dimensional vector, which represents the user's specific location in the attention state space at the current moment, i.e., the multidimensional saccade feature vector.

[0066] S3. Construct an attention state space defined by multidimensional saccade feature vectors and pupil-heart rate phase coupling coefficient, and preset an ideal attention state anchor point in the space; when the Mahalanobis distance between the current attention state point and the ideal attention state anchor point exceeds a threshold related to the pupil-heart rate phase coupling coefficient, determine an attention correction vector pointing from the current attention state point to the ideal attention state anchor point.

[0067] Specifically, the weighted saccade entropy, backtracking count, and standard deviation of pupil micro-fluctuations are combined with the pupil-heart rate phase coupling coefficient calculated within the current time window to form a new dimension, collectively constituting a four-dimensional attention state space. The ideal attention state anchor point is determined through a calibration phase: the user performs a task requiring high concentration in an undisturbed environment, the attention state point during the user's best performance period is recorded, and the centroid of this attention state point is calculated. The four-dimensional coordinates of this centroid are then set as the ideal attention state anchor point.

[0068] The covariance matrix between each feature dimension in the attention state space is calculated using data from the calibration phase. For each newly generated attention state point, the Mahalanobis distance between the attention state point and the ideal attention state anchor point is calculated, taking into account the correlation between features. The threshold is set as a quadratic function of the pupil-heart rate phase coupling coefficient C. For example, the threshold is equal to a constant a minus a constant b multiplied by C, then multiplied by 1 minus C, so that the threshold is low when C is too low (i.e., inattention) or too high (i.e., cognitive overload), and high when C is at a moderate level. Once the Mahalanobis distance exceeds this threshold, the coordinates of the current attention state point are subtracted from the coordinates of the ideal attention state anchor point to obtain a four-dimensional attention correction vector.

[0069] In an optional embodiment, an ideal focus state anchor point is preset within the space, including:

[0070] Before the method is implemented, users are guided to complete a 5-minute Stroop color word test as a high-focus calibration task;

[0071] During the high-attention calibration task, the user’s multidimensional saccade feature vector and pupil-heart rate phase coupling coefficient are continuously calculated to form a series of calibration state points;

[0072] Calculate the geometric center of a series of calibration state points in the attention state space, and set the coordinates of the geometric center as the ideal attention state anchor point.

[0073] Specifically, before the user begins performing the main task, a standard Stroop color word test interface is presented. For example, the word "red" written in blue ink appears on the screen, and the user needs to press the button representing blue. This process lasts for 5 minutes, requiring the user to maintain a high level of concentration. During this period, the user's eye movements and physiological signals are continuously collected in 30-second time windows, and the multidimensional saccade feature vector corresponding to each window is calculated.

[0074] After a 5-minute calibration task, a dataset containing ten 3D saccade feature vectors was obtained. For example, the first vector had values ​​of 1.9, 2.8, and 0.04; the second vector had values ​​of 1.8, 3.1, and 0.05; and so on. The average value of each of the ten vectors in each dimension was calculated. For instance, the average value for the first dimension was the arithmetic mean of the first-dimensional components of all ten vectors, resulting in 1.85; the average value for the second dimension was 3.0; and the average value for the third dimension was 0.045. A new 3D coordinate point of 1.85, 3.0, and 0.045 was obtained, and this point was defined as the ideal focus state anchor point for this user, serving as a benchmark for subsequent assessments of focus levels.

[0075] In an optional embodiment, when the Mahalanobis distance between the current focus state point and the ideal focus state anchor point exceeds a threshold related to the pupil-heart rate phase coupling coefficient, a focus correction vector is determined from the current focus state point to the ideal focus state anchor point, including:

[0076] The threshold T is determined by the following formula:

[0077] ;

[0078] in, The average Mahalanobis distance determined through the calibration task. This represents the pupil-heart rate phase coupling coefficient for the current time window.

[0079] When the calculated Mahalanobis distance is greater than the threshold T, intervention is triggered, and the vector pointing from the current focus state point to the ideal focus state anchor point is determined as the focus correction vector.

[0080] Specifically, during the calibration phase, the Mahalanobis distance from each calibration state point to the ideal focus anchor point is calculated, and the average Mahalanobis distance is set as the base distance. Assuming a value of 1.3, after entering the task phase, the current pupil-heart rate phase coupling coefficient is calculated within a certain time window. A value of 0.3 indicates a low cognitive load. In this case, the calculated threshold T is 1.65. The threshold adjusts according to cognitive load; it decreases when cognitive load is high, making the user more sensitive to distraction under high load.

[0081] The system calculates the user's current focus point, for example, coordinates 2.5, 4.8, 0.08, and then calculates the Mahalanobis distance between this point and the preset ideal focus anchor point. Let's assume the calculated Mahalanobis distance is 1.8. Since 1.8 is greater than the threshold of 1.65, it's determined that the user's focus has deviated from the ideal state, requiring intervention. At this point, a focus correction vector is defined. This vector starts at the current focus point (2.5, 4.8, 0.08) and ends at the ideal focus anchor point. This vector clearly defines the gap and direction between the current state and the ideal state.

