阅读说明：本技术 基于mvmd-cca的ssvep脑电信号识别方法 (MVMD-CCA-based SSVEP electroencephalogram signal identification method ) 是由 翟弟华 王康 胡乐云 夏元清 戴荔 邹伟东 张金会 闫莉萍 崔冰 孙中奇 郭泽华 于 2020-05-26 设计创作，主要内容包括：本发明公开了基于MVMD-CCA的SSVEP脑电信号识别方法,能够将脑电信号分解为多个多元调制分量,降低脑电信号中的非相关脑活动与伪迹的影响,提高分类精度。包括如下步骤：采集多通道稳态视觉诱发电位SSVEP脑电信号作为待识别脑电信号。设定带分解的分量个数K,构造变分问题,并采用ADMM算法求解所述变分问题,将所述待识别脑电信号分解为K个多元调制分量。据诱发所述待识别脑电信号的视觉刺激频率,定义参考信号。求解特定刺激频率f<Sub>i</Sub>的加权关联系数。取最大加权关联系数对应的频率即为所述待识别脑电信号的诱发刺激频率。(The invention discloses an SSVEP electroencephalogram signal identification method based on MVMD-CCA, which can decompose an electroencephalogram signal into a plurality of multivariate modulation components, reduce the influence of non-relevant brain activities and artifacts in the electroencephalogram signal and improve the classification precision. The method comprises the following steps: and collecting the multi-channel steady-state visual evoked potential SSVEP electroencephalogram signal as an electroencephalogram signal to be identified. Setting the number K of components with decomposition, constructing a variation problem, solving the variation problem by adopting an ADMM algorithm, and decomposing the electroencephalogram signal to be identified into K multivariate modulation components. Defining a reference according to the visual stimulation frequency inducing the EEG signal to be identifiedA signal. Solving for a specific stimulation frequency f i The weighted correlation coefficient of (2). And taking the frequency corresponding to the maximum weighting correlation coefficient as the evoked stimulus frequency of the electroencephalogram signal to be identified.)

1. The SSVEP electroencephalogram signal identification method based on MVMD-CCA is characterized by comprising the following steps:

s1, collecting a multichannel steady-state visual evoked potential (SSVEP) electroencephalogram signal as an electroencephalogram signal to be identified;

s2, setting the number K of components with decomposition, constructing a variational problem, solving the variational problem by adopting an ADMM algorithm, and decomposing the electroencephalogram signal to be identified into K multivariate modulation components;

s3, defining a reference signal according to the visual stimulation frequency for inducing the electroencephalogram signal to be identified;

s4, solving specific stimulation frequency f_iThe weighted correlation coefficient of (a);

and S5, taking the frequency corresponding to the maximum weighted correlation coefficient as the evoked stimulation frequency of the electroencephalogram signal to be identified.

2. The method of claim 1, wherein the multichannel steady-state visual evoked potential (SSVEP) electroencephalogram signal is used as an electroencephalogram signal to be identified, and specifically comprises:

and adopting a multi-lead electrode cap to acquire the EEG signals under SSVEP stimulation as the EEG signals to be identified.

3. The method according to claim 1 or 2, characterized in that the set number of components K of the band decomposition constructs a variational problem, in particular:

the electroencephalogram signal to be identified is x (t), and the number of channels is C; the number of the components to be decomposed is set to be K, and K multivariate modulation components u are obtained by decomposition_k(t), K1, 2.., K, from which a variational problem is constructed, comprising the steps of:

s201, solving a multivariate modulation vibration signal u by using Hilbert transform_k(t) analytic signal

WhereinRespectively are analytic signals of a 1 st channel to a C th channel; u. of_k,1(t)～u_k,C(t) are each u_k(t) data of 1 st to C th channels; (t) is an impulse function; j is a complex symbol;

s202, estimating the center frequency of the analytic signal to be w_kShifting the frequency spectrum of the analytic signal to a baseband by adopting a frequency shift operation:

s203, L using gradient²Norm structure constraint variantThe problems are as follows:

wherein u is_k,c(t) is u in the k component of the decomposition_kData of the c-th channel of (t), x_c(t) is data of the c channel of the electroencephalogram signal x (t) to be identified;

s204, converting the constraint variation problem into an unconstrained variation optimization problem by introducing a secondary penalty term alpha and a Lagrange multiplier lambda:

wherein the content of the first and second substances,<>represents the product of two elements; lambda [ alpha ]_cIs composed of

