EEG signal denoising method and system combining SSA and JADE

Abstract

The invention discloses an electroencephalogram signal denoising method and system combining SSA and JADE, and the method comprises the following steps: carrying out the preprocessing of an original electroencephalogram signal, extracting a signal mode of oscillation change related to artifacts, combining a preset threshold value with an extreme point of the signal mode, screening and extracting a segment only containing an electrooculogram signal, and carrying out the denoising of the electroencephalogram signal. The rest part is zeroed to serve as an input signal for blind source separation; the method comprises the following steps: decomposing a single-channel electroencephalogram signal into a plurality of components, constructing a trajectory matrix and performing SSA, preliminarily separating main components and noise components of the signal, and further processing the components obtained by singular spectrum analysis and decomposition by using a JADE algorithm; separating a signal and noise by using high-order statistics, processing a source signal of an artifact related source, and suppressing a noise component; and performing inverse JADE on the processed source signal matrix and the mixed matrix, and re-synthesizing the de-noised single-channel electroencephalogram signal. Compared with the prior art, under the condition that the noise level is high, noise and useful signals can still be effectively distinguished, and therefore the signal processing effect is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and more specifically, to an electroencephalogram (EEG) signal denoising method and system combining SSA and JADE. Background Art

[0002] EEG signals are signals generated by weak bioelectricity in the brain that are recorded and amplified by EEG acquisition equipment. These signals are widely used in the biomedical field, such as emotion recognition, Alzheimer's disease detection, activity recognition, and autism diagnosis. However, EEG signals are low-frequency, nonlinear, and non-stationary signals with a frequency range of 0.5 to 100 Hz and a signal amplitude of 5 to 300 μV. Since other physiological activities of the human body also generate electrical signals, EEG signals are extremely susceptible to noise interference, forming various artifacts, such as electrocardiogram, electrooculogram, and electromyogram. Although EEG signals are widely used in the biomedical field, their susceptibility to noise interference is still significant. Common interferences include electrocardiogram, electrooculogram, and electromyogram artifacts, which are difficult to eliminate directly during signal acquisition, seriously affecting experimental and research results. Therefore, removing EEG signal noise has become a crucial issue.

[0003] The existing denoising methods include using variational mode decomposition (VMD), singular spectrum analysis (SSA) or independent component analysis (ICA) to denoise EEG signals. VMD can effectively suppress the modal aliasing and endpoint effect caused by empirical mode decomposition (EMD) when processing signals. In the signal processing process, SSA needs to reconstruct the signal by distinguishing the components of noise signals and effective EEG signals. When ICA performs blind source separation on mixed signals, the results of the ICA algorithm may vary due to different estimation methods or parameter settings, resulting in different separation results.

[0004] However, a significant drawback of VMD is its dependence on the number of decompositions. Specifically, VMD needs to determine the number of decomposed modes based on the actual conditions of the signal. If the number of decompositions is not set properly, VMD may not be able to effectively distinguish between signals and noise, thus affecting the decomposition effect and the accuracy of signal processing. Due to the lack of an automated mechanism for selecting the number of decompositions, users need to determine the appropriate number of decompositions through experience or experiments, which increases the complexity and uncertainty of the operation. In addition, VMD is sensitive to the initial parameter settings. Different initial parameters may lead to significant differences in the decomposition results, affecting the stability and reliability of the algorithm. In addition, VMD has a high computational complexity, especially when processing large-scale data, and the computational time and resource consumption are large, which limits its application in real-time signal processing. It is worth noting that VMD extracts all possible modes from the input signal, which will generate a large computational burden, and it is not necessary to extract all modes in applications such as artifact removal. In the denoising stage, only the modes with the center frequency near the artifact type that the research is concerned about removing need to be extracted for processing. For example, electrooculography (EOG) is mainly concentrated in the low frequency band, usually between 0.1Hz and 10Hz; the frequency of electrocardiography (ECG) artifacts is mainly between 0.5Hz and 75Hz. The frequency range of electromyography (EMG) artifacts is wider, usually between 20Hz and 200Hz, and the main components are concentrated in 50Hz to 150Hz.

[0005] However, SSA has a major drawback when processing signals: when the original EEG signal is slightly contaminated by noise, the signal subspace can be constructed by extracting the larger singular values ​​in the trajectory matrix, while the smaller singular values ​​constitute the noise subspace, thereby achieving signal-to-noise separation. However, when the original EEG signal is severely contaminated by noise, even a small amount of large noise will cause the method to fail and fail to effectively distinguish between signal and noise.