[0082] S4 generates an asymmetric lens field as a visual intervention signal in the next potential gaze area predicted by the historical saccade path, based on the direction and magnitude of the attention correction vector.

[0083] Specifically, based on the location and velocity information of the user's five most recent gaze points, a Kalman filter is used to predict the region of the user's next most likely gaze point. The magnitude of the attention correction vector, i.e., its Euclidean length, is linearly mapped to the magnification and transparency of the asymmetric lens field. The larger the magnitude, the higher the magnification, the lower the transparency, and the greater the intervention intensity. The direction of the attention correction vector, i.e., the proportion of its components in the four dimensions, determines the shape of the lens field. For example, if the deviation of the saccadic entropy component is the largest, a laterally elongated elliptical lens field is generated to guide the user to explore a wider range; if the deviation of the cognitive load component is the largest, a circular lens field with a clear center and slightly blurred edges is generated to help the user focus on core information. The generated visual effect is superimposed as a temporary, semi-transparent layer on the predicted gaze area.

[0084] In an optional embodiment, the direction and magnitude of the focus correction vector include:

[0085] The direction of the focus correction vector is the direction from the current focus state point to the ideal focus state anchor point in the focus state space;

[0086] The magnitude of the focus correction vector is the Mahalanobis distance between the current focus state point and the ideal focus state anchor point.

[0087] Specifically, imagine a three-dimensional attention state space composed of weighted saccadic entropy, weighted backtracking count, and weighted pupil variability standard deviation. Within this space, there are two key points. One is the ideal attention state anchor point, whose coordinates are determined during the calibration phase, for example, 1.8, 3.0, 0.04. The other is the real-time calculated current attention state point; assuming that when the user is distracted, the coordinates of the current attention state point change to 2.9, 5.2, 0.09. The direction of the attention correction vector is defined as a direct line from the current attention state point (2.9, 5.2, 0.09) to the ideal attention state anchor point (1.8, 3.0, 0.04).

[0088] The length of the attention correction vector, or its modulus, is not calculated using simple geometric distance, but rather using Mahalanobis distance. Mahalanobis distance is a statistical distance that considers the correlation between different dimensions of the data. The Mahalanobis distance between the current attention state points (2.9, 5.2, 0.09) and the ideal attention state anchor points (1.8, 3.0, 0.04) is calculated. For example, the result is 1.7. This value of 1.7 becomes the modulus of the attention correction vector. The modulus indicates the severity of the deviation of the current attention state from the ideal state; a larger modulus value will result in a stronger visual intervention signal.

[0089] In an optional embodiment, an asymmetric lens field is generated as a visual intervention signal in the next potential gaze region predicted by historical saccades, including:

[0090] Obtain vector information of three consecutive saccades preceding the current fixation point, and predict the landing point of the next saccade based on a linear regression model;

[0091] A circular area with a radius of 2° visual angle, centered on the predicted landing point, is designated as the next potential fixation area.

[0092] An asymmetric lens field is generated within the next potential gaze region. The asymmetric lens field is a Gaussian blur field, and the blur intensity is proportional to the magnitude of the attention correction vector.

[0093] The direction of the blur gradient of the Gaussian blur field is set to point from the predicted landing point to the starting point of the previous scan, so that the area with the lowest blur level provides the user with visual guidance to return to the previous scan path.

[0094] Specifically, the user's three most recent saccades are recorded, for example, from screen coordinates A to B, then to C, and finally to D. Based on these three saccade vectors, a simple linear regression model is used to predict the location where the user is most likely to gaze next, for example, predicting the landing point as screen coordinate E. With the predicted point E as the center and a radius equivalent to a 2° visual angle in the user's field of vision, a virtual circular area is drawn on the screen; this area is the target area for visual intervention.

[0095] When intervention is needed, a Gaussian blur effect is applied within the circular area. The overall intensity of the Gaussian blur is directly related to the magnitude of the previously calculated attention correction vector; a larger magnitude indicates a more severe attention deviation, and a stronger blur effect. More importantly, the blur effect is non-uniform. Its blur level gradually changes from one side of the circular area to the other. The gradient direction is set from the predicted landing point E to the starting point D of the previous scan. The edge of the circular area near point D has the lightest blur, even being clear, while the side farthest from point D is the most blurred, thus visually creating a clear path that is not easily noticeable, guiding the user's gaze naturally back from the potentially deviated point E to the previous scan path.

[0096] An embodiment of the attention-adaptive intervention system based on cognitive load provided by this invention:

[0097] The cognitive load-based attention adaptive intervention system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the aforementioned cognitive load-based attention adaptive intervention method.

