EEG concentration degree-distraction degree two-dimensional classification method based on nonlinear mathematics

Abstract

The invention discloses an EEG concentration degree-distraction degree two-dimensional classification method based on nonlinear mathematics. The method comprises the following steps that 1, high-precision electroencephalogram signals are collected; 2, performing complex number transformation on the electroencephalogram signals, and separating concentration degree and distraction degree related signals; 3, extracting a two-dimensional phase angle, and constructing a concentration-distraction phase space; 4, performing feature extraction by applying a nonlinear coupling dynamic model; 5, quantifying the complex dynamic state of the cognitive state by using a chaos and fractal analysis method; and step 6, performing state classification by using a support vector machine in combination with dynamic system characteristics. The invention relates to the technical field of EEG concentration degree-distraction degree two-dimensional classification methods. According to the EEG concentration degree-distraction degree two-dimensional classification method based on nonlinear mathematics, more comprehensive cognitive state description can be realized, a complex cognitive state is captured through a two-dimensional model, the classification precision is higher, and the nonlinear model can identify fine state differences.

Description

Technical Field

[0001] The present invention relates to the technical field of EEG concentration-distraction dual-dimensional classification methods, and specifically to an EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics. Background Art

[0002] Traditional attention assessment methods often only focus on concentration and ignore the independent role of distraction. The existing technology has the following limitations:

[0003] Single dimension: Most methods (such as US10398370B2, CN110099025A) only consider a single attention dimension and cannot fully describe complex cognitive states.

[0004] Linear assumption: Many existing methods (such as EP3315069A1) use linear or simple nonlinear models, which are difficult to capture the complex dynamic characteristics of EEG signals.

[0005] Rough state classification: Existing technologies can usually only distinguish a limited number of attention states and have difficulty identifying subtle state differences.

[0006] Poor individual adaptability: Most methods lack effective adaptation mechanisms to individual differences.

[0007] The present invention aims to overcome these limitations and provide a more comprehensive and accurate method for cognitive status assessment. Summary of the invention

[0008] To achieve the above objectives, the present invention is implemented by the following technical solution: an EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics, comprising the following steps:

[0009] Step 1: Collect high-precision EEG signals;

[0010] Step 2: performing complex transformation on the EEG signal to separate the concentration and distraction related signals;

[0011] Step 3: Extract the two-dimensional phase angle and construct the concentration-distraction phase space;

[0012] Step 4: Apply nonlinear coupled dynamics model for feature extraction;

[0013] Step 5: Quantify the complex dynamics of cognitive states using chaos and fractal analysis methods;

[0014] Step 6: Use support vector machine combined with dynamic system characteristics to classify the state;

[0015] Step 7: Dynamically adjust model parameters through adaptive algorithms to adapt to individual differences.

[0016] Preferably, the nonlinear coupled dynamics model uses the following equation:

[0017] dA / dt=α1A-β1A^3-γ1AD+η1;

[0018] dD / dt=α2D-β2D^3-γ2AD+η2, where A represents concentration, D represents distraction, α, β, γ are coefficients, and η is the noise term.

[0019] Preferably, the chaos analysis includes calculating the maximum Lyapunov exponent, comprising the following steps:

[0020] (1) Select an initial point and a small perturbation point in the phase space, and calculate the rate of change of the distance between the two points as time evolves;

[0021] (2) Repeat this process for different initial conditions and take the average value to obtain the maximum Lyapunov exponent. If the exponent is positive, the system has chaotic characteristics.

[0022] The fractal analysis includes estimating the correlation dimension, comprising the following steps:

[0023] (1) Divide the points in the phase space corresponding to the EEG signal and calculate the probability that the distance between two points at different scales is less than a given threshold, i.e., the correlation integral;

[0024] (2) By changing the scale, a series of correlation integral values ​​are obtained;

[0025] (3) Use the log-log plot to fit a straight line, and the slope of the line is the correlation dimension.

