A multivariate modeling method for motor imagery EEG signals

Abstract

This invention discloses a multivariate modeling method for motor imagery EEG signals. EEG signals are non-stationary random signals, and single-channel signals are insufficient to fully reveal the dynamic processes of neural activity. Therefore, after denoising the EEG data, multi-channel, multivariate modeling analysis is necessary. This invention achieves a balance between computational complexity and convergence speed, making it easy to implement and meeting the needs for rapid EEG signal analysis. This invention can effectively model and analyze motor imagery EEG signals, revealing dynamic physiological information in the cognitive process.

Description

Technical Field

[0001] The invention relates to a time series modeling method, and in particular to a multi-channel motor imagery electroencephalogram signal modeling algorithm to fully mine the dynamic information of neuron activities. Background Art

[0002] EEG signals are the reflection of the electrophysiological activity of nerve cells on the scalp, and important information about cognitive activities can be extracted from them. Effective modeling of multi-channel EEG signals can not only improve the accuracy of feature extraction and classification, but also deepen the understanding of the dynamic process of cognitive activities (especially beneficial for the early rapid diagnosis of mental and psychological diseases). EEG signals are non-stationary random signals, and it is difficult to fully reveal the dynamic process of neural activity based on single-channel signals. Therefore, after denoising the EEG data, it is necessary to perform multi-channel multivariate modeling analysis. Summary of the invention

[0003] Aiming at the deficiencies of the existing methods, the present invention proposes a multivariate modeling and analysis method for motor imagery EEG signals.

[0004] The present invention proposes a multivariate modeling and analysis method for motor imagery EEG signals.

[0005] 1) The window width is w1, and the adjustment width is w2, where w2<w1; Assume that the EEG sampling signal vector of channel i and j is S i , S j , whose distance vector is defined as:

[0006]

[0007] Where N is the number of sampling points; d t For S i，t , S j，t Distance:

[0008]

[0009] where s i，t，m 、s j，t，n S i，t , S j，t The mth and nth components, S i，t For S i The tth subsequence, S j，t For S j The tth subsequence;

[0010] 2) D i，j Normalize the spacing vector:

[0011] D′ i，j =mean(D i，j )-std(D i，j ) (3)

[0012]

[0013] Where mean(·) is the expected value of the vector, std(·) is the standard deviation of the vector;

[0014] 3) Calculate the distance sequence energy E i，j , and retain the spacing sequence with a threshold greater than Th:

[0015]

[0016] C=C∪{D′ i，j} , if E i，j >Th (6)

[0017] Where C is the channel spacing sequence with a threshold greater than Th;

[0018] 4) Perform vector regression on the distance between the retained channels in C to determine the dynamic dependency Q between each pair of channels:

[0019] Q=C′∑ T (∑∑ T ) -1 (7)

[0020] in

[0021] Q represents the dynamic dependency between multiple EEG channel signal sequences, providing a basis for EEG signal feature extraction, where c represents the inter-channel distance vector regression coefficient, and (Mp) is the number of coefficients.

[0022] Beneficial effects of the invention: The invention strikes a balance between computational complexity and convergence speed, is easy to implement, and can meet the needs of rapid analysis of EEG signals. The invention can effectively model and analyze motor imagery EEG signals and discover dynamic physiological information in the cognitive process. DETAILED DESCRIPTION

[0023] A multivariate modeling and analysis method for motor imagery EEG signals, the method is as follows:

[0024] 1) The window width is w1, and the adjustment width is w2, where w2<w1; Assume that the EEG sampling signal vector of channel i and j is S i , S j , whose distance vector is defined as:

[0025]

[0026] Where N is the number of sampling points; d t For S i，t , S j，t Distance:

[0027]

[0028] where s i，t，m 、s j，t，n S i，t , S j，t The mth and nth components, S i，t For S i The tth subsequence, S j，t For S j The tth subsequence;

[0029] 2) D i，j Normalize the spacing vector:

[0030] D′ i，j =mean(D i，j )-std(D i，j ) (3)

[0031]

[0032] Where mean(·) is the expected value of the vector, std(·) is the standard deviation of the vector;

[0033] 3) Calculate the distance sequence energy E i，j , and retain the spacing sequence with a threshold greater than Th:

[0034]

[0035] C=C∪{D′ i，j}, if E i，j >Th (6)

[0036] Where C is the channel spacing sequence with a threshold greater than Th;

[0037] 4) Perform vector regression on the distance between the retained channels in C to determine the dynamic dependency Q between each pair of channels:

[0038] Q=C′∑ T (∑∑ T ) -1 (7)

[0039] in

[0040] Q represents the dynamic dependency between multiple EEG channel signal sequences, providing a basis for EEG signal feature extraction, where c represents the inter-channel distance vector regression coefficient, and (Mp) is the number of coefficients.

Claims

1. A multivariate modeling method for motor imagery EEG signals, characterized in that: The specific steps are as follows: 1) The window width is w1, and the adjustment width is w2, where w2<w1; Assume that the EEG sampling signal vector of channels i and j is S i , S j , whose distance vector is defined as: Where N is the number of sampling points; d t For S i，t , S j，t Distance: where s i，t，m 、s j，t，n S i，t , S j，t The mth and nth components, S i，t For i The t-th subsequence, S j，t For S j The tth subsequence; 2) D i，j Normalize the spacing vector: D′ i，j =mean(D i，j )-std(D i，j ) (3) Where mean(·) is the expected value of the vector, std(·) is the standard deviation of the vector; 3) Calculate the distance sequence energy E i，j , and retain the spacing sequence with a threshold greater than Th: C=C∪{D′ i，j }, if E i，j >Th (6) Where C is the channel spacing sequence with a threshold greater than Th; 4) Perform vector regression on the distance between the retained channels in C to determine the dynamic dependency Q between each pair of channels: Q=C′∑ T (∑∑ T ) -1 (7) Where C′=[c p+1 , c p+2 , c p+3 , …, c M ], Q represents the dynamic dependency between multiple EEG channel signal sequences, providing a basis for EEG signal feature extraction, where c represents the inter-channel distance vector regression coefficient, and (Mp) is the number of coefficients.

Images