阅读说明：本技术 一种状态相关的动态脑电信号中肌电伪迹的消除方法 (Method for eliminating myoelectric artifacts in state-related dynamic electroencephalogram signals ) 是由 刘爱萍 宋公正 陈勋 傅雪阳 吴枫 于 2020-06-22 设计创作，主要内容包括：本发明公开了一种状态相关的动态脑电信号中肌电伪迹的消除方法,其步骤包括：1、首先将收集到的脑电观测信号通过延时构造两个数据集；2、利用本发明提出的隐马尔科夫独立向量分析法进行动态的联合盲源分离,得到每个数据集在各个状态下的源信号矩阵和解混矩阵；3、选择脑电信号相对应的源信号矩阵和解混矩阵；4、依照自相关系数排序源信号矩阵中各个独立源成分,选择肌电噪声相关的独立源成分置零；5、盲源分离逆变换得到消除噪声后的干净脑电信号。本发明能在实际的动态环境中去除肌电噪声对脑电信号的影响,同时尽可能地保留脑电活动的信息不丢失,从而提高脑电信号的分析准确性,为脑电去噪提供一种新思路。(The invention discloses a method for eliminating myoelectric artifacts in state-related dynamic electroencephalogram signals, which comprises the following steps: 1. firstly, constructing two data sets by delaying collected electroencephalogram observation signals; 2. performing dynamic combined blind source separation by using a hidden Markov independent vector analysis method provided by the invention to obtain a source signal matrix and a demixing matrix of each data set in each state; 3. selecting a source signal matrix and a demixing matrix corresponding to the electroencephalogram signals; 4. sequencing each independent source component in the source signal matrix according to the autocorrelation coefficient, and selecting the independent source component related to the electromyographic noise to be set to zero; 5. and carrying out blind source separation inverse transformation to obtain a clean electroencephalogram signal after noise elimination. The method can remove the influence of the electromyographic noise on the electroencephalogram signal in the actual dynamic environment, and simultaneously keep the information of the electroencephalogram activity as far as possible without losing, thereby improving the analysis accuracy of the electroencephalogram signal and providing a new idea for the electroencephalogram denoising.)

1. A method for eliminating myoelectric artifacts in state-related dynamic electroencephalogram signals is characterized by comprising the following steps:

the method comprises the following steps: the method comprises the following steps of collecting and recording a pure brain electrical signal matrix of N channels by brain electrical measurement equipment, and recording as follows: s_EEG＝[s₁,s₂,...,s_n,...,s_N]∈R^N×T，1≤n≤N，s_nThe signal is a pure electroencephalogram signal of the nth channel, and T represents the total acquisition time;

collecting forearm muscles of both hands in contracting and relaxing fistThe electrical signal is denoted S_EMG∈R^N×T；

Electroencephalogram observation signal matrix X constructed by formula (1)_EEG＝[x₁,x₂,...,x_n,...,x_N]，x_nFor electroencephalogram observation signal matrix X_EEGObserving signals of the nth channel, wherein N is more than or equal to 1 and less than or equal to N:

X_EEG＝S_EEG+λ·X_EMG(1)

in the formula (1), lambda represents the degree of interference of the control myoelectric noise on the electroencephalogram signal; x_EMG∈R^N×TRepresents an electromyographic observation signal and has:

X_EMG＝[V₁S_EMG,V₂S_EMG,V₁S_EMG](2)

in the formula (2), V₁∈R^N×N，V₂∈R^N×NRepresenting two random mixing matrices;

when time t ∈ [0, t₁]∪[t₂,T]Temporal myoelectric observation signal X_EMGFrom a first random mixing matrix V₁Acting on muscle power supply signal matrix S_EMGGenerated and represented as a first blending mode;

when time t ∈ [ t ]₁,t₂]Temporal myoelectric observation signal X_EMGFrom a second random mixing matrix V₂Acting on muscle power supply signal matrix S_EMGGenerated and expressed as a second hybrid mode;

the signal-to-noise ratio SNR is obtained using equation (3):

in formula (3), RMS (-) represents a root mean square function;

step two: the electroencephalogram observation signal matrix X_EEGConstruction of K data sets X ═ X by delaying K-1 points₁,X₂,...,X_k,...,X_K]K is not less than 1 and not more than K, wherein X_kA brain electrical observation signal matrix representing the kth data set;

step three: using hidden Markov independent steeringCarrying out dynamic combined blind source separation on the K data sets X by a quantitative analysis method to obtain the kth data set X in different mixed modes_kBrain electrical source signal matrix S_kAnd an inverse mixing matrix W_k；

