Electroencephalogram signal denoising method and system based on MVMD and SOBI

Abstract

The invention relates to the technical field of electroencephalogram signal denoising, and provides an electroencephalogram signal denoising method based on MVMD and SOBI, comprising the following steps: acquiring an electroencephalogram signal, and performing multivariate variational mode decomposition on the electroencephalogram signal to obtain a plurality of intrinsic mode components corresponding to the electroencephalogram signal; performing second-order mode recognition on the intrinsic mode component to obtain a plurality of bottom layer source signals corresponding to the electroencephalogram signal; removing a bottom layer source signal corresponding to the noise; carrying out inverse operation of second-order mode recognition on the reserved bottom layer source signal to obtain a denoised intrinsic mode component; reconstructing based on the denoised intrinsic mode component to obtain a denoised electroencephalogram signal; the denoising method has the advantages that the requirement of the blind source separation technology for the input channels is met, meanwhile, the correlation between the channels is fully considered, and a good denoising effect can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of electroencephalogram (EEG) signal denoising, and more specifically, to an EEG signal denoising method and system based on MVMD and SOBI. Background Art

[0002] EEG signals are signals generated by amplifying and recording the weak bioelectricity of the human brain through EEG acquisition equipment. EEG signals are widely used in the field of biomedicine, such as sleep classification, epilepsy, concentration analysis, and diagnosis of depression. Because EEG signals are low-frequency, nonlinear, and non-stationary signals with a main frequency between 0.5 and 100 Hz and a signal amplitude of 5 to 300 μV, and the physiological activities of other parts of the human body will also generate corresponding electrical signals, EEG signals are very susceptible to various noise pollution, thereby generating various artifacts, such as the subject's electrocardiogram, electrooculogram, and electromyography, which will also cause corresponding artifacts.

[0003] In recent years, the development trend of medical devices is from clinical devices to portable devices. It is worth noting that the total number of channels obtained from portable devices is much less than that from clinical devices. That is, the total number of channels obtained by portable devices is usually less than 10.

[0004] The existing technology usually uses blind source separation technology to separate the noise in the EEG signal, but the existing blind source separation technology requires that the total number of channels of the EEG signal is sufficient, that is, the existing blind source separation technology is not suitable for portable devices whose total number of acquired channels is often less than 10.

[0005] Although EEG signals are widely used in the field of biomedicine, they are subject to a lot of noise interference, the most common of which are ECG artifacts, electrooculographic artifacts and electromyographic artifacts, which are difficult to be directly eliminated during signal acquisition, and these noises will affect the results of experiments and research, so EEG signal denoising is very important. Moreover, with the development trend of portable devices, obtaining effective EEG information from limited channels requires a signal processing method suitable for application on small channel data.

[0006] In summary, the prior art has the defect of being unable to obtain effective EEG information when the number of channels is small, that is, being unable to effectively denoise EEG signals with a small number of channels. Summary of the invention

[0007] In order to overcome the defect of the above-mentioned prior art that it is unable to effectively denoise EEG signals with a small number of channels, the present invention provides an EEG signal denoising method based on MVMD and SOBI, which can still maintain good denoising effect when the number of channels of the EEG signal is small.

[0008] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0009] Acquire an EEG signal, perform multivariate variational modal decomposition on the EEG signal, and obtain a plurality of intrinsic modal components corresponding to the EEG signal;

[0010] Performing second-order pattern recognition on the intrinsic mode component to obtain a plurality of underlying source signals corresponding to the EEG signal;

[0011] Remove the underlying source signal corresponding to the noise;

[0012] Perform the inverse operation of second-order pattern recognition on the retained underlying source signal to obtain the denoised intrinsic mode component;

[0013] A denoised electroencephalogram signal is reconstructed based on the denoised intrinsic mode component.

[0014] The present invention also proposes an EEG signal denoising system based on MVMD and SOBI for implementing the above-mentioned EEG signal denoising method based on MVMD and SOBI, the system comprising:

[0015] An EEG signal decomposition module is used to obtain an EEG signal, perform multivariate variational modal decomposition on the EEG signal, and obtain a number of intrinsic modal components corresponding to the EEG signal;

[0016] A source signal acquisition module, used for performing second-order pattern recognition on the intrinsic mode component to obtain a plurality of underlying source signals corresponding to the EEG signal;

[0017] A denoising module is used to remove the underlying source signal corresponding to the noise;

[0018] A first reconstruction module is used to perform an inverse operation of second-order pattern recognition on the retained underlying source signal to obtain a denoised intrinsic mode component;

[0019] The second reconstruction module is used to reconstruct the denoised EEG signal based on the denoised intrinsic mode component.

