PSO-EEMD-ICA preprocessing method for electroencephalogram signal denoising

Abstract

The invention relates to an electroencephalogram signal preprocessing method based on particle swarm optimization (PSO), ensemble empirical mode decomposition (EEMD) and independent component analysis (ICA), and aims to improve the quality of electroencephalogram signals and enhance analyzability of the signals. The method comprises the following steps: firstly, carrying out EEMD (Ensemble Empirical Mode Decomposition) on an original electroencephalogram signal, generating a plurality of noise auxiliary signals by adding white noise with different intensities, carrying out EMD on each noise auxiliary signal, and then averaging intrinsic mode functions (IMF) of all the noise auxiliary signals to obtain a final IMF, thereby reducing the problem of mode aliasing and improving the reliability of the electroencephalogram signal. Useful components and noise components are preliminarily separated out; secondly, optimizing EEMD parameters by using a PSO algorithm; a particle swarm is initialized, each particle represents a possible parameter combination (such as Gaussian white noise standard deviation and noise adding times), a fitness function is defined, a quality index of a signal is taken as a target, positions and speeds of the particles are iteratively updated, the parameter combination is optimized, and finally optimal parameters are applied to perform EEMD decomposition, so that an optimized IMF is obtained. Therefore, the complexity of manual parameter adjustment is avoided, and the decomposition accuracy and stability are improved. Then, the sample entropy of each IMF is calculated, a sample entropy threshold value is set, the IMFs with the sample entropy values higher than the threshold value are screened out, IMF components with low information content are effectively removed, effective components with high information content are reserved, and the analyzability of the signals is further improved. And finally, combining the screened effective IMF component with the original electroencephalogram signal to generate a virtual multi-channel signal, performing Fast ICA (Independent Component Analysis), separating out independent electroencephalogram signal components, further removing noise, and improving the purity and the signal-to-noise ratio of the signal. Through the steps, the quality of the electroencephalogram signals can be remarkably improved, and a solid foundation is provided for subsequent signal analysis and application.

Description

Technical Field

[0001] The present invention relates to the field of electroencephalogram (EEG) signal processing, and in particular to a PSO-EEMD-ICA preprocessing method for EEG signal denoising. Background Art

[0002] Electroencephalogram (EEG) signals are important physiological signals reflecting brain neural activity and are widely used in fields such as brain science research, clinical diagnosis, and brain-computer interfaces. However, EEG signals are often affected by various noise and artifacts, such as myoelectric noise, oculoscopic artifacts, and environmental noise, which degrade signal quality and hinder subsequent analysis and processing. Therefore, how to effectively remove noise from EEG signals and extract useful information is a key topic in current EEG signal processing research.

[0003] Existing EEG denoising methods include wavelet transform, empirical mode decomposition (EMD), and independent component analysis (ICA), but these methods still have certain limitations when dealing with complex noise and nonlinear signals. Ensemble empirical mode decomposition (EEMD), as an improved method of EMD, can effectively alleviate the problem of modal aliasing by adding a white noise auxiliary signal, but its decomposition results depend on the noise intensity and the number of additions. Particle swarm optimization (PSO), a global optimization algorithm, can effectively optimize the parameters in EEMD and improve the decomposition effect. Fast independent component analysis (FastICA) can quickly separate independent components and remove noise. Summary of the Invention

[0004] The purpose of this paper is to provide a PSO-EEMD-ICA preprocessing method for EEG signal denoising, which aims to improve the signal decomposition effect by optimizing the EEMD decomposition parameters, combine sample entropy screening and FastICA analysis, remove noise from EEG signals, and extract useful EEG signal components.

[0005] To achieve the above object, the present invention provides a PSO-EEMD-ICA preprocessing method for EEG signal denoising, comprising the following steps:

[0006] S1: EEMD decomposition: Perform ensemble empirical mode decomposition (EEMD) on the EEG signal to decompose the signal into several intrinsic mode function (IMF) components;

[0007] S2: PSO optimization: The particle swarm optimization (PSO) algorithm is used to optimize the Gaussian white noise standard deviation and the number of noise additions in EEMD to minimize the ratio of the envelope entropy to the mutual information entropy between the separated signal and the actual signal.