[0098] The cognitive load-based attention-adaptive intervention system also includes other components well known to those skilled in the art, such as communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0099] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained by such a computer-readable medium.

[0100] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A cognitive load-based adaptive intervention method for attention, characterized in that, Includes the following steps: S1, acquires eye-tracking data and heart rate signal data of the user within a preset time window; S2. Extract the pupil diameter time-series signal and heart rate variability time-series signal from the data. Obtain the pupil-heart rate phase coupling coefficient representing cognitive load by performing complementary ensemble empirical mode decomposition and phase-locking value calculation on the two signals. This includes: using complementary ensemble empirical mode decomposition to decompose the pupil diameter time-series signal and heart rate variability time-series signal into a set of intrinsic mode functions (IMFs); selecting target IMFs with frequencies in the range of 0.08Hz to 0.15Hz from the two sets of IMFs; applying Hilbert transform to the selected two sets of target IMFs to extract the instantaneous phase signal; and calculating the phase-locking value based on the phase difference distribution of the instantaneous phase signal within the time window, using the phase-locking value as the pupil-heart rate phase coupling coefficient. Based on eye-tracking data, a multidimensional saccade feature vector is constructed, including: Define a nonlinear weighting function : Where x is the pupil-heart rate phase coupling coefficient of the previous time window; The calculated saccade entropy, saccade path backtracking times, and pupil diameter micro-fluctuation standard deviation are multiplied by weight values ​​calculated by a nonlinear weighting function to generate weighted saccade features. The three weighted saccade features are used as three dimensions to form a multidimensional saccade feature vector. S3 constructs a focus state space defined by multidimensional saccade feature vectors and pupil-heart rate phase coupling coefficients, and presets ideal focus state anchor points within the space. This includes: before the method is executed, guiding the user to complete a 5-minute Stroop color word test as a high focus calibration task; during the high focus calibration task, continuously calculating the user's multidimensional saccade feature vectors and pupil-heart rate phase coupling coefficients to form a series of calibration state points; calculating the geometric center of the series of calibration state points in the focus state space, and setting the coordinates of the geometric center as the ideal focus state anchor point. When the Mahalanobis distance between the current focus state point and the ideal focus state anchor point exceeds a threshold related to the pupil-heart rate phase coupling coefficient, a focus correction vector is determined pointing from the current focus state point to the ideal focus state anchor point. S4 generates an asymmetric lens field as a visual intervention signal in the next potential gaze area predicted by the historical saccade path, based on the direction and magnitude of the attention correction vector.

2. The attention-adaptive intervention method based on cognitive load according to claim 1, characterized in that, When the Mahalanobis distance between the current focus point and the ideal focus point exceeds a threshold related to the pupil-heart rate phase coupling coefficient, a focus correction vector is determined pointing from the current focus point to the ideal focus point, including: The threshold T is determined by the following formula: ； in, The average Mahalanobis distance determined through the calibration task. This represents the pupil-heart rate phase coupling coefficient for the current time window. When the calculated Mahalanobis distance is greater than the threshold T, intervention is triggered, and the vector pointing from the current focus state point to the ideal focus state anchor point is determined as the focus correction vector.

3. The attention-adaptive intervention method based on cognitive load according to claim 1, characterized in that, In the next potential gaze region predicted by historical saccades, an asymmetric lens field is generated as a visual intervention signal, including: Obtain vector information of three consecutive saccades preceding the current fixation point, and predict the landing point of the next saccade based on a linear regression model; A circular area with a radius of 2° visual angle, centered on the predicted landing point, is designated as the next potential fixation area. An asymmetric lens field is generated within the next potential gaze region. The asymmetric lens field is a Gaussian blur field, and the blur intensity is proportional to the magnitude of the attention correction vector. The direction of the blur gradient of the Gaussian blur field is set to point from the predicted landing point to the starting point of the previous scan, so that the area with the lowest blur level provides the user with visual guidance to return to the previous scan path.

4. The attention-adaptive intervention method based on cognitive load according to claim 1, characterized in that, The direction and magnitude of the focus correction vector include: The direction of the focus correction vector is the direction from the current focus state point to the ideal focus state anchor point in the focus state space; The magnitude of the focus correction vector is the Mahalanobis distance between the current focus state point and the ideal focus state anchor point.

5. The attention-adaptive intervention method based on cognitive load according to claim 1, characterized in that, In S1, eye-tracking data includes the diameter of both pupils and the screen coordinates of the gaze point.

6. The attention-adaptive intervention method based on cognitive load according to claim 1, characterized in that, In S4, a Kalman filter is used to predict the location of the next gaze based on the vector information of the user's five most recent gaze points.

7. A cognitive load-based attentional adaptive intervention system, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the attention-adaptive intervention method based on cognitive load as described in any one of claims 1-6 is implemented.

Images