[0020] The present invention also proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the above-mentioned EEG signal denoising method based on MVMD and SOBI.

[0021] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0022] Multivariate Variational Mode Decomposition (MVMD) is used to decompose the EEG signal into several intrinsic mode components to meet the requirements of blind source separation technology for the number of input channels. Then, the second-order blind identification (SOBI) blind source separation technology is used to denoise the intrinsic mode components, which fully considers the correlation between channels. It has the advantages of meeting the requirements of blind source separation technology for input channels while fully considering the correlation between channels, achieving good denoising effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of the EEG signal denoising method based on MVMD and SOBI proposed in Example 1;

[0024] Figure 2 This is a schematic diagram of the standard electrode placement method proposed in Example 1;

[0025] Figure 3 This is the overall framework diagram of the EEG signal denoising system based on MVMD and SOBI proposed in Example 3. DETAILED DESCRIPTION

[0026] The accompanying drawings are only used for illustrative purposes and should not be construed as limiting the present embodiment;

[0027] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0028] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0029] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0030] Example 1

[0031] This embodiment proposes an EEG signal denoising method based on MVMD and SOBI. Figure 1 Schematic diagram of the process of the EEG signal denoising method based on MVMD and SOBI in this embodiment;

[0032] The present embodiment proposes an EEG signal denoising method based on MVMD and SOBI, comprising the following steps:

[0033] S1: Acquire an EEG signal, perform multivariate variational modal decomposition on the EEG signal, and obtain a number of intrinsic modal components corresponding to the EEG signal;

[0034] S2: performing second-order pattern recognition on the intrinsic mode component to obtain a plurality of underlying source signals corresponding to the EEG signal;

[0035] S3: remove the underlying source signal corresponding to the noise;

[0036] S4: performing an inverse operation of second-order pattern recognition on the retained underlying source signal to obtain a denoised intrinsic mode component;

[0037] S5: Reconstruct a denoised EEG signal based on the denoised intrinsic mode component.

[0038] In the specific implementation process, Multivariate Variational Mode Decomposition (MVMD) is used to decompose the EEG signal into several intrinsic mode components to meet the requirements of blind source separation technology for the number of input channels, and then the second-order pattern recognition (SOBI) blind source separation technology is used to denoise the intrinsic mode component (IMF), which fully considers the correlation between channels. It has the advantages of meeting the requirements of blind source separation technology for input channels while fully considering the correlation between channels, and achieves good denoising effect.

[0039] In an optional embodiment, when collecting the EEG signal, a plurality of electrodes are used to collect the brain signal at a plurality of collection moments to form an EEG signal containing a plurality of channels, wherein one electrode corresponds to one channel.

[0040] As an example, Figure 2 This is a schematic diagram of the standard electrode placement method proposed in this embodiment. Figure 2 The standard electrode placement method prescribed by the International Electroencephalographic Society is demonstrated.

[0041] In an optional embodiment, the calculation expression of the eigenmode component includes:

[0042]

[0043] x(t)=[x1(t),…,x c (t)…,x C (t)] T

[0044] In the formula, x(t) represents the EEG signal, C represents the total number of channels of the EEG signal x(t), and x c (t) represents the cth component of the EEG signal x(t), and also represents the data corresponding to the cth channel of the EEG signal x(t); u k,c(t) represents x c The kth eigenmode component of (t), also represented by x c The kth component of (t), K represents x c (t) is the total number of eigenmode components.

[0045] In an optional embodiment, before performing second-order pattern recognition on the intrinsic mode component, the objective function is to minimize the sum of the bandwidths of all components of all channels of the electroencephalogram signal, and the objective function is iteratively solved. When the objective function value is minimum or the number of iterations reaches a preset number, the iteration is stopped to obtain the intrinsic mode component with the minimum bandwidth sum. When performing second-order pattern recognition, the intrinsic mode component with the minimum bandwidth sum is used as the operation object for operation;

[0046] The expression of the objective function includes:

[0047]

[0048] u k (t) = {u k,1 (t),…,u k,c (t),…u k,C (t)}

[0049] In the formula, ω k Indicates u k (t) corresponds to the center frequency, ||·|| represents the L2 norm, represents the partial derivative operation with respect to time, e represents a natural constant, j represents an imaginary number, and t represents time.