[0026] Preferably, the state classification includes but is not limited to five states: high concentration and low distraction, high concentration and high distraction, low concentration and low distraction, low concentration and high distraction, and medium concentration and medium distraction, and in the classification of these five states, the overall accuracy is not less than 90%.

[0027] Preferably, the adaptive algorithm comprises the following steps:

[0028] a) Set the initial parameter set;

[0029] b) Calculate the error between the model prediction and the actual observed state;

[0030] c) Update parameters using gradient descent method;

[0031] d) Apply adaptive learning rate;

[0032] e) Continuously monitor the performance in real applications and trigger parameter readjustment when needed.

[0033] Preferably, the method can achieve reliable status detection within a time window of no more than 2 seconds.

[0034] Preferably, under the condition that the signal-to-noise ratio is 10 dB, the classification accuracy is not less than 85%.

[0035] Preferably, after 4 weeks of continuous use, the classification accuracy is improved by no less than 5% relative to the initial level, and the "high concentration and high distraction" state can be identified and classified with a classification accuracy of no less than 90%.

[0036] The device of the dual-dimensional classification method of EEG concentration and distraction based on nonlinear mathematics includes:

[0037] Signal acquisition module, used to collect high-precision EEG signals;

[0038] A signal processing module for complex number transformation and phase angle extraction;

[0039] Dynamics analysis module for applying nonlinear coupled dynamics models;

[0040] Feature extraction module for chaos and fractal analysis;

[0041] Classification module, used for status classification;

[0042] Parameter adjustment module, used to dynamically adjust model parameters.

[0043] The present invention provides an EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics. It has the following beneficial effects:

[0044] The present invention is significantly superior to the prior art in several key aspects:

[0045] A more comprehensive description of cognitive states, capturing complex cognitive states through a two-dimensional model;

[0046] Higher classification accuracy, nonlinear models can identify subtle state differences;

[0047] Better individual adaptability, dynamic parameter adjustment to adapt to individual differences;

[0048] Stronger noise immunity, chaotic dynamics improves robustness to EEG signal noise;

[0049] The present invention has wide application potential in the following fields:

[0050] High-precision cognitive monitoring, such as real-time status monitoring of pilots and air traffic controllers;

[0051] Personalized learning system that dynamically adjusts teaching strategies based on learners’ cognitive states;

[0052] Brain-computer interface optimization to improve the accuracy and response speed of brain-controlled devices;

[0053] Cognitive impairment diagnosis, providing more precise diagnostic tools for disorders such as attention deficit. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] In the first embodiment, the present invention provides a technical solution: an EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics, comprising the following steps:

[0056] Step 1: Collect high-precision EEG signals

[0057] (1) Use high-precision EEG equipment to collect signals with a sampling rate of ≥1000 Hz;

[0058] (2) Applying adaptive filters to remove artifacts;

[0059] (3) Use independent component analysis (ICA) to separate the EEG components related to concentration and distraction;

[0060] Step 2: Perform a complex transformation on the EEG signal to separate the concentration and distraction related signals, and perform a complex transformation on the preprocessed signal s(t): z(t)=A(t)+i*D(t), where A(t) represents the concentration related signal and D(t) represents the distraction related signal;

[0061] Step 3: Extract the two-dimensional phase angle and construct the concentration-distraction phase space. The two-dimensional phase angle extraction process includes:

[0062] (1) Calculate the complex phase: φ(t) = arg(z(t));

[0063] (2) Define the concentration phase angle: θA(t) = Re(φ(t));

[0064] (3) Define the distraction phase angle: θD(t) = Im(φ(t));

[0065] Step 4: Apply the nonlinear coupled dynamics model for feature extraction. Dynamic system feature extraction: Construct a two-dimensional phase space reconstruction: X(t) = [θA(t), θD(t), θA(t+τ), θD(t+τ), ..., θA(t+(m-1)τ), θD(t+(m-1)τ)] where τ is the time delay and m is the embedding dimension. Calculate the two-dimensional Lyapunov index spectrum. The nonlinear coupled dynamics model uses the following equation:

[0066] dA / dt=α1A-β1A^3-γ1AD+η1;