Step four: corresponding to each mixed mode, the kth brain electrical source signal matrix S_kZeroing the channel signal with a small middle autocorrelation coefficient to obtain an electroencephalogram source signal matrix with the zero electromyographic noise component

Step five: obtaining the kth de-noised electroencephalogram signal by using the formula (4)

In the formula (4), (W)_k)^-1Is the k-th inverse mixing matrix W_kThe inverse matrix of (c).

2. The electroencephalogram signal denoising method according to claim 1, wherein the third step is performed according to the following process:

step 3.1, recording the transition state of the hidden markov model as Q ═ Q₁,q₂,...,q_m,...,q_M}，q_mM is the mth transition state, M is more than or equal to 1 and less than or equal to M, and M is the total number of the transition states;

performing Viterbi algorithm on K data sets X to obtain a transition state sequence

using said transition state sequenceIdentify outA change in K data sets X blend modes;

step 3.2, in each mixed mode, performing combined blind source separation on the K data sets X to obtain K unmixed matrices W ═ W₁,W₂,...,W_k,...W_K]，W_kFor the kth data set X in the current mixing mode_kThe unmixing matrix of (a);

obtaining the kth data set X in the current mixed mode by using the formula (5)_kBrain electrical source signal matrix S_k：

S_k＝W_kX_k，k＝1,2,...,K (5)。

Technical Field

The invention relates to the field of medical signal processing, in particular to a method for removing noise in a neural signal in a complex environment.

Background

Electroencephalography (EEG) can non-invasively and conveniently observe and record brain activity by measuring electrical signals generated by neuronal activity, and has been widely used in neuroscience research, disease diagnosis, and medical health monitoring. However, because the brain electrical signal is weak in amplitude, it is often contaminated by various non-neurophysiological factors derived from the activity of the eye, heart and muscles. In practical long-term medical monitoring, these inevitable noises can significantly interfere with the recorded electroencephalogram signals, thereby adversely affecting subsequent analysis. Therefore, under the condition of not influencing real electroencephalogram data, the development of an effective noise removal algorithm has important significance. Compared with the electrooculogram noise and the electrocardio noise, the myoelectricity noise is more difficult to effectively remove due to complex characteristics of the myoelectricity noise, such as high amplitude, wide frequency spectrum and the like.

In order to remove electromyographic noise from electroencephalogram records, the blind signal separation and combined blind signal separation method draws great attention and obtains better effect. For example, an Independent Component Analysis (ICA) model is widely used for electroencephalogram denoising as a blind signal separation method. The method decomposes the electroencephalogram data into independent components, regards independent components with low autocorrelation as electromyographic noise, sets the electromyographic noise to zero, and finally reconstructs a source signal to achieve the purpose of removing the noise. Further, in order to meet the requirement of multi-data joint analysis, methods of joint blind signal separation such as canonicalculation analysis (CCA), Independent component analysis (IVA) and the like are proposed as reliable methods of electroencephalogram denoising. However, most of them are designed for electroencephalogram noise reduction under ideal conditions, and cannot meet the requirements of electroencephalogram noise reduction for complex changes in actual mobile medical monitoring. In practical situations, various myoelectric noises change with time along with the electroencephalogram activity, and various mixed modes may appear in the electroencephalogram record. The existing electroencephalogram signal denoising method is mostly suitable for static situations, namely, the model assumes that the mixed mode of electroencephalogram recording does not change in the whole observation process. When the traditional static electroencephalogram denoising method is applied to actual mobile medical monitoring, the electromyographic noise cannot be effectively removed, so that serious interference is caused to subsequent research on electroencephalogram recording. It is therefore desirable to develop new methods for eliminating noise under dynamically changing conditions.

Disclosure of Invention

The invention provides a method for eliminating electromyographic artifacts in state-related dynamic electroencephalogram signals in order to overcome the defects of the prior art, so that the influence of electromyographic noise on the electroencephalogram signals can be eliminated in the actual dynamic environment, and meanwhile, information of electroencephalogram activity is kept as far as possible without being lost, so that the analysis accuracy of the electroencephalogram signals is improved, and a new idea is provided for electroencephalogram denoising.