[0050] In an optional embodiment, the expression of the underlying source signal includes:

[0051] S(t)=X(t)QU

[0052] S(t)=[S1(t),…,S n (t)]

[0053] R z (τ i )=Q T R X (τ i )Q=D i

[0054] R X (τ i )=E(X(t+τ i ) T X(t)

[0055] i∈{1,2,3,…,a}

[0056] Where X(t) represents the intrinsic mode matrix composed of all the intrinsic mode components in any channel of the EEG signal x(t), S(t) represents the source matrix composed of n underlying source signals corresponding to X(t), n represents the total number of elements in the source matrix, S n (t) represents the nth element of the source matrix S(t), and also represents the nth underlying source signal; Q represents the whitening matrix corresponding to X(t); D i represents a diagonal matrix, T represents the transposed symbol; τ i represents the i-th sampling moment; a represents a positive integer, which also represents the total number of sampling moments; U represents the number of sampling moments. Z (τ1)~R Z (τ k ) is a unitary matrix that is jointly diagonalized.

[0057] In an optional embodiment, the calculation expression of the whitening matrix Q includes:

[0058]

[0059] In the formula, ∧ represents the X (0), P represents the diagonal matrix composed of the singular values ​​of ∧, and T represents the transposed sign.

[0060] In an optional embodiment, when removing the underlying source signal corresponding to the noise, the fuzzy entropy of the underlying source signal and the Pearson correlation coefficient of the EEG signal are calculated, and based on the fuzzy entropy and the Pearson correlation coefficient, it is determined whether the underlying source signal meets the denoising condition. If so, the underlying source signal is removed, otherwise the underlying source signal is retained.

[0061] In an optional embodiment, the expression of the denoising condition includes:

[0062] cc+f>r

[0063] Where cc represents the Pearson correlation coefficient of the EEG signal, f represents the fuzzy entropy of the underlying source signal, and r represents the preset threshold.

[0064] Example 2

[0065] This embodiment is based on the EEG signal denoising method based on MVMD and SOBI proposed in Embodiment 1, and proposes the following specific implementation examples:

[0066] The basic idea of ​​this embodiment includes:

[0067] S1: Perform multivariate variational mode decomposition on the original EEG signal to obtain enough multivariate IMFs.

[0068] S2: Perform second-order pattern recognition on multivariate IMFs with the same channels to obtain the underlying source signal.

[0069] S3: Calculate the fuzzy entropy and Pierce coefficient to set the threshold to remove the underlying source signal related to the noise.

[0070] S4: Perform inverse SOBI reconstruction on the denoised underlying source signal into a clean multivariate IMF.

[0071] S5: Use clean multivariate IMF to reconstruct clean multi-channel EEG signals for subsequent applications.

[0072] The specific implementation steps of this embodiment include:

[0073] Assume x(t) is a multivariate input. Let C be the number of channels of x(t), and let x c (t) becomes the cth component of x(t). So x(t)=[x1(t)…x C (t)] T Here, the MVMD method takes x(t) as input and decomposes it into various multivariate components. Let K be the total number of multivariate components x(t), let u k (t) are these multivariate components. Let u k,c (t) Become u k The cth component of (t).

[0074] So it can be expressed as u k (t)=[u k,1 (t)…u k,C (t)] T .

[0075] The sum of the bandwidths of all components of all channels is minimized and, in addition, the resulting components are decomposed to achieve a perfect reconstruction of the input. That is:

[0076] Therefore, the goal of the variational optimization problem is to minimize the sum of the bandwidths of all components of all channels to satisfy the exact perfect reconstruction constraint of the input. That is:

[0077]

[0078] The total number of linear equality constraints in the above equation is equal to the total number of channels. Let λ c (t) is the Lagrange multiplier. Let u be when the number of iterations is 0 k,c (t) is expressed as and and They are and λ c (t) is the Fourier transform of

[0079] The corresponding augmented Lagrangian function is:

[0080]

[0081] The above optimization problem can be solved by the ADMM method. That is, each MIMF and its bandwidth are continuously updated during the iteration process until the termination condition of the algorithm is met. Let u at the n+1th iteration be k The corresponding Fourier transform of (t) is expressed as u at the n+1th iteration k The center frequency corresponding to (t) is expressed as but, and The expressions include:

[0082]

[0083]

[0084] Thus, the multivariate IMF of all channels is obtained.

