CN109009101B - An adaptive real-time denoising method for EEG signals - Google Patents

Abstract

The invention discloses an adaptive real-time denoising method for electroencephalogram signals; the invention first performs centralization processing on a sampling matrix X, calculates the covariance matrix of the sampling matrix X, calculates its eigenvalues and eigenvectors, and calculates and filters the self-adaptive Coefficients, filter coefficients, noise signal y source signal; finally, the source signal matrix is obtained; the present invention achieves a balance between computational complexity and convergence speed, is easy to implement in PFGA, and can meet the real-time acquisition requirements of EEG signals. The invention has fast convergence speed, is not easy to change the waveform shape, and can effectively remove physiological artifacts and circuit noise.

Description

Electroencephalogram signal self-adaptive real-time denoising method

Technical Field

The invention relates to a denoising method, in particular to an electroencephalogram signal self-adaptive real-time denoising method.

Background

Electroencephalogram (EEG) is a reflection of electrophysiological activity of brain nerve cells on the cerebral cortex or scalp and can be collected by electrodes. The electroencephalogram signals contain rich physiological information, and the accuracy of feature extraction and classification can be improved by effectively filtering the electroencephalogram signals. EEG signals have the following characteristics: the EEG signal is a weak physiological signal, the magnitude is microvolt, and the amplitude is generally 0.1-100 muV; EEG signal frequency is distributed between 0.5 Hz and 3000Hz, and signal energy is mainly concentrated between 2 Hz and 30 Hz; the system is easy to be interfered by noise, including electro-oculogram interference, myoelectricity interference, instrument high-frequency electromagnetic noise interference, acquisition environment power frequency interference and the like; EEG signals are non-stationary random signals, and the neural activity they represent is difficult to detect from the signal itself, making identification of neural commands or clinical diagnosis difficult. It is therefore necessary to de-noise when acquiring and analyzing EEG data. EEG signals are typically complex non-stationary signals whose instantaneous frequency and instantaneous energy need to be analyzed.

Disclosure of Invention

The invention provides an electroencephalogram signal self-adaptive real-time denoising method aiming at the defects of the prior art.

The invention discloses an electroencephalogram signal self-adaptive real-time denoising method, which comprises the following steps:

suppose N is the EEG signal sampling channel number, Q is the EEG signal sampling number, and the source signal matrix is S ═ S₁,s₂,…,s_N]_Q×NThe EEG sampling matrix is X ═ X₁,x₂,…,x_N]_Q×N，Y_Q×NFor an additive noise matrix, the following equation is given:

X＝S+Y (1)

the objective is to obtain a source signal matrix S from a sampling matrix X so as to achieve the purpose of denoising. The method comprises the following specific steps:

step 1: carrying out centralization processing on the sampling matrix X:

X_O＝X-U^TH (2)

wherein

H[n]1,2, …, Q, N, 1,2, …, N. And X (m, n) is the nth voltage value acquired by the mth channel.

Step 2: calculating a covariance matrix of the sampling matrix X, and calculating an eigenvalue and an eigenvector of the sampling matrix X:

C＝E[XX^T] (3)

where E [. cndot. ] is the desired value. Since the covariance matrix C is a positive definite symmetric matrix, its eigenvectors are real and orthogonal. Let D be a vector of eigenvalues of the X covariance matrix C used to calculate the filter coefficients.

And step 3: for m-1, …, Q-1:

step 3.1: computing filter adaptation coefficients

Wherein 0<α<1，0<β，

And

respectively represent coefficient matrixes

The m +1 th and m-th row vectors. X_O(m,: represents X)_OThe m-th row vector of (2).

Step 3.2: calculating a filter coefficient w:

wherein w (m + 1:) and w (m:) represent the m +1 and m row vectors of w, respectively. D (m) represents the mth component of D. w (1) ═ 0, …, 0.

Step 3.3: estimating the noise signal y:

y(m,:)＝w^T(m,:)D(m) (6)

the superscript T denotes transpose. Y (m,: indicates the Y mth row vector.

Step 3.4: calculating a source signal:

an mth row evaluation value representing S, X (m:) being an Xmth row vector.

And 4, step 4: and finally, obtaining a source signal matrix S:

the invention has the beneficial effects that: the invention balances the calculation complexity and the convergence rate, is convenient to realize in PFGA, and can meet the real-time acquisition requirement of electroencephalogram signals. The invention has fast convergence speed, is not easy to change the waveform shape and can effectively remove physiological artifacts and circuit noise.