In order to achieve the purpose, the invention adopts the following technical scheme:

the invention relates to a method for eliminating myoelectric artifacts in state-related dynamic electroencephalogram signals, which is characterized by comprising the following steps of:

the method comprises the following steps: the method comprises the following steps of collecting and recording a pure brain electrical signal matrix of N channels by brain electrical measurement equipment, and recording as follows: s_EEG＝[s₁,s₂,...,s_n,...,s_N]∈R^N×T，1≤n≤N，s_nThe signal is a pure electroencephalogram signal of the nth channel, and T represents the total acquisition time;

collecting forearm electromyographic signals of both hands in contraction and relaxation of fist and recording as S_EMG∈R^N×T；

Electroencephalogram observation signal matrix X constructed by formula (1)_EEG＝[x₁,x₂,...,x_n,...,x_N]，x_nFor electroencephalogram observation signal matrix X_EEGObserving signals of the nth channel, wherein N is more than or equal to 1 and less than or equal to N:

X_EEG＝S_EEG+λ·X_EMG(1)

in the formula (1), λ represents a control muscleThe degree of interference of electrical noise with the electroencephalogram signal; x_EMG∈R^N×TRepresents an electromyographic observation signal and has:

X_EMG＝[V₁S_EMG,V₂S_EMG,V₁S_EMG](2)

in the formula (2), V₁∈R^N×N，V₂∈R^N×NRepresenting two random mixing matrices;

when time t ∈ [0, t₁]∪[t₂,T]Temporal myoelectric observation signal X_EMGFrom a first random mixing matrix V₁Acting on muscle power supply signal matrix S_EMGGenerated and represented as a first blending mode;

when time t ∈ [ t ]₁,t₂]Temporal myoelectric observation signal X_EMGFrom a second random mixing matrix V₂Acting on muscle power supply signal matrix S_EMGGenerated and expressed as a second hybrid mode;

obtaining signal-to-noise ratio using equation (3)^SNR：

In formula (3), RMS (-) represents a root mean square function;

step two: the electroencephalogram observation signal matrix X_EEGConstruction of K data sets X ═ X by delaying K-1 points₁,X₂,...,X_k,...,X_K]K is not less than 1 and not more than K, wherein X_kA brain electrical observation signal matrix representing the kth data set;

step three: performing dynamic joint blind source separation on the K data sets X by using a hidden Markov independent vector analysis method to obtain the kth data set X in different mixed modes_kBrain electrical source signal matrix S_kAnd an inverse mixing matrix W_k；

Step four: corresponding to each mixed mode, the kth brain electrical source signal matrix S_kZeroing the channel signal with a small middle autocorrelation coefficient to obtain an electroencephalogram source signal matrix with the zero electromyographic noise component

Step five: obtaining the kth de-noised electroencephalogram signal by using the formula (4)

In the formula (4), (W)_k)^-1Is the k-th inverse mixing matrix W_kThe inverse matrix of (c).

The electroencephalogram signal denoising method is also characterized in that the third step is carried out according to the following processes:

step 3.1, recording the transition state of the hidden markov model as Q ═ Q₁,q₂,...,q_m,...,q_M}，q_mM is the mth transition state, M is more than or equal to 1 and less than or equal to M, and M is the total number of the transition states;

performing Viterbi algorithm on K data sets X to obtain a transition state sequenceq′_tIs the transition state of the time t;

using said transition state sequenceIdentifying changes in K sets of data, X, of the hybrid mode;

step 3.2, in each mixed mode, performing combined blind source separation on the K data sets X to obtain K unmixed matrices W ═ W₁,W₂,...,W_k,...W_K]，W_kFor the kth data set X in the current mixing mode_kThe unmixing matrix of (a);

obtaining the kth data set X in the current mixed mode by using the formula (5)_kBrain electrical source signal matrix S_k：

S_k＝W_kX_k，k＝1,2,...,K (5)。

Compared with the traditional static blind separation method such as CCA, ICA and IVA, the method has the advantages that the method can not only remove the influence of myoelectric noise on the electroencephalogram in a dynamic change environment, but also well retain the electroencephalogram information, and has better denoising capability compared with a dynamic algorithm HMM-ICA. The performance of the method of the invention is improved as follows:

1. in the second step and the third step of the invention, a plurality of data sets are obtained through time delay, the observation data sets are subjected to state division by utilizing a Viterbi algorithm, and signals in the divided time periods are subjected to combined blind source separation to extract noise source signals. Compared with the traditional denoising methods such as ICA, CCA and IVA, the method has the advantages that the joint blind signal separation performed in a segmented mode meets the requirement of long-time dynamic monitoring in the actual health monitoring, the problem of noise removal changing under a complex environment is solved, and the precision and the accuracy of the noise removal are improved.