[0006] As for ICA, it needs to assume that the components are independent of each other, but in practical applications, this assumption is not always true. When processing EEG signals, the correlation between channels needs to be considered. In addition, ICA is very sensitive to noise in the data, and the presence of noise may affect the accuracy of the separation results. The performance of ICA depends on the high-order statistical characteristics of the signal. For non-Gaussian and nonlinear data, ICA may not be able to extract independent components well. For EEG signals with non-Gaussian, non-stationary, nonlinear and complex physiological signals, the performance of ICA in separating artifact noise may be affected. It is worth noting that if the signal segments containing EOG and the signal segments without EOG are input into the artifact removal method together, the signal segments without EOG will cause large distortion and changes. This is because the blind source separation algorithm may mix some signal-related source signals into the EOG source signal when separating the source signal, and the processing of the EOG source signal will have a greater impact on these source signals. Because EOG has a sharp jump feature, the amplitude of the corresponding source signal will also change greatly. In order to weaken the amplitude jump of the EOG segment, a large suppression change will be produced when processing these source signals. If signal-related information is mixed into the EOG source signal at this time, this information will suffer a large loss.

[0007] The prior art discloses a method for denoising and de-aliasing myoelectric signals based on second-generation wavelet and independent component analysis, which uses a second-generation wavelet denoising algorithm to first filter out the noise in the signal, and then performs ICA separation on the denoised mixed signal to complete the feature extraction of the myoelectric signal. The defect of this solution is that the signal segments containing EOG and the signal segments without EOG are input into the artifact removal method together, which will cause the signal segments without EOG to produce large distortion and changes.

[0008] To this end, in combination with the above requirements and the defects of the prior art, the present application proposes an EEG signal denoising method and system combining SSA and JADE. Summary of the invention

[0009] The present invention provides an EEG signal denoising method and system combining SSA and JADE, using VME to extract specific modes in EEG signals, especially modes related to electro-oculogram (EOG) interference, and process them without affecting other EEG signal fragments. Combined with SSA, the EEG signal is subjected to time-frequency analysis, and the signal is decomposed into multiple components to provide input for subsequent blind source separation. The JADE algorithm is used to perform blind source separation on the components decomposed by SSA to remove artifact interference such as electrocardiogram and electromyography.

[0010] The primary purpose of the present invention is to solve the above technical problems. The technical solution of the present invention is as follows:

[0011] The first aspect of the present invention provides an EEG signal denoising method combining SSA and JADE, the method comprising the following steps:

[0012] S1. Preprocess the original EEG signal, use variational mode to extract the signal mode of oscillation change related to the artifact, combine the preset threshold and the extreme point of the signal mode, filter and extract the fragment containing only the electrooculogram signal, and set the rest to zero as the input signal of blind source separation (BSS).

[0013] S2. Decompose the single-channel EEG signal containing the electrooculogram signal into multiple components, construct a trajectory matrix and perform singular value decomposition.

[0014] S3. According to the size of the singular value, the main components and noise components of the signal are preliminarily separated, and the components obtained by singular spectrum analysis decomposition are further processed using the characteristic matrix combined with the similarity diagonalization algorithm.

[0015] S4. Use high-order statistics to separate signals and noise, process source signals of artifact-related sources, and suppress noise components.

[0016] S5. Perform inverse feature matrix joint similarity diagonal transformation on the processed source signal matrix and the mixing matrix to resynthesize the denoised single-channel EEG signal.

[0017] Furthermore, step S1 includes the following specific steps: extracting the center frequency ω of the original EEG signal x(n) that meets the desired artifact correlation d The nearby mode u d (n), decompose the signal x(n) into the desired mode u d (n) and the residual x r (n).

[0018] Furthermore, the desired mode u d (n) is calculated by minimizing the bandwidth near its center frequency, specifically by first using the Hilbert transform to obtain the analytical signal of the desired mode:

[0019]

[0020] Where δ(n) is the Dirac function, and * represents convolution, which moves the frequency of the analytical signal to the center frequency, that is, multiplying the analytical signal by an exponential function:

[0021]

[0022] Calculate the gradient of the square of the second norm of the frequency-shifted analytical signal as the bandwidth of the mode; minimizing the bandwidth requires satisfying the following constraints:

[0023]

[0024] A filter is used to filter the part of the spectrum that overlaps with the desired mode, and the filter is at the desired center frequency ω d Nearby has infinite sensitivity, specifically:

[0025]

[0026] Where α is used to adjust the bandwidth of the filter.