[0008] S3: Sample entropy screening: Use sample entropy as a threshold to screen the IMF components and obtain valid components;

[0009] S4: Virtual multi-channel signal generation: combining the effective component with the observation signal to generate a virtual multi-channel signal;

[0010] S5: FastICA analysis: Use the fast independent component analysis (FastICA) method to perform independent component analysis on the virtual multi-channel signal until each group of vibration data is completely separated to obtain the denoised EEG signal.

[0011] Preferably, the EEG signal is decomposed using ensemble empirical mode decomposition (EEMD), and the specific process is as follows:

[0012] S1.1: Add white noise: Add white noise of different intensities to the original EEG signal to generate multiple noise auxiliary signals;

[0013] The formula is:

[0014] x i (t) = x(t) + n i (t)

[0015] Among them, x(t) is the original signal, n i (t) is the white noise in the i-th noise auxiliary signal.

[0016] S1.2: EMD decomposition: Perform empirical mode decomposition (EMD) on each noise auxiliary signal to obtain a series of IMFs;

[0017] The formula is:

[0018]

[0019] Among them, the IMF ij (t) is the jth IMF component of the i-th noise auxiliary signal, r i (t) is the residual.

[0020] S1.3: IMF averaging: average the IMFs of all noise-assisted signals to obtain the final IMF;

[0021] The formula is:

[0022]

[0023] Where N is the number of noise auxiliary signals.

[0024] Preferably, the particle swarm optimization (PSO) algorithm is used to optimize the Gaussian white noise standard deviation and the number of noise additions in the EEMD. The specific process is as follows:

[0025] S2.1: Initialize the particle swarm: each particle represents a possible parameter combination (Gaussian white noise standard deviation σσ and number of noise additions NN);

[0026] S2.2: Define the fitness function: take the signal quality index (such as the ratio of envelope entropy to mutual information entropy) as the target;

[0027] The fitness function is expressed as:

[0028]

[0029] Among them, H env is the envelope entropy of the optimized IMF, and IMI is the mutual information entropy between the optimized IMF and the original signal.

[0030] S2.3: Iteratively update particles: It iteratively update the position and velocity of particles to optimize the parameter combination;

[0031] The update formula is:

[0032] v i (t+1)=wv i (t)+c1r1(p i -x i (t))+c2r2(gx i (t))

[0033] x i (t+1)=x i (t)+v i (t+1)

[0034] where vi(t) is the velocity of the ith particle at the tth iteration, xi(t) is the position, pi is the historical optimal position of the ith particle, gg is the global optimal position, ww is the inertia weight, c1 and c2 are acceleration constants, and r1 and r2 are random numbers.

[0035] S2.4: Apply optimal parameters: Apply the optimal parameters to perform EEMD decomposition and obtain the optimized IMF.

[0036] Preferably, the sample entropy is used to screen the IMF components. The specific process is as follows:

[0037] S3.1: Calculate sample entropy: Calculate the sample entropy of each IMF;

[0038] The calculation formula of sample entropy is:

[0039]

[0040] Where m is the embedding dimension, r is the tolerance, N is the data length, A is the number of matching template pairs, and B is the number of all template pairs.

[0041] S3.2: Screening IMF: Set the sample entropy threshold T and screen out IMFs whose sample entropy values ​​are higher than the threshold;

[0042] The screening formula is:

[0043] IMF eff ={IMF j |SampEn(IMF j )>T}

[0044] Preferably, the virtual multi-channel signal is generated by combining the effective component with the observation signal, and the specific process is as follows:

[0045] S4.1: Combined signal: Combine the filtered valid IMF components with the original EEG signal to generate a virtual multi-channel signal;

[0046] The combined signal is represented as:

[0047] X virtual ={IMF eff ,x(t)}

[0048] Among them, the IMF eff is the effective IMF component, and x(t) is the original EEG signal.