[0085] The SOBI algorithm is an ICA (Independent component analysis) algorithm based on second-order statistics. Therefore, it is also a type of BSS (Blind source separation) algorithm. In particular, the SOBI algorithm estimates the mixing matrix and the source matrix by finding an approximation to the diagonal of the covariance matrix. Here, the SOBI algorithm does not require any information about the distribution of the source signal. Therefore, the separation performance does not depend on whether the source signal obeys a Gaussian distribution. Moreover, the SOBI algorithm does not have any constraints on the mixing matrix as long as the mixing matrix is ​​full rank. Therefore, it can achieve BSS.

[0086] The SOBI algorithm mainly includes four steps. They are pre-whitening the signal, calculating the covariance matrix of the original signal, finding the whitening matrix, and estimating the mixing matrix and the source matrix. The specific steps are as follows:

[0087] Step 1: Pre-whiten the signal. Let m be the total number of channels. Let n be the total number of sources. Let T be the total number of measurements. Let p be the total number of features. Let S = [s1…s n ] is a source matrix. X=[x1…x m ] is a p×T mixing matrix. Assume A is an n×T mixing matrix. So X=SA. Pre-whitening is to transform X into another matrix so that the correlation matrix is ​​a diagonal matrix.

[0088] Let Q be the whitening matrix. Let Z be the transformed matrix. So we get Z = XQ. Let R Z (τ), R X(τ) and R S (τ) are the correlation matrices of Z, X, and S respectively. Let τ1,…,τ k is the sampling time, let D i From i=1,…,a is a diagonal matrix. Then, we have:

[0089] R X (τ i )=E(X(t+τ i ) Τ X(t))=A Τ R S (τ i )A i=1,…,k

[0090] in addition

[0091] R Z (τ i )=E(Z(t+τ i ) Τ Z(t))=Q T R X (τ i )Q=D i i=1,…,k

[0092] Because D i is a diagonal matrix, so Z is whitened.

[0093] Step 2: Calculate the covariance matrix of the original signal; assume R Z (0) is the identity matrix. So we have:

[0094] R Z (0) = E((S(t)AQ) T (S(t)AQ))=Q T A T R S (0)AQ=I

[0095] Because A Τ R S (0)A=R X (0), R X (0) is symmetric. Let P and Λ(0) be singular and diagonal matrices, respectively, and apply the singular value decomposition to R X (0). So

[0096] R X (0) = PΛ(0)P T =A Τ R S (0)A

[0097] Step 3: Find the whitening matrix; Q Τ It can be found by:

[0098] Q T PΛ(0)P T Q=I

[0099] Assume Q is a unitary matrix. So we get:

[0100]

[0101] Step 4: Estimate the mixing matrix and source matrix; find R Z (τ i )i=1,…,k, let U be the obtained unitary matrix. Then the estimates of A and S are as follows:

[0102] A=U T Q #

[0103] S=XQU

[0104] The superscript # indicates pseudo-inverse.

[0105] Calculate the fuzzy entropy f of each underlying source signal and the Pearson correlation coefficient cc with the original signal. So r = cc + f. By setting a threshold, discard the underlying source signals with r higher than the threshold, thus achieving a denoising effect.

[0106] The existing variational mode decomposition (VMD) is an adaptive, completely non-recursive mode variation and signal processing method, which overcomes the problems of endpoint effect and modal component aliasing in the empirical mode decomposition method. VMD can self-determine the number of modal function decompositions and reduce the non-stationarity of time series with high complexity and strong nonlinearity. VMD can determine the number of signal decompositions based on the number of self-determined modal function decompositions, and then adaptively match the optimal center frequency and limited bandwidth of each modal function, and can achieve effective separation of modal function components and division of signal frequency domain, and then reconstruct the signal by selecting the modal function to achieve the effect of denoising.

[0107] Existing BBS technologies such as ICA and CCA (canonical correlation analysis) can detect and separate the underlying source signals of EEG signals, and remove artifact-related source signals in the subsequent reconstruction step to achieve a denoising effect.

[0108] Disadvantages of the prior art: EEG signal denoising method based on variational modal decomposition (VMD), which improves the problems of modal aliasing and endpoint effect by performing VMD decomposition and noise reduction on EEG signals. When using VMD to process EEG signals, the number of signal decompositions can be determined based on the self-determined number of modal function decompositions, and then the optimal center frequency and limited bandwidth of each modal function can be adaptively matched, and the effective separation of modal function components and the division of signal frequency domain can be achieved. Then, the signal is reconstructed by selecting the modal function, thereby achieving the denoising effect. However, because VMD can only input channel signals one by one when processing multi-channel data, the correlation between channels is not considered.