[0027] Furthermore, the filter is constrained using a penalty equation, where β(n) is the impulse response of the filter:

[0028]

[0029] In order to make the original signal completely reconstructed by the expected mode and the residual signal, it is necessary to add the constraint that the sum of the expected mode and the residual signal is equal to the original signal:

[0030]

[0031] The alternating direction method of multipliers is used to solve the above equation.

[0032] Furthermore, the specific process of step S2 is: using singular spectrum analysis to decompose the mode extracted in step S1 into multiple components, and the singular spectrum analysis decomposes the signal into several components by constructing a trajectory matrix and performing singular value decomposition to obtain the main components and noise components of the signal.

[0033] Furthermore, the singular spectrum analysis includes the following steps: embedding, singular value decomposition, grouping and reconstruction; the embedding adopts K=N-L+1, X i =[x i ,x i+1 ,...,x i+L-1 ] Τ , (1≤i≤K) represents the lag vector after the i-th window slides, then the signal x is embedded and mapped into an L×K trajectory matrix:

[0034]

[0035] The purpose of the singular value decomposition is to find X = UΣV Τ The left and right unitary matrices U and V and the diagonal matrix Σ in the covariance matrix XX of X T Perform singular value decomposition, and the obtained singular values ​​and singular vectors are the squares of the singular values ​​and left singular vectors of X, that is:

[0036] XXT =UΛU Τ

[0037] where Λ is a diagonal matrix with diagonal elements λ1,λ2,...,λ L For XX T Arrange the eigenvalues ​​of in descending order, and let but:

[0038]

[0039]

[0040] The right unitary matrix is ​​composed of X,U i and λ i Indicates that the grouping is based on the preset standard S i Grouping is performed, I = 1, 2, ..., L is divided into M groups, and the trajectory matrix X can be expressed as:

[0041]

[0042] in is the mth submatrix, which is used to represent the different components of the time series. The original N-point signal x is expressed as:

[0043]

[0044] where x m represents a subsequence of length N obtained by performing anti-diagonal averaging on the submatrix.

[0045] Furthermore, the step S3 includes the following specific steps: performing a blind source separation algorithm on the EEG components and noise components initially separated by singular value decomposition through joint similarity diagonalization of the feature matrix, and the joint similarity diagonalization of the feature matrix includes the following steps: calculating the sample covariance And whitening matrix W, calculate the sample fourth-order cumulant Q z (M i ) and apply joint diagonalization, using Compute an estimate of the source signal.

[0046] Furthermore, the calculation of sample covariance The whitening matrix W is calculated using the following formula:

[0047] R x =E[x·x * ]

[0048]

[0049]

[0050] Where * represents complex conjugate, λ i is the characteristic value, h i For R x The corresponding eigenvector, the superscript H represents the complex conjugate transpose, s is the source signal, A is the mixing matrix, z represents x = As left multiplied by the whitening matrix W; the calculation sample fourth-order cumulant Q z (M i ) is defined as:

[0051]

[0052] Among them, M i , i=1,2,…N is a collection of different matrices used to make is a diagonal matrix, and after applying joint diagonalization we get:

[0053]

[0054] Among them, ||diag(·)|| 2 represents the sum of the squares of the diagonal, the matrix By Q z (M i ), which is obtained by maximizing J(·).

[0055] Furthermore, the step S5 is specifically as follows: processing the estimation of the source signal separated by the feature matrix joint similarity diagonalization to remove obvious noise components, and then performing the inverse feature matrix joint similarity diagonalization to complete the reconstructed signal.

[0056] The second aspect of the present invention provides an EEG signal denoising system combining SSA and JADE, which is used for the EEG signal denoising method combining SSA and JADE, and includes: a signal preprocessing module, a singular spectrum analysis module, a feature matrix joint similarity diagonalization module and a signal reconstruction module.

[0057] The signal preprocessing module preprocesses the original EEG signal, uses variational mode to extract the signal mode of oscillation change related to the artifact, combines the preset threshold and the extreme point of the signal mode, screens and extracts the fragment containing only the electrooculogram signal, and sets the rest to zero as the input signal for blind source separation. The singular spectrum analysis module decomposes the single-channel EEG signal containing the electrooculogram signal into multiple component components, constructs a trajectory matrix and performs singular value decomposition, and preliminarily separates the main components and noise components of the signal according to the size of the singular values. The characteristic matrix joint similarity diagonalization module uses the characteristic matrix joint similarity diagonalization algorithm to further process the components decomposed by the singular spectrum analysis, separates the signal and the noise by high-order statistics, processes the source signal of the artifact-related source, and suppresses the noise component. The signal reconstruction module performs an inverse characteristic matrix joint similarity diagonalization transformation on the processed source signal matrix and the mixing matrix, and re-synthesizes the denoised single-channel EEG signal.