2. In the third step of the invention, an observation part of hidden Markov is set as an independent vector analysis model, when the method carries out combined blind source separation on signals of multiple data sets, the separated source signals are mutually independent in the same data set, and in different data sets, the corresponding source signals have the maximum correlation, namely, IVA simultaneously uses Second Order Statistics (SOS) and High Order Statistics (HOS). Compared with ICA and CCA methods, the IVA model is more effective in separating myoelectric noise, and loss of electroencephalogram signal information in the denoising process is reduced. The method combines the IVA and the state estimation, improves the accuracy of noise removal in the complex environment, and meets the requirement of electroencephalogram signal pretreatment in practical application.

Drawings

FIG. 1 is a flow chart of the method of the present invention;

FIG. 2a is a diagram of a real electroencephalogram signal with 19 channels for 10 seconds;

FIG. 2b is a diagram of a real electromyographic signal for 10 seconds;

fig. 2c is a schematic diagram of the observed signal at SNR 1.5, 19 channels for 10 seconds;

fig. 3 is a diagram of independent source signals and transfer paths obtained by dynamic blind separation in SNR 1.5, 19 channels for 10 seconds;

FIG. 4 is a schematic diagram of the denoised EEG signal and original observed signal of the present invention at SNR of 1.5 and T7 channel for 10 seconds;

FIG. 5 is a graph comparing RRMSE and ACC denoising performance of the method of the present invention and CCA, ICA, IVA and HMM-ICA methods.

Detailed Description

In this embodiment, as shown in fig. 1, a method for eliminating myoelectric artifacts in state-dependent dynamic electroencephalogram signals includes: firstly, delaying the collected electroencephalogram observation signals to obtain a plurality of data sets, and then carrying out combined blind source separation on the plurality of data sets by using a hidden Markov independent vector analysis method; obtaining a source signal matrix and an inverse mixing matrix of the observation signal; setting the independent source signals related to partial noise to zero according to the autocorrelation coefficient, and reserving an electroencephalogram independent signal source; and finally, carrying out blind source separation inverse transformation reconstruction to obtain the electroencephalogram signal with noise eliminated.

The following describes a specific embodiment by taking a semi-simulation electroencephalogram signal as an example and combining the accompanying drawings.

A semi-simulation data set is constructed through real electroencephalogram and electromyogram signal data, the implementation mode of the invention is described in detail, and the method is compared with ICA, CCA, IVA and HMM-ICA processing methods.

The method comprises the following steps: the method comprises the following steps of collecting and recording a pure brain electrical signal matrix of N channels by brain electrical measurement equipment, and recording as follows: s_EEG＝[s₁,s₂,...,s_n,...,s_N]∈R^N×T，1≤n≤N，s_nThe signal is a pure electroencephalogram signal of the nth channel, and T represents the total acquisition time;

collecting forearm electromyographic signals of both hands in contraction and relaxation of fist and recording as S_EMG∈R^N×T；

Electroencephalogram observation signal matrix X constructed by formula (1)_EEG＝[x₁,x₂,...,x_n,...,x_N]，x_nFor electroencephalogram observation signal matrix X_EEGObserving signals of the nth channel, wherein N is more than or equal to 1 and less than or equal to N:

X_EEG＝S_EEG+λ·X_EMG(1)

in the formula (1), lambda represents the degree of interference of the control myoelectric noise on the electroencephalogram signal; x_EMG∈R^N×TRepresents an electromyographic observation signal and has:

X_EMG＝[V₁S_EMG,V₂S_EMG,V₁S_EMG](2)

in the formula (2), V₁∈R^N×N，V₂∈R^N×NRepresenting two random mixing matrices;

when time t ∈ [0, t₁]∪[t₂,T]Temporal myoelectric observation signal X_EMGFrom a first random mixing matrix V₁Acting on muscle power supply signal matrix S_EMGGenerated and represented as a first blending mode;

when time t ∈ [ t ]₁,t₂]Temporal myoelectric observation signal X_EMGFrom a second random mixing matrix V₂Acting on muscle power supply signal matrix S_EMGGenerated and expressed as a second hybrid mode;

the signal-to-noise ratio SNR is obtained using equation (3):

in formula (3), RMS (-) represents a root mean square function;

in this embodiment, fig. 2a shows a clean electroencephalogram signal matrix with a time length T of 10s and N of 19 channels, which are acquired and recorded by an electroencephalogram measuring device. To obtain diverse electromyographic signals, twenty-three volunteers performed three types of muscle movements as required, namely (1) continue to clench for 10 s; (2) relaxing for 5s, and continuing to clench the fist for 5 s; (3) the user momentarily clenches the finger for a time less than 1S and then relaxes the finger, and the data is recorded as S_EMGAs shown in fig. 2 b. The data size is matched with the electroencephalogram signal. Next, the electroencephalogram signal and the electromyogram signal are acted on in different mixed modes at different time periods. Specifically, two mixing matrices V are provided₁，V₂. Wherein, V₁Acting at t ∈ [0,2.5 ]]And t ∈ [7.5,10 ]]，V₂Acting at t ∈ [2.5,7.5 ]]Root of Chinese characterConstructing an observed Signal X according to equation (1)_EMG＝[V₁S_EMG,V₂S_EMG,V₁S_EMG]，X_EEG＝S_EEG+λ×X_EMGAs shown in fig. 2 c.

Step two: observing the brain electrical signal matrix X_EEGConstruction of K data sets X ═ X by delaying K-1 points₁,X₂,...,X_k,...,X_K]K is not less than 1 and not more than K, wherein X_kRepresenting the electroencephalogram observed signal of the kth data set at time t;

step three: performing dynamic joint blind source separation on the K data sets X by using a hidden Markov independent vector analysis method to obtain the kth data set X in different mixed modes_kBrain electrical source signal matrix S_kAnd an inverse mixing matrix W_k：

Step 3.1, recording the transition state of the hidden markov model as Q ═ Q₁,q₂,...,q_m,...,q_M}，q_mM is the mth transition state, M is more than or equal to 1 and less than or equal to M, and M is the total number of the transition states;

performing Viterbi algorithm on K data sets X to obtain a transition state sequenceq′_tIs the transition state of the time t;

using said transition state sequenceIdentifying changes in K sets of data, X, of the hybrid mode;

step 3.2, in each mixed mode, performing combined blind source separation on the K data sets X to obtain K unmixed matrices W ═ W₁,W₂,...,W_k,...W_K]，W_kFor the kth data set X in the current mixing mode_kThe unmixing matrix of (a);

obtaining the kth data set X in the current mixed mode by using the formula (5)_kBrain electrical source signal matrix S_k：

S_k＝W_kX_k，k＝1,2,...,K (5)

In this example, the independent vector analysis model (IVA) is as follows:

S_k＝W_kX_k,k＝1....K, (6)

in the formula (6), W_kA solution mixing matrix, S, representing the kth data set_kFor the decomposed source signal matrix, let B ═ W_kK is 1, 2. Model assurance S_kEach independent from the other, and the corresponding sources between the data sets maintain correlation. The relationship between the decomposed source signal and the observed signal can be obtained according to the probability density function of the multivariate distribution, namely:

in the formula (7), the reaction mixture is,

an ith source representing a kth decomposed source signal matrix,representing the ith channel of the kth observed data set. sⁱRepresenting the ith SCV of the model. The invention realizes the identification of dynamic environment by Hidden Markov Model (HMM). Specifically, the state transition portion is set as follows, initial probability distribution pi, and state transition matrix a, where the transition matrix size is set to M × M, for controlling the model to transition between M states. As before, the hidden markov observation part is set to the IVA model, and the observation signal generation process follows the IVA model corresponding to each hidden state of the model. In order to solve the parameters set by the method of the invention, firstly, the log-likelihood function of the HMM model is given:

in the formula (8), T represents the observation data length, X (T) is a value of the electroencephalogram observation signal at time T, and X (T) ([ X ])₁(t),X₂(t),...,X_K(t)]。The value is q 'for the observation sequence of the whole model'_iM, i 1, 2. To solve the parameters of the model set, an optimization of the auxiliary function is requiredWherein, the notation theta is { pi, A, B }. The initial distribution of the model, pi, the state transition matrix, a, can be solved according to a forward-backward algorithm. The solution process for the solution mixture matrix is as follows, according to equation (6), hiding the state q'_tThe multivariate probability density function of the following observations can be expressed as:

and (3) obtaining an update formula of the unmixing matrix by derivation of the unmixing matrix of the formula:

in the formula (10), the compound represented by the formula (10),

to be in a transition state q_mThe unmixing matrix for the next kth data set, μ is the learning rate used to control the update rate. Gamma ray_m[t]Indicating that the observed signal is in a hidden state q 'at time t'_mThe probability of (c). Phi^k(s(t))＝[Φ^k(s¹(t)),Φ^k(s²(t)),...,Φ^k(s^N(t))]And is and

carrying out dynamic combined blind source separation on 2 data sets X by using a hidden Markov independent vector analysis method, and dividing the state of an observed signal by using a Viterbi algorithm to obtain an electroencephalogram source signal matrix S of the electroencephalogram observed signal matrix X_kAnd k is 1, 2. Due to the observation partThe second-order correlation of corresponding source signals among data sets is considered by the multivariate Gaussian distribution independent vector analysis method, so that the separated source signals are automatically arranged from large to small according to autocorrelation coefficients. Fig. 3 is an independent source signal obtained via a dynamic blind source separation process. The first line in the figure marks the transition path of the hidden state, so that the model clearly identifies different mixed modes. And in the given segment, completing the dynamic joint blind source separation.

Step four: corresponding to each mixed mode, the kth brain electrical source signal matrix S_kZeroing the channel signal with a small middle autocorrelation coefficient to obtain an electroencephalogram source signal matrix with the zero electromyographic noise component

In this embodiment, according to the physical characteristic that the self-correlation coefficient of the myoelectric noise is low, the independent signal source arranged at the tail end can be set to zero, so as to achieve the purpose of eliminating the myoelectric noise.

Step five: obtaining the kth de-noised electroencephalogram signal by using the formula (4)

In the formula (4), (W)_k)^-1Is the k-th inverse mixing matrix W_kThe inverse matrix of (c).

Fig. 4 is a comparison result of the noise-removed electroencephalogram signal obtained by reconstruction through blind source separation inverse transformation processing and other methods. The gray background is the original electroencephalogram observation signal, and the black thin line is the reconstructed electroencephalogram signal. It can be seen visually that the algorithm provided by the invention can remove myoelectric noise more thoroughly while preserving the electroencephalogram signal.

In order to quantitatively evaluate the effect of the present invention, the method of the present invention was compared with four algorithms of ICA, HMM-ICA, IVA and CCA for this purpose. Two performance indexes of a Relative Root Mean Square Error (RRMSE) and an Average Correlation Coefficient (ACC) are selected as evaluation indexes. The relative root mean square error is defined as follows:

the ACC is used for calculating the average value of the autocorrelation coefficients of all channels of the clean electroencephalogram and the denoised electroencephalogram. The smaller the RRMSE value is, the better the RRMSE value is, the smaller the difference between the de-noised signal and the original clean electroencephalogram signal is; the larger the value of ACC, the better, the larger the representation of the denoised signal is closer to the clean electroencephalogram signal, and the cleaner the noise is removed.

According to the steps, the simulation experiment is repeated for 100 times, RRMSE and ACC comparison graphs of various methods are drawn, as shown in FIG. 5, and as can be seen from FIG. 5, under the condition of various signal-to-noise ratios, the method disclosed by the invention is remarkably stronger than other comparison methods, and the method shows excellent noise elimination capability in a dynamic environment.

In conclusion, the invention designs a new algorithm, namely a hidden Markov independent vector analysis algorithm, can solve the problem of eliminating the electromyographic noise and can keep the electroencephalogram information as much as possible. The method is suitable for preprocessing the electroencephalogram signals and provides convenience for subsequent diagnosis and research. Compared with other methods, the method can obtain better denoising effect in a dynamic environment, solves the problem of dynamic noise removal of complex electroencephalogram signals in actual health monitoring, and has important significance for further developing long-term health monitoring.