[0049] Preferably, the fast independent component analysis (FastICA) method is used to perform independent component analysis on the virtual multi-channel signal, and the specific process is as follows:

[0050] S5.1: Inputting the FastICA algorithm: inputting the virtual multi-channel signal into the FastICA algorithm;

[0051] S5.2: Independent component analysis: Perform independent component analysis to separate independent EEG signal components;

[0052] The goal of FastICA is to find an unmixing matrix W such that:

[0053] S=WX

[0054] Where S is the independent component matrix and X is the virtual multi-channel signal matrix.

[0055] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0056] Reducing modal aliasing: EEMD effectively reduces the modal aliasing problem in traditional EMD methods by adding a white noise auxiliary signal. PSO further optimizes the parameters of EEMD, making the decomposition results more accurate and stable. Automated parameter optimization: The PSO algorithm is used to automatically optimize the noise intensity and number of additions in EEMD, avoiding the tedious manual parameter adjustment and improving the automation and efficiency of the decomposition process. Enhancing the analyzability of the signal: Through sample entropy screening, the low-information IMF components are effectively eliminated, while the high-information effective components are retained, further improving the analyzability of the signal. Virtual multi-channel signal generation: The effective IMF components are combined with the original signal to generate a virtual multi-channel signal, which enhances the diversity and independence of the signal and provides a richer information source for subsequent independent component analysis. Fast independent component analysis: The FastICA method is used to perform independent component analysis on the virtual multi-channel signal, which can quickly separate the independent EEG signal components, further remove noise, and improve the purity of the signal. Wide Applicability: This method is suitable for preprocessing various EEG signals. It can be used not only in brain science research, but also in clinical diagnosis and brain-computer interface, and has broad application prospects. Improved Processing Efficiency: By combining advanced algorithms such as PSO, EEMD, and FastICA, the overall processing flow is efficient and robust, and can complete complex signal preprocessing tasks in a relatively short time. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 : Flowchart of the EEG signal preprocessing method based on PSO-EEMD-ICA. DETAILED DESCRIPTION

[0058] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0059] like Figure 1 As shown, this embodiment proposes a method for denoising an EEG signal, comprising the following steps:

[0060] S1: EEMD decomposition: Perform ensemble empirical mode decomposition (EEMD) on the EEG signal to decompose the signal into several intrinsic mode function (IMF) components;

[0061] S2: PSO optimization: The particle swarm optimization (PSO) algorithm is used to optimize the Gaussian white noise standard deviation and the number of noise additions in EEMD to minimize the ratio of the envelope entropy to the mutual information entropy between the separated signal and the actual signal.

[0062] S3: Sample entropy screening: Use sample entropy as a threshold to screen the IMF components and obtain valid components;

[0063] S4: Virtual multi-channel signal generation: combining the effective component with the observation signal to generate a virtual multi-channel signal;

[0064] S5: FastICA analysis: Use the fast independent component analysis (FastICA) method to perform independent component analysis on the virtual multi-channel signal until each group of vibration data is completely separated to obtain the denoised EEG signal.

[0065] S1.1: Add white noise: Add white noise of different intensities to the original EEG signal to generate multiple noise auxiliary signals;

[0066] The formula is:

[0067] x i (t) = x(t) + n i (t)

[0068] Among them, x(t) is the original signal, n i (t) is the white noise in the i-th noise auxiliary signal.

[0069] S1.2: EMD decomposition: Perform empirical mode decomposition (EMD) on each noise auxiliary signal to obtain a series of IMFs;

[0070] The formula is:

[0071]

[0072] Among them, the IMF ij (t) is the jth IMF component of the i-th noise auxiliary signal, r i (t) is the residual.

[0073] S1.3: IMF averaging: average the IMFs of all noise-assisted signals to obtain the final IMF;

[0074] The formula is:

[0075]

[0076] Where N is the number of noise auxiliary signals.

[0077] S2.1: Initialize the particle swarm: each particle represents a possible parameter combination (Gaussian white noise standard deviation σσ and number of noise additions NN);

[0078] S2.2: Define the fitness function: take the signal quality index (such as the ratio of envelope entropy to mutual information entropy) as the target;

[0079] The fitness function is expressed as:

[0080]

[0081] Among them, H envis the envelope entropy of the optimized IMF, and IMI is the mutual information entropy between the optimized IMF and the original signal.