[0058] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0059] The present invention provides an EEG signal denoising method and system combining SSA and JADE, using the JADE algorithm to further process the components decomposed by SSA, using high-order statistics (fourth-order cumulants) to separate signals and noise, processing source signals of artifact-related sources, and suppressing noise components. Finally, the signal is reconstructed, the processed source signal matrix and the mixing matrix are subjected to inverse JADE transformation, and the denoised single-channel EEG signal is resynthesized, which can effectively distinguish noise and useful signals under high noise levels, thereby significantly improving the effect of signal processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 The present invention is a flow chart of an EEG signal denoising method combining SSA and JADE.

[0061] Figure 2 This is a schematic diagram of an EEG signal denoising system combining SSA and JADE according to the present invention. DETAILED DESCRIPTION

[0062] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0063] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0064] Example 1

[0065] like Figure 1 As shown, the present invention provides an EEG signal denoising method combining SSA and JADE, and the method comprises the following steps:

[0066] S1. Preprocess the original EEG signal, use variational mode to extract the signal mode of oscillation change related to the artifact, combine the preset threshold and the extreme point of the signal mode, filter and extract the fragment containing only the electrooculogram signal, and set the rest to zero as the input signal of blind source separation (BSS).

[0067] S2. Decompose the single-channel EEG signal containing the electrooculogram signal into multiple components, construct a trajectory matrix and perform singular value decomposition.

[0068] S3. According to the size of the singular value, the main components and noise components of the signal are preliminarily separated, and the components obtained by singular spectrum analysis decomposition are further processed using the characteristic matrix combined with the similarity diagonalization algorithm.

[0069] S4. Use high-order statistics to separate signals and noise, process source signals of artifact-related sources, and suppress noise components.

[0070] S5. Perform inverse feature matrix joint similarity diagonal transformation on the processed source signal matrix and the mixing matrix to resynthesize the denoised single-channel EEG signal.

[0071] It should be noted that the present invention uses variational mode extraction combined with singular spectrum analysis and characteristic matrix joint similarity diagonalization algorithm to denoise the EEG signal. Since the signal segments containing EOG and the signal segments without EOG are input into the method of blind source separation artifact removal together, the signal segments without EOG will cause large distortion and changes. In response to this problem, VME can only extract the required mode of signal oscillation, and according to the time domain and frequency domain waveform characteristics of the electrooculogram artifact, the EOG segment in the signal can be screened out in combination with the amplitude threshold and the signal frequency, and only the EEG segment containing EOG artifact interference is processed, avoiding the situation that the EEG segment that does not contain EOG artifact interference after processing is severely distorted after denoising. Secondly, the algorithm of singular spectrum analysis and characteristic matrix joint similarity diagonalization is used to remove the artifact interference of electrocardiogram, electromyography, etc. of the EEG signal. As a blind source separation (BSS) algorithm, characteristic matrix joint similarity diagonalization (JADE) implicitly assumes that the number of channels is more than the number of source signals. If the number of channels is less than the number of source signals, the BSS method will not be able to obtain a unique separation matrix, resulting in unreliable separation results. As a complex mixed signal, a single-channel EEG signal contains not only EEG-related source signals that need to be retained, but also multiple artifact-related source signals.

[0072] In order to solve this problem, a single-channel decomposition method, singular spectrum analysis, is used to decompose the single-channel EEG signal into multiple components. These components can be used as inputs to the blind source separation algorithm, thereby improving the separation effect and the accuracy of signal processing.

[0073] Specifically, SSA decomposes a single-channel signal into several components by constructing a trajectory matrix and performing singular value decomposition. Larger singular values ​​correspond to the signal subspace, and smaller singular values ​​correspond to the noise subspace. Then, the JADE algorithm is used to further process these components, and higher-order statistics are used to achieve more accurate signal separation. JADE can more accurately separate independent components by jointly diagonalizing multiple matrices. Especially for non-Gaussian signals, JADE's higher-order statistics can better capture signal characteristics, thereby effectively suppressing noise. By combining the SSA and JADE algorithms, the advantages of both can be fully utilized: the initial decomposition ability of SSA and the higher-order statistics separation ability of JADE, so that noise and signal can still be effectively distinguished under high noise levels, and the effect of signal processing can be improved. The problem of SSA's high sensitivity to noise is overcome. When the original signal is severely contaminated by noise, SSA may not be able to effectively separate the signal and noise, resulting in analysis failure. In this way, EEG signals and artifact signals can be effectively distinguished, and the effect of signal processing can be significantly improved even under high noise levels.