4. The method of claim 3, wherein the variational problem is solved using an ADMM algorithm to decompose the electroencephalogram signal to be identified into K multivariate modulation components, comprising the steps of:

s2001, initializationThe iteration number n is set to 0;is u_k,c(t) the corresponding initial value of the iteration,for each center frequency w_k(K ═ 1, 2.., K) corresponding to the initial iteration value,for lagrange multiplier function lambda_cCorresponding iteration initial values;

s2002, for the (n + 1) th iteration, each multivariate modulation component u_k,c(t), K is updated as follows:

whereinIs u_k,c(t) the corresponding (n + 1) th iteration value; i is a designation for the subscript k,is to include all u when i < k_i,c(t) a corresponding set of n +1 th iteration values,is intended to include all u when i ≧ k_i,c(t) a corresponding set of nth iteration values;is w_kThe value of the nth iteration of (a),to include all w_iA corresponding set of nth iteration values;is λ_cThe corresponding nth iteration value;

the solved frequency domain updating formula is as follows:

wherein w is a frequency domain variable;

s2003, aiming at each central frequency w_k(k＝1,2, 1, K) with the following iterations:

the obtained frequency domain updating formula is as follows:

s2004, Lagrange multiplier function lambda_cCorresponding n +1 th iteration valueThe update is as follows:

s2005, repeating iteration according to S2002-S2005 until the stop condition of the iteration is satisfied:

finally K multivariate modulation components u are obtained_k(t)，k＝1,2,...,K。

5. The method according to claim 4, characterized in that said defining a reference signal according to the frequency of visual stimulation inducing said electroencephalogram signal to be identified is:

according to the N visual stimulus frequencies f inducing SSVEP_iWhere i ═ 1,2, …, N, defines the reference signal as:

wherein N is_hIndicating the number of harmonics.

6. As claimed in claimThe method of claim 5, wherein the solution for the specific stimulation frequency f_iThe weighted correlation coefficient of (a) is specifically:

s401, aiming at specific stimulation frequency f_i(i ═ 1,2, …, N), multivariate modulation component u is solved using classical correlation analysis CCA algorithm_k(t), K ═ 1, 2.., K, and reference signal Y_iMaximum typical correlation coefficient of 1,2, …, N

S402, calculating stimulation frequency f_iWeighted correlation coefficient of:

Technical Field

The invention relates to the technical field of pattern recognition, in particular to an SSVEP electroencephalogram signal recognition method based on MVMD-CCA.

Background

The brain-computer interface is a technology for directly utilizing signals generated by human brain activities to communicate with external environment equipment without depending on the pathways such as peripheral nerve, muscle tissues and the like. In recent years, brain-computer interfaces have gradually become research hotspots in the fields of brain science, biomedicine, artificial intelligence and the like, and are paid attention to research in various fields in the world, and related research plans are started in many countries successively. Brain-computer interface technology has great potential for development in a number of areas: in the field of rehabilitation medicine, the rehabilitation training device can help patients with the diseases such as amyotrophic lateral sclerosis and cerebral apoplexy to perform rehabilitation training, and is helpful for the patients to recover nervous consciousness; in the military field, the individual combat capability of soldiers can be enhanced by controlling the military exoskeleton, and methods for controlling external equipment such as unmanned aerial vehicles, unmanned vehicles and the like can also be provided; in the field of entertainment and living, the household electronic equipment can be controlled by people, and the life quality of people is improved.

Steady-state Visual Evoked potentials (SSVEPs) are commonly used input signals for brain-computer interface devices. When the human eye is subjected to a higher frequency of visual stimulation, periodic rhythmic activity, SSVEP, occurs in the occipital region of the cerebral cortex. The main indexes of the brain-computer interface based on the SSVEP are signal identification precision and Information Transfer Rate (ITR), and the key point of the system is the feature extraction and classification identification of the SSVEP electroencephalogram signals.