[0082] S2.3: Iteratively update particles: It iteratively update the position and velocity of particles to optimize the parameter combination;

[0083] The update formula is:

[0084] v i (t+1)=wv i (t)+c1r1(p i -x i (t))+c2r2(gx i (t))

[0085] x i (t+1)=x i (t)+v i (t+1)

[0086] where vi(t) is the velocity of the ith particle at the tth iteration, xi(t) is the position, pi is the historical optimal position of the ith particle, gg is the global optimal position, ww is the inertia weight, c1 and c2 are acceleration constants, and r1 and r2 are random numbers.

[0087] S2.4: Apply optimal parameters: Apply the optimal parameters to perform EEMD decomposition and obtain the optimized IMF.

[0088] S3.1: Calculate sample entropy: Calculate the sample entropy of each IMF;

[0089] The calculation formula of sample entropy is:

[0090]

[0091] Where m is the embedding dimension, r is the tolerance, N is the data length, A is the number of matching template pairs, and B is the number of all template pairs.

[0092] S3.2: Screening IMF: Set the sample entropy threshold T and screen out IMFs whose sample entropy values ​​are higher than the threshold;

[0093] The screening formula is:

[0094] IMF eff ={IMF j |SampEn(IMF j )>T}

[0095] S4.1: Combined signal: Combine the filtered valid IMF components with the original EEG signal to generate a virtual multi-channel signal;

[0096] The combined signal is represented as:

[0097] X virtual ={IMF eff ,x(t)}

[0098] Among them, the IMF eff is the effective IMF component, and x(t) is the original EEG signal.

[0099] S5.1: Inputting the FastICA algorithm: inputting the virtual multi-channel signal into the FastICA algorithm;

[0100] S5.2: Independent component analysis: Perform independent component analysis to separate independent EEG signal components;

[0101] The goal of FastICA is to find an unmixing matrix W such that:

[0102] S=WX

[0103] Where S is the independent component matrix and X is the virtual multi-channel signal matrix.

[0104] This paper presents a PSO-EEMD-ICA preprocessing method for EEG signal denoising. By optimizing EEMD decomposition parameters and combining sample entropy screening with FastICA analysis, it effectively removes noise from EEG signals and extracts useful EEG signal components. This method improves the signal-to-noise ratio of EEG signals and enhances their analyzability, providing strong support for subsequent processing and analysis of EEG signals.

[0105] In this example, the original EEG signal is first subjected to EEMD decomposition. A white noise auxiliary signal is added to reduce modal aliasing and initially separate the useful and noise components. Secondly, the PSO algorithm is used to optimize the EEMD parameters, automatically adjusting the noise intensity and number of additions, avoiding the tedious manual parameter adjustment and improving the accuracy and stability of the decomposition. Then, the sample entropy of each IMF is calculated to screen out high-information IMFs, effectively eliminating low-information components and enhancing the analyzability of the signal. Finally, the screened-out valid IMF components are combined with the original EEG signal to generate a virtual multi-channel signal, which is then subjected to FastICA analysis to quickly separate the independent EEG signal components, further remove noise, and improve signal purity and signal-to-noise ratio. Through these steps, the method of the present invention significantly improves the quality of EEG signals and provides a solid foundation for subsequent signal analysis and application. Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention and are not intended to limit the implementation methods of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to exhaustively enumerate all implementation methods here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A PSO-EEMD-ICA preprocessing method for EEG signal denoising, comprising the following steps: S1: EEMD decomposition: Perform ensemble empirical mode decomposition (EEMD) on the EEG signal to decompose the signal into several intrinsic mode function (IMF) components; S2: PSO optimization: The particle swarm optimization (PSO) algorithm is used to optimize the Gaussian white noise standard deviation and the number of noise additions in EEMD to minimize the ratio of the envelope entropy to the mutual information entropy between the separated signal and the actual signal. S3: Sample entropy screening: Use sample entropy as a threshold to screen the IMF components and obtain valid components; S4: Virtual multi-channel signal generation: combining the effective component with the observation signal to generate a virtual multi-channel signal; S5: FastICA analysis: Use the fast independent component analysis (FastICA) method to perform independent component analysis on the virtual multi-channel signal until each group of vibration data is completely separated to obtain the denoised EEG signal.