[0074] According to the above technical features, the present invention processes EEG signals through variational mode extraction, and can perform denoising on specific areas interfered by artifacts without affecting other normal EEG signal fragments. This method effectively avoids the distortion and twisting of EEG signals that may be caused during the denoising process, thereby retaining the integrity and accuracy of the signal. By combining singular spectrum analysis with the joint similarity diagonalization algorithm of the characteristic matrix, the advantages of both are fully utilized. SSA has the ability to preliminarily decompose the signal, while JADE uses high-order statistics for accurate blind source separation. This combined method can still effectively distinguish between noise and useful signals under high noise levels, thereby significantly improving the effect of signal processing. Using the characteristic matrix to jointly similarly diagonalize the blind source separation problem of EEG signals, JADE uses the fourth-order cumulant of the signal to achieve blind source separation. The high-order cumulant can capture the non-Gaussian characteristics of the signal. In addition, JADE has good robustness to changes in the noise and signal mixing matrix. This means that in practical applications, JADE can stably separate the source signal. This makes JADE perform well when processing non-Gaussian signals such as EEG signals.

[0075] Compared with the prior art, the present invention uses variational mode extraction combined with singular spectrum analysis and characteristic matrix joint similarity diagonalization to process single-channel EEG signals, which can not only efficiently and accurately remove related artifact interference, but also avoid EEG signal distortion and distortion that may be caused in the denoising process, and retain the original physiological characteristics and information of the EEG signal as much as possible. The singular spectrum analysis is combined with the characteristic matrix joint similarity diagonalization algorithm to denoise the single-channel EEG signal, which meets the requirement that the number of channels is implicitly assumed to be more than the number of source signals in blind source separation, and simultaneously utilizes the advantages of the two algorithms. SSA performs a preliminary decomposition of the EEG signal, while JADE uses high-order statistics for accurate blind source separation, which can more effectively distinguish artifacts from EEG signals. JADE uses the fourth-order cumulant of the signal to achieve blind source separation. The high-order cumulant can capture the non-Gaussian characteristics of the signal. In addition, JADE has better robustness to changes in the noise and signal mixing matrix, avoiding the influence of noise.

[0076] Example 2

[0077] Based on the above-mentioned embodiment 1, this embodiment elaborates the specific content of the present invention.

[0078] Furthermore, step S1 includes the following specific steps: extracting the center frequency ω of the original EEG signal x(n) that meets the desired artifact correlation d The nearby mode u d (n), decompose the signal x(n) into the desired mode u d (n) and the residual x r (n).

[0079] Furthermore, the desired mode u d (n) is calculated by minimizing the bandwidth near its center frequency, specifically by first using the Hilbert transform to obtain the analytical signal of the desired mode:

[0080]

[0081] Where δ(n) is the Dirac function, and * represents convolution, which moves the frequency of the analytical signal to the center frequency, that is, multiplying the analytical signal by an exponential function:

[0082]

[0083] Calculate the gradient of the square of the second norm of the frequency-shifted analytical signal as the bandwidth of the mode; minimizing the bandwidth requires satisfying the following constraints:

[0084]

[0085] A filter is used to filter the part of the spectrum that overlaps with the desired mode, and the filter is at the desired center frequency ωd Nearby has infinite sensitivity, specifically:

[0086]

[0087] Where α is used to adjust the bandwidth of the filter.

[0088] Furthermore, the filter is constrained using a penalty equation, where β(n) is the impulse response of the filter:

[0089]

[0090] In order to make the original signal completely reconstructed by the expected mode and the residual signal, it is necessary to add the constraint that the sum of the expected mode and the residual signal is equal to the original signal:

[0091]

[0092]

[0093] The alternating direction method of multipliers is used to solve the above equation.

[0094] Furthermore, the specific process of step S2 is: using singular spectrum analysis to decompose the mode extracted in step S1 into multiple components, and the singular spectrum analysis decomposes the signal into several components by constructing a trajectory matrix and performing singular value decomposition to obtain the main components and noise components of the signal.