At present, an algorithm based on CCA and variants thereof is widely applied to identification of SSVEP electroencephalogram signals, wherein the most typical algorithm is an FBCCA algorithm proposed by Chenshinggai and other people, and the accuracy of about 92 percent and the information transmission rate of 151 bits/min are achieved in an online experiment. At present, the recognition rate of the SSVEP has a further space for improvement.

Spontaneous electroencephalogram activity and artifacts occurring in the process of recording the SSVEP electroencephalogram signals can affect the recognition performance based on the CCA algorithm, and the extraction of sub-bands related to the SSVEP can reduce the influence of irrelevant brain activity and artifacts, and the FBCCA algorithm is also based on the point.

At present, the identification aiming at the SSVEP electroencephalogram signals is greatly influenced by non-related brain activities and artifacts in the electroencephalogram signals, and the classification precision is not high.

Disclosure of Invention

In view of this, the invention provides an SSVEP electroencephalogram signal identification method based on MVMD-CCA, which can decompose an electroencephalogram signal into multiple multivariate modulation components, reduce the influence of non-related brain activities and artifacts in the electroencephalogram signal, and improve classification accuracy.

In order to achieve the purpose, the technical scheme of the invention is as follows: the SSVEP electroencephalogram signal identification method based on MVMD-CCA comprises the following steps:

and S1, acquiring the multichannel steady-state visual evoked potential SSVEP electroencephalogram signal as the electroencephalogram signal to be identified.

S2, setting the number K of components with decomposition, constructing a variational problem, solving the variational problem by adopting an ADMM algorithm, and decomposing the electroencephalogram signal to be identified into K multivariate modulation components.

And S3, defining a reference signal according to the visual stimulation frequency for inducing the electroencephalogram signal to be recognized.

S4, solving specific stimulation frequency f_iIs weighted offAnd (4) a joint coefficient.

And S5, taking the frequency corresponding to the maximum weighted correlation coefficient as the evoked stimulation frequency of the electroencephalogram signal to be identified.

Further, the multichannel steady-state visual evoked potential SSVEP electroencephalogram signal is used as an electroencephalogram signal to be identified, and the method specifically comprises the following steps: and adopting a multi-lead electrode cap to acquire the EEG signals under SSVEP stimulation as the EEG signals to be identified.

Further, the number K of the components with decomposition is set, and the variation problem is constructed, specifically: the electroencephalogram signal to be identified is x (t), and the number of channels is C; the number of the components to be decomposed is set to be K, and K multivariate modulation components u are obtained by decomposition_k(t), K1, 2.., K, from which a variational problem is constructed, comprising the steps of:

s201, solving a multivariate modulation vibration signal u by using Hilbert transform_k(t) analytic signal

WhereinRespectively are analytic signals of a 1 st channel to a C th channel; u. of_k,1(t)～u_k,C(t) are each u_k(t) data of 1 st to C th channels; (t) is an impulse function; j is a complex symbol.

S202, estimating the center frequency of the analytic signal to be w_kShifting the frequency spectrum of the analytic signal to a baseband by adopting a frequency shift operation:

s203, L using gradient²The norm construction constraint variation problem is as follows:

wherein u is_k,c(t) is u in the k component of the decomposition_kData of the c-th channel of (t), x_cAnd (t) is data of the c channel of the electroencephalogram signal x (t) to be identified.

S204, converting the constraint variation problem into an unconstrained variation optimization problem by introducing a secondary penalty term alpha and a Lagrange multiplier lambda:

wherein the content of the first and second substances,<>represents the product of two elements; lambda [ alpha ]_cIs composed of

Further, the variation problem is solved by adopting an ADMM algorithm, the electroencephalogram signal to be identified is decomposed into K multivariate modulation components, and the method specifically comprises the following steps:

s2001, initializationThe iteration number n is set to 0;is u_k,c(t) the corresponding initial value of the iteration,for each center frequency w_k(K ═ 1, 2.., K) corresponding to the initial iteration value,for lagrange multiplier function lambda_cCorresponding iteration initial values;

s2002, for the (n + 1) th iteration, each multivariate modulation component u_k,c(t), K is updated as follows:

whereinIs u_k,c(t) the corresponding (n + 1) th iteration value; i is a designation for the subscript k,is to include all u when i < k_i,c(t) a corresponding set of n +1 th iteration values,is intended to include all u when i ≧ k_i,c(t) a corresponding set of nth iteration values;is w_kThe value of the nth iteration of (a),to include all w_iA corresponding set of nth iteration values;is λ_cThe corresponding nth iteration value;

the solved frequency domain updating formula is as follows:

wherein w is a frequency domain variable;

s2003, aiming at each central frequency w_kThe n +1 th iteration of (K ═ 1, 2.., K) is updated as follows:

the obtained frequency domain updating formula is as follows:

s2004, Lagrange multiplier function lambda_cCorresponding n +1 th iteration valueThe update is as follows:

s2005, repeating iteration according to S2002-S2005 until the stop condition of the iteration is satisfied:

finally K multivariate modulation components u are obtained_k(t)，k＝1,2,...,K。

Further, according to the visual stimulation frequency for inducing the electroencephalogram signal to be identified, a reference signal is defined, which specifically comprises the following steps:

according to the N visual stimulus frequencies f inducing SSVEP_iWhere i ═ 1,2, …, N, defines the reference signal as:

wherein N is_hIndicating the number of harmonics.

Further, the specific stimulation frequency f is solved_iThe weighted correlation coefficient of (a) is specifically:

s401, aiming at specific stimulation frequency f_i(i ═ 1,2, …, N), multivariate modulation component u is solved using classical correlation analysis CCA algorithm_k(t), K ═ 1, 2.., K, and reference signal Y_iMaximum typical correlation coefficient of 1,2, …, N

S402, calculating stimulation frequency f_iWeighted correlation coefficient of:

has the advantages that:

the SSVEP electroencephalogram signal identification method based on MVMD-CCA provided by the embodiment of the invention combines the MVMD algorithm and the CCA algorithm, wherein the MVMD is used as a decomposition algorithm for processing multichannel, nonlinear and non-stable signals, has natural advantages in the aspect of analysis and processing of electroencephalogram signals, and the advantages of the two algorithms can be effectively played by combining the MVMD algorithm and the CCA algorithm. The MVMD algorithm can decompose an original electroencephalogram signal into a plurality of multivariate modulation components, and defined weighting correlation coefficients can reduce the influence of irrelevant components on identification and reduce the influence of irrelevant brain activities and artifacts in the signal to a certain extent. The quality of the multivariate modulation component decomposed by the MVMD algorithm is high, and the classification precision is obviously improved by combining the recognition result of the CCA algorithm compared with that of the FBCCA algorithm.

Drawings

Fig. 1 is a schematic diagram of an SSVEP electroencephalogram signal identification method based on MVMD-CCA according to an embodiment of the present invention.

Detailed Description

The invention is described in detail below by way of example with reference to the accompanying drawings.

The invention provides an SSVEP electroencephalogram signal identification method based on MVMD-CCA, the principle of which is shown in figure 1, and the method comprises the following steps:

and S1, acquiring the multichannel steady-state visual evoked potential SSVEP electroencephalogram signal as the electroencephalogram signal to be identified. In the embodiment of the invention, a multi-lead electrode cap is adopted to collect the EEG signals under SSVEP stimulation as the EEG signals to be identified.

S2, setting the number K of components with decomposition, constructing a variational problem, solving the variational problem by adopting an ADMM algorithm, and decomposing the electroencephalogram signal to be identified into K multivariate modulation components.

And S3, defining a reference signal according to the visual stimulation frequency for inducing the electroencephalogram signal to be recognized.

S4, solving specific stimulation frequency f_iThe weighted correlation coefficient of (2).

And S5, taking the frequency corresponding to the maximum weighted correlation coefficient as the evoked stimulation frequency of the electroencephalogram signal to be identified.

Wherein S2 adopts an improved variable modal decomposition MVMD algorithm to construct a variational problem, S4 adopts a typical correlation analysis CCA algorithm to solve a specific stimulation frequency f_iAlthough MVMD is determined as a decomposition algorithm for processing multichannel, nonlinear and non-stationary signals at present, the MVMD has no application in the aspect of analysis and processing of electroencephalogram signals, the analyzed MVMD has natural advantages in the aspect of analysis and processing of electroencephalogram signals, the advantages of the two algorithms can be effectively played by combining the MVMD with a CCA algorithm, and a good identification effect on SSVEP electroencephalogram signals can be generated.

