CN114757236B - Electroencephalogram signal denoising optimization method and system based on TQWT and SVMD - Google Patents
Electroencephalogram signal denoising optimization method and system based on TQWT and SVMD Download PDFInfo
- Publication number
- CN114757236B CN114757236B CN202210659050.5A CN202210659050A CN114757236B CN 114757236 B CN114757236 B CN 114757236B CN 202210659050 A CN202210659050 A CN 202210659050A CN 114757236 B CN114757236 B CN 114757236B
- Authority
- CN
- China
- Prior art keywords
- components
- signal
- modal
- tqwt
- subband
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/369—Electroencephalography [EEG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7253—Details of waveform analysis characterised by using transforms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention provides an electroencephalogram signal denoising optimization method and system based on TQWT and SVMD, and relates to the technical field of electroencephalogram signal denoising optimization.
Description
Technical Field
The invention relates to the technical field of electroencephalogram signal denoising optimization, in particular to an electroencephalogram signal denoising optimization method and system based on TQWT and SVMD.
Background
With the progress of science and the rapid development of biomedical technology, biological signals are applied to research in various fields, such as monitoring blood pressure and blood sugar by using pulse electrical signals, sleep classification by using electro-oculogram and electrocardiogram, and besides, the application of electroencephalogram signals is more extensive, such as sleep classification by using electroencephalogram signals, epilepsy and auxiliary diagnosis of depression.
However, for electroencephalogram signals, because the amplitude of the acquired electroencephalogram signals is small, generally, the acquired electroencephalogram signals are only dozens to hundreds of microvolts, various noise interferences can be caused in the acquisition process, most common noises such as ocular artifacts and myoelectric artifacts are difficult to be directly eliminated in the signal acquisition process, and the noises can influence the experimental and research effects, so the denoising of the electroencephalogram signals is very important.
Empirical Mode Decomposition (EMD) is a time-frequency domain signal processing method that performs signal Decomposition according to the time scale characteristics of data itself without presetting any basis function, is suitable for processing nonlinear and non-stationary signal sequences, has a high signal-to-noise ratio, performs EMD processing on an original electroencephalogram signal, and then performs signal reconstruction, thereby achieving the effect of denoising. Discrete Wavelet Transform (DWT) can also be used for denoising of electroencephalogram signals, and complete reconstruction of electroencephalogram signals can be achieved, but in the face of signals with oscillation behaviors, Q factors (ratio of oscillation signal center frequency to bandwidth) of oscillation signals cannot be flexibly adjusted, free division of frequency bands and bandwidths of decomposed signal components cannot be achieved, and flexibility is poor. For example, the prior art discloses a method for removing ocular artifacts in a single-channel electroencephalogram signal, which comprises the steps of firstly carrying out empirical wavelet transform processing on the single-channel electroencephalogram signal, and carrying out improved adaptive noise-complete empirical mode decomposition on the electroencephalogram signal of a delta frequency band to obtain a plurality of inherent modal functions; calculating the sample entropy of each inherent mode function, and setting a sample entropy threshold value to identify the inherent mode function containing the ocular artifacts; removing intrinsic mode functions containing ocular artifacts, and performing improved adaptive noise complete empirical mode decomposition inverse operation reconstruction on the remaining intrinsic mode functions to obtain delta frequency band signals after filtering; and finally, performing empirical wavelet transform-based inverse transformation on the filtered delta frequency band signal and the filtered high frequency band signal, and finally reconstructing to obtain the electroencephalogram signal without the ocular artifacts. However, the EMD generates a mode aliasing phenomenon in the process of decomposing the electroencephalogram signal, the mode aliasing is caused by signal interruption, and the interruption confuses time-frequency distribution, thereby destroying the physical significance of the intrinsic mode function, and making the feature extraction, model training and mode identification difficult. The Variable Mode Decomposition (VMD) can well suppress the modal aliasing phenomenon generated when the EMD processes a signal, but the VMD has high computational complexity and the initial value of the center frequency has a large influence on the result of VMD processing.