[0095] Furthermore, the singular spectrum analysis includes the following steps: embedding, singular value decomposition, grouping and reconstruction; the embedding adopts K=N-L+1, X i =[x i ,x i+1 ,...,x i+L-1 ] Τ , (1≤i≤K) represents the lag vector after the i-th window slides, then the signal x is embedded and mapped into an L×K trajectory matrix:

[0096]

[0097] The purpose of the singular value decomposition is to find X = UΣV Τ The left and right unitary matrices U and V and the diagonal matrix Σ in the covariance matrix XX of X T Perform singular value decomposition, and the obtained singular values ​​and singular vectors are the squares of the singular values ​​and left singular vectors of X, that is:

[0098] XX T =UΛU Τ

[0099] where Λ is a diagonal matrix with diagonal elements λ1,λ2,...,λ L For XX T Arrange the eigenvalues ​​of in descending order, and let but:

[0100]

[0101]

[0102] The right unitary matrix is ​​composed of X,U i and λ i Indicates that the grouping is based on the preset standard S i Grouping is performed, I = 1, 2, ..., L is divided into M groups, and the trajectory matrix X can be expressed as:

[0103]

[0104] in is the mth submatrix, which is used to represent the different components of the time series. The original N-point signal x is expressed as:

[0105]

[0106] where x m represents a subsequence of length N obtained by performing anti-diagonal averaging on the submatrix.

[0107] Furthermore, the step S3 includes the following specific steps: performing a blind source separation algorithm on the EEG components and noise components initially separated by singular value decomposition through joint similarity diagonalization of the feature matrix, and the joint similarity diagonalization of the feature matrix includes the following steps: calculating the sample covariance And whitening matrix W, calculate the sample fourth-order cumulant Q z (M i ) and apply joint diagonalization, using Compute an estimate of the source signal.

[0108] Furthermore, the calculation of sample covariance The whitening matrix W is calculated using the following formula:

[0109] R x =E[x·x * ]

[0110]

[0111]

[0112] Where * represents complex conjugate, λ i is the characteristic value, h i For Rx The corresponding eigenvector, the superscript H represents the complex conjugate transpose, s is the source signal, A is the mixing matrix, z represents x = As left multiplied by the whitening matrix W; the calculation sample fourth-order cumulant Q z (M i ) is defined as:

[0113]

[0114] Among them, M i , i=1,2,…N is a collection of different matrices used to make is a diagonal matrix, and after applying joint diagonalization we get:

[0115]

[0116] Among them, ||diag(·)|| 2 represents the sum of the squares of the diagonal, the matrix By Q z (M i ), which is obtained by maximizing J(·).

[0117] Furthermore, the step S5 is specifically as follows: processing the estimation of the source signal separated by the feature matrix joint similarity diagonalization to remove obvious noise components, and then performing the inverse feature matrix joint similarity diagonalization to complete the reconstructed signal.

[0118] In a specific embodiment, the parameters of singular spectrum analysis are selected according to the characteristics of the processed signal. EEG signal-related sources may include artifact sources such as EMG, ECG, and EOG, and the blind source separation algorithm implicitly assumes that the number of input channels is greater than the number of source signals. Therefore, the SSA window length is set to 6, and 6 components are separated and input into the JADE algorithm for further blind source separation. In variational mode extraction, due to the time-frequency characteristics of EOG artifacts, the amplitude of the EOG segment will quickly jump to the maximum amplitude, and then slowly oscillate back to the normal range. Generally speaking, the post-blink peak lasts longer than the pre-blink peak, and the duration of the post-peak is about 3-4 times that of the pre-peak.

[0119] In this embodiment, the first 15.625ms and the last 62.5ms of the maximum peak are taken as EOG segments. Based on this feature, VME can filter out EOG segments and only process the segments containing EOG, avoiding the situation where non-EOG segments are changed after processing. First, use VME to extract the mode of signal change and find all the extreme points of the mode. Then, combined with the threshold, 1.5 times the standard deviation of the VME mode is used as the threshold, and the formula for locating the blink peak is referred to:

[0120]

[0121] Finally, the frequency of the selected EOG segment is calculated before BSS is performed on the segment. Because EOG has the characteristics of slow changes and low frequency, generally 0.3Hz-15Hz, only the segments below 12Hz are processed here. In this way, only the EOG oscillation part can be processed, which can avoid the information part of the signal being mixed in the artifact source and removed. By selecting the above appropriate signal processing algorithm parameters according to the artifact characteristics contained in the processed signal, the artifact can be removed more effectively, thereby ensuring that the denoised signal still retains the original physiological characteristics and information.