In the embodiment of the present invention, in step S2, the number K of components with decomposition is set to construct a variation problem, specifically:

the electroencephalogram signal to be identified is x (t), and the number of channels is C; the number of the components to be decomposed is set to be K, and K multivariate modulation components u are obtained by decomposition_k(t), K1, 2.., K, from which a variational problem is constructed, comprising the steps of:

s201, solving a multivariate modulation vibration signal u by using Hilbert transform_k(t) analytic signal

WhereinRespectively are analytic signals of a 1 st channel to a C th channel; u. of_k,1(t)～u_k,C(t) are each u_k(t) data of 1 st to C th channels; (t) is an impulse function; j is a complex symbol.

S202, estimating the center frequency of the analytic signal to be w_kShifting the frequency spectrum of the analytic signal to a baseband by adopting a frequency shift operation:

s203, L using gradient²The norm construction constraint variation problem is as follows:

wherein u is_k,c(t) is u in the k component of the decomposition_kData of the c-th channel of (t), x_cAnd (t) is data of the c channel of the electroencephalogram signal x (t) to be identified.

S204, converting the constraint variation problem into an unconstrained variation optimization problem by introducing a secondary penalty term alpha and a Lagrange multiplier lambda:

wherein the content of the first and second substances,<>represents the product of two elements; lambda [ alpha ]_cIs composed of

In the step S2, the ADMM algorithm is used to solve the variational problem, and the electroencephalogram signal to be identified is decomposed into K multivariate modulation components, specifically including the following steps:

s2001, initializationThe iteration number n is set to 0;is u_k,c(t) the corresponding initial value of the iteration,for each center frequency w_k(K ═ 1, 2.., K) corresponding to the initial iteration value,for lagrange multiplier function lambda_cCorrespond toThe initial value of iteration of (1);

s2002, for the (n + 1) th iteration, each multivariate modulation component u_k,c(t), K is updated as follows:

whereinIs u_k,c(t) the corresponding (n + 1) th iteration value; i is a designation for the subscript k,is to include all u when i < k_i,c(t) a corresponding set of n +1 th iteration values,is intended to include all u when i ≧ k_i,c(t) a corresponding set of nth iteration values;is w_kThe value of the nth iteration of (a),to include all w_iA corresponding set of nth iteration values;is λ_cThe corresponding nth iteration value;

the solved frequency domain updating formula is as follows:

wherein w is a frequency domain variable;

s2003, aiming at each central frequency w_kThe n +1 th iteration of (K ═ 1, 2.., K) is updated as follows:

the obtained frequency domain updating formula is as follows:

s2004, Lagrange multiplier function lambda_cCorresponding n +1 th iteration valueThe update is as follows:

s2005, repeating iteration according to S2002-S2005 until the stop condition of the iteration is satisfied:

finally K multivariate modulation components u are obtained_k(t)，k＝1,2,...,K。

In the embodiment of the present invention, in step S3, a reference signal is defined according to a visual stimulation frequency for inducing an electroencephalogram signal to be recognized, and specifically, the reference signal is defined as follows:

according to the N visual stimulus frequencies f inducing SSVEP_iWhere i ═ 1,2, …, N, defines the reference signal as:

wherein N is_hIndicating the number of harmonics.

In the embodiment of the present invention, in step S4, the specific stimulation frequency f is solved_iThe weighted correlation coefficient of (a) is specifically:

s401, aiming at specific stimulation frequency f_i(i ═ 1,2, …, N), multivariate modulation component u is solved using classical correlation analysis CCA algorithm_k(t), K ═ 1, 2.., K, and reference signalY_iMaximum typical correlation coefficient of 1,2, …, N

S402, calculating stimulation frequency f_iWeighted correlation coefficient of:

the table 1 is a comparison table of simulation results of the FBCCA and the MVMD-CCA, wherein the accuracy of the SSVEP electroencephalogram recognition algorithm based on the MVMD-CCA is obviously improved in the same reference data set compared with the accuracy of the algorithm based on the FBCCA, and the effectiveness of the algorithm is verified.

TABLE 1

In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.