[0067] dD / dt=α2D-β2D^3-γ2AD+η2, where A represents concentration, D represents distraction, α, β, γ are coefficients, and η is the noise term;

[0068] Step 5: quantify the complex dynamics of cognitive states using chaos and fractal analysis methods, wherein the chaos analysis includes calculating the maximum Lyapunov exponent, including the following steps:

[0069] (1) Select an initial point and a small perturbation point in the phase space, and calculate the rate of change of the distance between the two points as time evolves;

[0070] (2) Repeat this process for different initial conditions and take the average value to obtain the maximum Lyapunov exponent. If the exponent is positive, the system has chaotic characteristics.

[0071] The specific algorithm formula is as follows:

[0072] Assume that the state equation of the nonlinear coupled dynamic system is: Where A represents concentration, D represents distraction, and f and g are corresponding function expressions;

[0073] Select the initial point X0 = (A0, D0) and a small perturbation point X0 + δX0 = (A0 + δA0, D0 + δD0) in the phase space;

[0074] After time t, the two points evolve to X(t) = (A(t), D(t)) and X(t) + δX(t) = (A(t) + δA(t), D(t) + δD(t)) respectively;

[0075] Calculate the distance between two points

[0076] The maximum Lyapunov exponent is calculated as:

[0077] Usually numerical methods are used for calculation. In actual calculation, the above process is repeated for different initial conditions, and the average value is taken to obtain a more accurate maximum Lyapunov exponent;

[0078] The fractal analysis includes estimating the correlation dimension, comprising the following steps:

[0079] (1) Divide the points in the phase space corresponding to the EEG signal and calculate the probability that the distance between two points at different scales is less than a given threshold, i.e., the correlation integral;

[0080] (2) By changing the scale, a series of correlation integral values ​​are obtained;

[0081] (3) Use the log-log plot to fit a straight line, and the slope of the line is the correlation dimension;

[0082] (4) State classification optimization;

[0083] The specific algorithm formula is as follows:

[0084] Suppose the point set in phase space is {X i =(A i , D i )}, i = 1, 2, ..., N;

[0085] Define the correlation integral C(r) as: Where Θ is the Heaviside function, when x≥0, Θ(x)=1; when x<0, Θ(x)=0;

[0086] By changing the scale r, a series of associated integral values ​​C(r) are obtained;

[0087] The calculation formula of correlation dimension D is:

[0088] In the log-log plot, lnC(r) and lnr are linearly related, and the slope of the line is the correlation dimension D, which is

[0089] Through the above formula, the maximum Lyapunov exponent and correlation dimension can be calculated relatively accurately, thus realizing the quantification of the complex dynamics of cognitive states;

[0090] Step 6: Use support vector machine combined with dynamic system characteristics to classify the state;

[0091] State classification optimization, combining support vector machine (SVM) with dynamic system characteristics to improve classification accuracy:

[0092] a) Feature extraction: Extract dynamic system features such as Lyapunov exponents, correlation dimensions, entropy measures, and phase synchronization indices;

[0093] Supplementary statistical characteristics such as mean, standard deviation, skewness, and kurtosis;

[0094] b) Feature selection: The most relevant features are selected using the recursive feature elimination (RFE) method;

[0095] c) SVM model construction: Use SVM with RBF kernel and optimize hyperparameters through grid search;

[0096] d) Classification process: including feature extraction, feature selection, feature normalization and SVM classification;

[0097] e) Model evaluation and optimization: Use k-fold cross validation to evaluate model performance and analyze classification results through confusion matrix;

[0098] Status classification steps:

[0099] a) Extract dynamic system features and statistical features;

[0100] b) Feature selection using recursive feature elimination method;

[0101] c) Build and optimize support vector machine (SVM) models;

[0102] d) Use the optimized SVM model for state classification;

[0103] Step 7: Dynamically adjust model parameters to accommodate individual differences through adaptive algorithms. Adapt model parameters to accommodate individual differences and temporal changes in EEG features through adaptive algorithms:

[0104] a) Initial parameter setting: setting the initial parameter set of the nonlinear coupling equation;

[0105] b) Error calculation: Calculate the mean square error between the model prediction and the actual observed state;

[0106] c) Gradient descent update: Use gradient descent to update parameters and minimize the prediction error;

[0107] d) Adaptive learning rate: Introduce an adaptive learning rate that decays over time to optimize the parameter update process;

[0108] e) Real-time adjustment mechanism: Continuously monitor model performance and trigger parameter readjustment when performance degrades.