[0122] Example 3

[0123] like Figure 2 As shown, the present invention also provides an EEG signal denoising system combining SSA and JADE, which is used for the EEG signal denoising method combining SSA and JADE, and includes: a signal preprocessing module, a singular spectrum analysis module, a feature matrix joint similarity diagonalization module and a signal reconstruction module.

[0124] The signal preprocessing module preprocesses the original EEG signal, uses variational mode to extract the signal mode of oscillation change related to the artifact, combines the preset threshold and the extreme point of the signal mode, screens and extracts the fragment containing only the electrooculogram signal, and sets the rest to zero as the input signal for blind source separation. The singular spectrum analysis module decomposes the single-channel EEG signal containing the electrooculogram signal into multiple component components, constructs a trajectory matrix and performs singular value decomposition, and preliminarily separates the main components and noise components of the signal according to the size of the singular values. The characteristic matrix joint similarity diagonalization module uses the characteristic matrix joint similarity diagonalization algorithm to further process the components decomposed by the singular spectrum analysis, separates the signal and the noise by high-order statistics, processes the source signal of the artifact-related source, and suppresses the noise component. The signal reconstruction module performs an inverse characteristic matrix joint similarity diagonalization transformation on the processed source signal matrix and the mixing matrix, and re-synthesizes the denoised single-channel EEG signal.

[0125] In the embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium, which, when executed, executes the steps of the above method embodiments; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, and other media capable of storing program codes.

[0126] Alternatively, if the above-mentioned embodiments of the present invention are implemented in the form of software function modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media capable of storing program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0127] 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. The icons describing the structural positional relationships in the accompanying drawings are only used for exemplary purposes and are not to be construed as limiting the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and is not possible to list all the embodiments here. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the claims of the present invention.

Claims

1. A method for denoising an EEG signal by combining SSA and JADE, characterized in that: The following steps are involved: S1. Preprocess the original EEG signal, use variational mode to extract the signal mode of oscillation change related to the artifact, combine the preset threshold and the extreme point of the signal mode, screen and extract the fragment containing only the electrooculogram signal, and set the rest to zero as the input signal of blind source separation; S2, decomposing the single-channel EEG signal including the electrooculogram signal into multiple components, constructing a trajectory matrix and performing singular value decomposition; S3. According to the size of the singular value, the main component and the noise component of the signal are preliminarily separated, and the components obtained by singular spectrum analysis decomposition are further processed using the characteristic matrix combined with the similarity diagonalization algorithm; S4, using high-order statistics to separate signals and noise, process source signals of artifact-related sources, and suppress noise components; S5. Perform inverse feature matrix joint similarity diagonal transformation on the processed source signal matrix and the mixing matrix to resynthesize the denoised single-channel EEG signal.

2. The method for denoising an EEG signal by combining SSA and JADE according to claim 1, characterized in that: The step S1 includes the following specific steps: extracting the center frequency ω of the original EEG signal x(n) that meets the desired artifact correlation d The nearby mode u d (n), decompose the signal x(n) into the desired mode u d (n) and the residual x r (n).

3. The method for denoising an EEG signal by combining SSA and JADE according to claim 2, characterized in that: The desired mode u d (n) is calculated by minimizing the bandwidth near its center frequency, specifically by first using the Hilbert transform to obtain the analytical signal of the desired mode: Where δ(n) is the Dirac function, and * represents convolution, which moves the frequency of the analytical signal to the center frequency, that is, multiplying the analytical signal by an exponential function: Calculate the gradient of the square of the second norm of the frequency-shifted analytical signal as the bandwidth of the mode; minimizing the bandwidth requires satisfying the following constraints: A filter is used to filter the part of the spectrum that overlaps with the desired mode, and the filter is at the desired center frequency ω d Nearby has infinite sensitivity, specifically: Where α is used to adjust the bandwidth of the filter.

4. The method for denoising an EEG signal by combining SSA and JADE according to claim 3, characterized in that: The filter is constrained using a penalty equation, where β(n) is the impulse response of the filter: In order to make the original signal completely reconstructed by the expected mode and the residual signal, it is necessary to add the constraint that the sum of the expected mode and the residual signal is equal to the original signal: s.t.:u d (n)+x r (n)=x(n) The alternating direction method of multipliers is used to solve the above equation.