2. The method according to 1, characterized in that The EEMD decomposition step (S1) comprises: S1.1: Add white noise: Add white noise of different intensities to the original EEG signal to generate multiple noise auxiliary signals; The formula is: x i (t)=x(t)+n i (t) Among them, x(t) is the original signal, n i (t) is the white noise in the i-th noise auxiliary signal. S1.2: EMD decomposition: Perform empirical mode decomposition (EMD) on each noise auxiliary signal to obtain a series of IMFs; The formula is: Among them, the IMF ij (t) is the jth IMF component of the i-th noise auxiliary signal, r i (t) is the residual. S1.3: IMF averaging: average the IMFs of all noise-assisted signals to obtain the final IMF; The formula is: Where N is the number of noise auxiliary signals.

3. The method according to 2, characterized in that The PSO optimization step (S2) includes: S2.1: Initialize the particle swarm: each particle represents a possible parameter combination (Gaussian white noise standard deviation σσ and number of noise additions NN); S2.2: Define the fitness function: take the signal quality index (such as the ratio of envelope entropy to mutual information entropy) as the target; The fitness function is expressed as: Among them, H env is the envelope entropy of the optimized IMF, and IMI is the mutual information entropy between the optimized IMF and the original signal. S2.3: Iteratively update particles: It iteratively update the position and velocity of particles to optimize the parameter combination; The update formula is: v i (t+1)=wv i (t)+c1r1(p i -x i (t))+c2r2(g-x i (t)) x i (t+1)=x i (t)+v i (t+1) where vi(t) is the velocity of the ith particle at the tth iteration, xi(t) is the position, pi is the historical optimal position of the ith particle, gg is the global optimal position, ww is the inertia weight, c1 and c2 are acceleration constants, and r1 and r2 are random numbers. S2.4: Apply optimal parameters: Apply the optimal parameters to perform EEMD decomposition and obtain the optimized IMF.

4. The method according to 3, characterized in that The sample entropy screening step (S3) includes: S3.1: Calculate sample entropy: Calculate the sample entropy of each IMF; The calculation formula of sample entropy is: Where m is the embedding dimension, r is the tolerance, N is the data length, A is the number of matching template pairs, and B is the number of all template pairs. S3.2: Screening IMF: Set the sample entropy threshold T and screen out IMFs whose sample entropy values ​​are higher than the threshold; The screening formula is: IMF eff ={IMF j ∣SampEn(IMF j )>T} 5. The method according to 4, characterized in that The virtual multi-channel signal generating step (S4) includes: S4.1: Combined signal: Combine the filtered valid IMF components with the original EEG signal to generate a virtual multi-channel signal; The combined signal is represented as: X virtual ={IMF eff ,x(t)} Among them, the IMF eff is the effective IMF component, and x(t) is the original EEG signal.

6. The method according to 5, characterized in that The FastICA separation step (S5) comprises: S5.1: inputting the FastICA algorithm: inputting the virtual multi-channel signal into the FastICA algorithm; S5.2: Independent component analysis: Perform independent component analysis to separate independent EEG signal components; The goal of FastICA is to find an unmixing matrix W such that: S=WX Where S is the independent component matrix and X is the virtual multi-channel signal matrix.

7. The method according to 6, characterized in that The PSO algorithm simulates the foraging behavior of a flock of birds to find the optimal solution, ensuring that the ratio of the envelope entropy to the mutual information entropy of the separated signal and the actual signal is minimized.

8. The method according to item 7, characterized in that The sample entropy is an indicator for measuring signal complexity. A higher sample entropy value generally indicates that the signal contains more useful information.

9. The method according to 8, characterized in that The EEMD method generates multiple noise auxiliary signals by adding white noise of different intensities, and performs EMD decomposition on each noise auxiliary signal to obtain a series of intrinsic mode functions (IMFs).

10. The method according to 9, characterized in that The virtual multi-channel signal includes a denoised EEG signal component.

Images