Drawings

FIG. 1 is a flow chart of the adaptive real-time filtering of electrical signals according to the present invention.

Detailed Description

As shown in fig. 1, a method for adaptively denoising an electroencephalogram signal in real time specifically comprises the following steps:

suppose N is the EEG signal sampling channel number, Q is the EEG signal sampling number, and the source signal matrix is S ═ S₁,s₂,…,s_N]_Q×NThe EEG sampling matrix is X ═ X₁,x₂,…,x_N]_Q×N，Y_Q×NFor an additive noise matrix, the following equation is given:

X＝S+Y (1)

the objective is to obtain a source signal matrix S from a sampling matrix X so as to achieve the purpose of denoising. The method comprises the following specific steps:

step 1: carrying out centralization processing on the sampling matrix X:

X_O＝X-U^TH (2)

wherein

H[n]1,2, …, Q, N, 1,2, …, N. And X (m, n) is the nth voltage value acquired by the mth channel.

Step 2: calculating a covariance matrix of the sampling matrix X, and calculating an eigenvalue and an eigenvector of the sampling matrix X:

C＝E[XX^T] (3)

where E [. cndot. ] is the desired value. Since the covariance matrix C is a positive definite symmetric matrix, its eigenvectors are real and orthogonal. Let D be a vector of eigenvalues of the X covariance matrix C used to calculate the filter coefficients.

And step 3: for m-1, …, Q-1:

step 3.1: computing filter adaptation coefficients

Wherein 0<α<1，0<β，

And

respectively represent coefficient matrixes

The m +1 th and m-th row vectors. X_O(m,: represents X)_OThe m-th row vector of (2).

Step 3.2: calculating a filter coefficient w:

wherein w (m + 1:) and w (m:) represent the m +1 and m row vectors of w, respectively. D (m) represents the mth component of D. w (1) ═ 0, …, 0.

Step 3.3: estimating the noise signal y:

y(m,:)＝w^T(m,:)D(m) (6)

the superscript T denotes transpose. Y (m,: indicates the Y mth row vector.

Step 3.4: calculating a source signal:

an mth row evaluation value representing S, X (m:) being an Xmth row vector.

And 4, step 4: and finally, obtaining a source signal matrix S:

Claims (1)

1. an electroencephalogram signal self-adaptive real-time denoising method comprises the following steps:

suppose N is the EEG signal sampling channel number, Q is the EEG signal sampling number, and the source signal matrix is S ═ S₁,s₂,…,s_N]_Q×NThe EEG sampling matrix is X ═ X₁,x₂,…,x_N]_Q×N，Y_Q×NFor an additive noise matrix, the following equation is given:

X＝S+Y (1)

the target is to obtain a source signal matrix S from a sampling matrix X so as to achieve the purpose of denoising; the method comprises the following specific steps:

step 1: carrying out centralization processing on the sampling matrix X:

X_O＝X-U^TH (2)

wherein

H[n]1,2, …, Q, N1, 2, …, N; x (m, n) is the nth voltage value collected by the mth channel;

step 2: calculating a covariance matrix of the sampling matrix X, and calculating an eigenvalue and an eigenvector of the sampling matrix X:

C＝E[XX^T] (3)

wherein E [. cndot. ] is a desired value; the covariance matrix C is a positive definite symmetric matrix, and the eigenvectors of the covariance matrix C are real numbers and are orthogonal; assuming that D is a vector formed by eigenvalues of the X covariance matrix C;

and step 3: for m-1, …, Q-1:

step 3.1: computing filter adaptation coefficients

Wherein 0<α<1，0<β，

And

respectively represent coefficient matrixes

The m +1 th and m-th row vectors of (a); x_O(m,: represents X)_OThe m-th row vector of (1);

step 3.2: calculating a filter coefficient w:

wherein w (m + 1:) and w (m:) represent the m +1 and m row vectors of w, respectively; d (m) denotes the m-th component of D; w (1) ═ 0, …, 0;

step 3.3: estimating the noise signal y:

y(m,:)＝w^T(m,:)D(m) (6)

superscript T denotes transpose; y (m,: means the Y mth row vector;

step 3.4: calculating a source signal:

an mth row evaluation value representing S, X (m:) being an Xmth row vector;

and 4, step 4: and finally, obtaining a source signal matrix S:

Images