5. The method for denoising an EEG signal by combining SSA and JADE according to claim 1, characterized in that: The specific process of step S2 is: using singular spectrum analysis to decompose the mode extracted in step S1 into multiple components. The singular spectrum analysis decomposes the signal into several components by constructing a trajectory matrix and performing singular value decomposition to obtain the main components and noise components of the signal.

6. The method for denoising an EEG signal by combining SSA and JADE according to claim 5, characterized in that: The singular spectrum analysis includes the following steps: embedding, singular value decomposition, grouping and reconstruction; the embedding adopts K=N-L+1, X i =[x i ,x i+1 ,...,x i+L-1 ] Τ , (1≤i≤K) represents the lag vector after the i-th window slides, then the signal x is embedded and mapped into an L×K trajectory matrix: The purpose of the singular value decomposition is to find X = UΣV Τ The left and right unitary matrices U and V and the diagonal matrix Σ in the covariance matrix of X Perform singular value decomposition, and the obtained singular values ​​and singular vectors are the squares of the singular values ​​and left singular vectors of X, that is: where Λ is a diagonal matrix with diagonal elements λ1,λ2,...,λ L for Arrange the eigenvalues ​​of in descending order, and let but: The right unitary matrix is ​​composed of X,U i and λ i Indicates that the grouping is based on the preset standard S i Grouping is performed, I = 1, 2, ..., L is divided into M groups, and the trajectory matrix X can be expressed as: in is the mth submatrix, which is used to represent the different components of the time series. The original N-point signal x is expressed as: where x m represents a subsequence of length N obtained by performing anti-diagonal averaging on the submatrix.

7. The method for denoising an EEG signal by combining SSA and JADE according to claim 6, characterized in that: The step S3 comprises the following specific steps: The blind source separation algorithm is performed on the EEG components and noise components initially separated by singular value decomposition through the joint similarity diagonalization of the feature matrix. The joint similarity diagonalization of the feature matrix includes the following steps: calculating the sample covariance And whitening matrix W, calculate the sample fourth-order cumulant Q z (M i ) and apply joint diagonalization, using Compute an estimate of the source signal.

8. The method for denoising an EEG signal by combining SSA and JADE according to claim 7, characterized in that: The calculation of sample covariance The whitening matrix W is calculated using the following formula: R x =E[x·x * ] Where * represents complex conjugate, λ i is the characteristic value, h i For R x The corresponding eigenvector, the superscript H represents the complex conjugate transpose, s is the source signal, A is the mixing matrix, z represents x = As left multiplied by the whitening matrix W; the calculation sample fourth-order cumulant Q z (M i ) is defined as: Among them, M i , i=1,2,…N is a collection of different matrices used to make is a diagonal matrix, and after applying joint diagonalization we get: Among them, ||diag(·)|| 2 represents the sum of the squares of the diagonal, the matrix It is obtained by the joint approximation of Qz(Mi) to diagonalize, that is, by maximizing J(·).

9. The method for denoising an EEG signal by combining SSA and JADE according to claim 1, characterized in that: The step S5 specifically includes: processing the estimation of the source signal separated by the feature matrix joint similarity diagonalization to remove the obvious noise component, and then performing the inverse feature matrix joint similarity diagonalization to complete the reconstruction of the signal.

10. An EEG signal denoising system combining SSA and JADE, the system being used for the EEG signal denoising method combining SSA and JADE as claimed in any one of claims 1 to 9, characterized in that: It includes: signal preprocessing module, singular spectrum analysis module, characteristic matrix joint similarity diagonalization module and signal reconstruction module; The signal preprocessing module preprocesses the original EEG signal, uses variational mode to extract the signal mode of oscillation change related to the artifact, combines the preset threshold and the extreme point of the signal mode, screens and extracts the fragment containing only the electrooculogram signal, and sets the rest to zero as the input signal for blind source separation. The singular spectrum analysis module decomposes the single-channel EEG signal containing the electrooculogram signal into multiple component components, constructs a trajectory matrix and performs singular value decomposition, and preliminarily separates the main components and noise components of the signal according to the size of the singular values. The characteristic matrix joint similarity diagonalization module uses the characteristic matrix joint similarity diagonalization algorithm to further process the components decomposed by the singular spectrum analysis, separates the signal and the noise by high-order statistics, processes the source signal of the artifact-related source, and suppresses the noise component. The signal reconstruction module performs an inverse characteristic matrix joint similarity diagonalization transformation on the processed source signal matrix and the mixing matrix, and re-synthesizes the denoised single-channel EEG signal.

Images