[0109] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without creative work should fall within the scope of protection of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention are implemented according to the conventional means in the field unless otherwise specified and limited.

Claims

1. A dual-dimensional classification method of EEG concentration and distraction based on nonlinear mathematics, characterized by: The following steps are involved: Step 1: Collect high-precision EEG signals; Step 2: performing complex transformation on the EEG signal to separate the concentration and distraction related signals; Step 3: Extract the two-dimensional phase angle and construct the concentration-distraction phase space; Step 4: Apply nonlinear coupled dynamics model for feature extraction; Step 5: Quantify the complex dynamics of cognitive states using chaos and fractal analysis methods; Step 6: Use support vector machine combined with dynamic system characteristics to classify the state; Step 7: Dynamically adjust model parameters through adaptive algorithms to adapt to individual differences.

2. The EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics according to claim 1 is characterized in that: The nonlinear coupled dynamics model uses the following equation: dA / dt=α1A-β1A^3-γ1AD+η1; dD / dt=α2D-β2D^3-γ2AD+η2, where A represents concentration, D represents distraction, α, β, γ are coefficients, and η is the noise term.

3. The EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics according to claim 2 is characterized in that: The chaos analysis includes calculating the maximum Lyapunov exponent, including the following steps: (1) Select an initial point and a small perturbation point in the phase space, and calculate the rate of change of the distance between the two points as time evolves; (2) Repeat this process for different initial conditions and take the average value to obtain the maximum Lyapunov exponent. If the exponent is positive, the system has chaotic characteristics. The fractal analysis includes estimating the correlation dimension, comprising the following steps: (1) Divide the points in the phase space corresponding to the EEG signal and calculate the probability that the distance between two points at different scales is less than a given threshold, i.e., the correlation integral; (2) By changing the scale, a series of correlation integral values ​​are obtained; (3) Use the log-log plot to fit a straight line, and the slope of the line is the correlation dimension.

4. The EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics according to claim 3 is characterized in that: The state classification includes but is not limited to five states: high concentration and low distraction, high concentration and high distraction, low concentration and low distraction, low concentration and high distraction, and medium concentration and medium distraction, and in the classification of these five states, the overall accuracy rate is not less than 90%.

5. The EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics according to claim 4 is characterized in that: The adaptive algorithm comprises the following steps: a) Set the initial parameter set; b) Calculate the error between the model prediction and the actual observed state; c) Update parameters using gradient descent method; d) Apply adaptive learning rate; e) Continuously monitor the performance in real applications and trigger parameter readjustment when needed.

6. The EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics according to claim 5, characterized in that: The method can realize reliable status detection within a time window of no more than 2 seconds.

7. The EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics according to claim 6, characterized in that: Under the condition of a signal-to-noise ratio of 10dB, the classification accuracy is no less than 85%.

8. The EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics according to claim 7, characterized in that: After 4 weeks of continuous use, the classification accuracy rate increased by no less than 5% compared to the initial level, and it was able to identify and classify the "high concentration and high distraction" state, with a classification accuracy rate of no less than 90%.

9. The device for the EEG concentration-distraction dual-dimensional classification method based on nonlinear mathematics according to claim 8, characterized in that: include: Signal acquisition module, used to collect high-precision EEG signals; A signal processing module for complex number transformation and phase angle extraction; Dynamics analysis module for applying nonlinear coupled dynamics models; Feature extraction module for chaos and fractal analysis; Classification module, used for status classification; Parameter adjustment module, used to dynamically adjust model parameters.

Images