[0109] BBS technology, such as CCA, aims to study the connection between two groups of variables, that is, to study the linear correlation between the linear combination of one group of variables and the linear combination of another group of variables, that is, the Pearson correlation coefficient. The Pearson correlation coefficient is used to measure the degree of correlation between two variables (X and Y), and its value is between -1 and 1. The closer its absolute value is to 1, the higher the linear correlation between X and Y; the closer it is to 0, the lower the linear correlation between X and Y. It is worth noting that electromyographic signals and electrooculographic signals are broadband signals similar to Gaussian noise. Therefore, the correlation between these signals and their delayed versions is very small, so the underlying source signals with small correlation coefficients are removed by setting a threshold. The underlying source signals of EEG signals can be detected and separated, but the BBS technology requires that the number of observed signals is greater than the number of source signals. When the EEG signal is a few-channel signal, it is difficult to infer the underlying source signal from a small number of observed signals. That is, it is only effective when the number of channels is large enough.

[0110] In summary, compared with the prior art, the present application considers the correlation between channels through multivariate variational mode decomposition, solves the convex optimization problem of minimizing the sum of the bandwidths of all modes on all channels under the premise of satisfying multiple constraints of accurate reconstruction of signals of all channels, and decomposes the multi-channel signal into multiple multivariate modulated oscillations. Thus, enough IMFs are obtained to meet the input requirements of blind source separation technology. SOBI is used to identify and discard the underlying source signals related to the noise, thereby achieving a better denoising effect on less-channel data.

[0111] Example 3

[0112] This embodiment proposes an EEG signal denoising system based on MVMD and SOBI, which is used to implement the EEG signal denoising method based on MVMD and SOBI proposed in Embodiment 1.

[0113] Figure 3 4 is an overall framework diagram of the EEG signal denoising system based on MVMD and SOBI in this embodiment.

[0114] The EEG signal denoising system based on MVMD and SOBI comprises:

[0115] An EEG signal decomposition module is used to obtain an EEG signal, perform multivariate variational modal decomposition on the EEG signal, and obtain a number of intrinsic modal components corresponding to the EEG signal;

[0116] A source signal acquisition module, used for performing second-order pattern recognition on the intrinsic mode component to obtain a plurality of underlying source signals corresponding to the EEG signal;

[0117] A denoising module is used to remove the underlying source signal corresponding to the noise;

[0118] A first reconstruction module is used to perform an inverse operation of second-order pattern recognition on the retained underlying source signal to obtain a denoised intrinsic mode component;

[0119] The second reconstruction module is used to reconstruct the denoised EEG signal based on the denoised intrinsic mode component.

[0120] This embodiment proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the EEG signal denoising method based on MVMD and SOBI described in Example 1.

[0121] It can be understood that the EEG signal denoising system and computer device based on MVMD and SOBI in this embodiment are used to implement the method of Example 1. The options in the above-mentioned Example 1 are also applicable to this embodiment, so they will not be repeated here.

[0122] The same or similar reference numerals correspond to the same or similar components;

[0123] The terms used to describe the positional relationship in the drawings are only used for illustrative purposes and should not be construed as limiting the present embodiment;

[0124] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for denoising EEG signals based on MVMD and SOBI, characterized in that: The following steps are involved: Acquire an EEG signal, perform multivariate variational modal decomposition on the EEG signal, and obtain a plurality of intrinsic modal components corresponding to the EEG signal; Performing second-order pattern recognition on the intrinsic mode component to obtain a plurality of underlying source signals corresponding to the EEG signal; Remove the underlying source signal corresponding to the noise; Perform the inverse operation of second-order pattern recognition on the retained underlying source signal to obtain the denoised intrinsic mode component; A denoised electroencephalogram signal is reconstructed based on the denoised intrinsic mode component.

2. The method for denoising an EEG signal based on MVMD and SOBI according to claim 1, characterized in that: When collecting the EEG signal, a plurality of electrodes are used to collect the brain signal at a plurality of collection moments to form an EEG signal containing a plurality of channels, wherein one electrode corresponds to one channel.

3. The method for denoising an EEG signal based on MVMD and SOBI according to claim 2, characterized in that: The calculation expression of the eigenmode component includes: x(t)=[x1(t),…,x c (t)…,x C (t)] T In the formula, x(t) represents the EEG signal, C represents the total number of channels of the EEG signal x(t), and x c (t) represents the cth component of the EEG signal x(t), and also represents the data corresponding to the cth channel of the EEG signal x(t); u k,c (t) represents x c The kth eigenmode component of (t), also represented by x c The kth component of (t), K represents x c (t) is the total number of eigenmode components.

4. The method for denoising an EEG signal based on MVMD and SOBI according to claim 3, characterized in that: Before performing second-order pattern recognition on the intrinsic mode component, the objective function is to minimize the sum of the bandwidths of all components of all channels of the electroencephalogram signal, and the objective function is iteratively solved. When the objective function value is minimum or the number of iterations reaches a preset number, the iteration is stopped to obtain the intrinsic mode component with the minimum bandwidth sum. When performing second-order pattern recognition, the intrinsic mode component with the minimum bandwidth sum is used as the operation object for operation; The expression of the objective function includes: u k (t)={u k,1 (t),…,u k,c (t),…u k,C (t)} In the formula, ω k Indicates u k (t) corresponds to the center frequency, ||·|| represents the L2 norm, represents the partial derivative operation with respect to time; e represents a natural constant, j represents an imaginary number, and t represents time.

5. The method for denoising an EEG signal based on MVMD and SOBI according to claim 2, characterized in that: The expression of the underlying source signal includes: S(t)=X(t)QU S(t)=[S1(t),…,S n (t)] R Z (t i )=Q T R X (t i )Q=D i R X (t i )=E(X(t+τ i ) T X(t)) i∈{1,2,3,…,a} Where X(t) represents the intrinsic mode matrix composed of all the intrinsic mode components in any channel of the EEG signal x(t), S(t) represents the source matrix composed of n underlying source signals corresponding to X(t), n represents the total number of elements in the source matrix, S n (t) represents the nth element of the source matrix S(t), and also represents the nth underlying source signal; Q represents the whitening matrix corresponding to X(t); D i represents a diagonal matrix, T represents the transposed symbol; τ i represents the i-th sampling moment; a represents a positive integer, which also represents the total number of sampling moments; U represents the number of sampling moments. Z (τ1)~R Z (τ k ) is a unitary matrix that is jointly diagonalized.

6. The method for denoising an EEG signal based on MVMD and SOBI according to claim 5, characterized in that: The calculation expression of the whitening matrix Q includes: In the formula, ∧ represents the X (0), P represents the diagonal matrix composed of the singular values ​​of ∧, and T represents the transposed sign.

7. The method for denoising an EEG signal based on MVMD and SOBI according to any one of claims 1 to 6, characterized in that: When removing the underlying source signal corresponding to the noise, the fuzzy entropy of the underlying source signal and the Pearson correlation coefficient of the EEG signal are calculated, and based on the fuzzy entropy and the Pearson correlation coefficient, it is determined whether the underlying source signal meets the denoising condition. If so, the underlying source signal is removed, otherwise the underlying source signal is retained.

8. The method for denoising an EEG signal based on MVMD and SOBI according to claim 7, characterized in that: The expression of the denoising condition includes: co+f>r Where cc represents the Pearson correlation coefficient of the EEG signal, f represents the fuzzy entropy of the underlying source signal, and r represents the preset threshold.

9. An EEG signal denoising system based on MVMD and SOBI, used to implement the EEG signal denoising method based on MVMD and SOBI according to any one of claims 1 to 8, characterized in that: include: An EEG signal decomposition module is used to obtain an EEG signal, perform multivariate variational modal decomposition on the EEG signal, and obtain a number of intrinsic modal components corresponding to the EEG signal; A source signal acquisition module, used for performing second-order pattern recognition on the intrinsic mode component to obtain a plurality of underlying source signals corresponding to the EEG signal; A denoising module is used to remove the underlying source signal corresponding to the noise; A first reconstruction module is used to perform an inverse operation of second-order pattern recognition on the retained underlying source signal to obtain a denoised intrinsic mode component; The second reconstruction module is used to reconstruct the denoised EEG signal based on the denoised intrinsic mode component.

10. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that: When the computer-readable instructions are executed by the processor, the processor executes the steps of the EEG signal denoising method based on MVMD and SOBI as claimed in any one of claims 1 to 8.

Images