Disclosure of Invention
In order to solve the problems of poor flexibility, high calculation complexity and poor operation speed of the current EEG signal denoising method, the invention provides an EEG signal denoising optimization method and system based on TQWT and SVMD.
In order to achieve the technical effects, the technical scheme of the invention is as follows:
a TQWT and SVMD based electroencephalogram signal denoising optimization method comprises the following steps:
s1, collecting original electroencephalogram signals, and preprocessing the original electroencephalogram signals to obtain preprocessed discrete electroencephalogram signals;
s2, performing wavelet transformation TQWT decomposition operation with adjustable Q factors on the preprocessed discrete electroencephalogram signals to obtain sub-band components;
s3, analyzing the energy of each sub-band component, and selecting a plurality of sub-band components from each sub-band to reconstruct a signal to obtain a reconstructed electroencephalogram signal;
and S4, carrying out continuous variational modal decomposition (SVMD) optimization and reprocessing on the reconstructed electroencephalogram signal, and decomposing out a plurality of modal components.
In the technical scheme, the original electroencephalogram signal is preprocessed, wavelet transformation TQWT decomposition operation with adjustable Q factors is carried out on the preprocessed discrete electroencephalogram signal, the Q factors can be adjusted according to the oscillation degree of the electroencephalogram signal, so that the frequency band and the bandwidth of decomposed sub-band components are freely divided, the flexibility is high, then the sub-band components used for reconstruction are selected based on the energy of each sub-band component, the calculation complexity is reduced, the reconstructed electroencephalogram signal is subjected to continuous variational modal decomposition (SVMD) optimization reprocessing, the modal aliasing phenomenon caused by the application of the traditional variational modal decomposition is inhibited, the center frequency is initialized randomly, the SVMD can be completely converged to the optimal center frequency mode, the robustness of the center frequency is stronger, the electroencephalogram signal is optimized, the iteration times are few, and the operation speed is accelerated.
Preferably, in step S1, the preprocessing operation performed on the raw brain electrical signal is:
setting a signal extraction frequency segment, performing discrete Fourier transform processing on an original electroencephalogram signal, extracting the electroencephalogram signal which is subjected to the discrete Fourier transform processing and is positioned in the signal extraction frequency segment by using a low-pass filter, and performing inverse discrete Fourier transform on the extracted electroencephalogram signal to obtain a preprocessed discrete electroencephalogram signalx[n],nRepresenting a time index of the discrete brain electrical signal.
Preferably, in step S2, it is assumed that the preprocessed discrete electroencephalogram signal is paired with a filter bank with a tree structurex[n]Performing a wavelet transform TQWT decomposition operation with an adjustable Q factor, andx[n]the decomposition into J +1 subband components, J representing the total number of layers of TQWT decomposition, each level of decomposition in the tree-structured filter bank consisting of a two-channel filter bank,αandβlow pass filters in dual channel filter bankCoefficient and high-pass filterThe coefficient of (a) is determined,a frequency-domain variable is represented by,,and is and;andby a power-complementary function of period 2 piDefinition of, forAnd satisfies the following conditions:
preferably, is provided withAndare respectively asjThe equivalent low-pass filter and the equivalent high-pass filter for which the layer is being decomposed are:
and
wherein the content of the first and second substances,is shown asjThe low-pass scaling factor of the layer low-pass filter,jindicating the layer that is currently being decomposed and,is shown asjLow-pass scaling factor of the layer high-pass filter, in the second placejDiscrete electroencephalogram signal after pretreatment in layer decompositionx[n]Via an equivalent low-pass filterObtaining a signal c after low-pass scaling j [n]Pre-processed discrete electroencephalogram signalsx[n]Via an equivalent high-pass filterLow pass scaling, high pass filterIs high-pass scaled to obtain a signal d j [n](ii) a During the TQWT decomposition operation, three parameters are set, which are: quality factor Q, total number of layers decomposed J, and redundancy r, where:
wherein the content of the first and second substances,f c andBWthe center frequency and the bandwidth of the low-pass filter and the high-pass filter respectively, Q represents an adjustable factor and is the ratio of the center frequency to the bandwidth, the center frequency and the bandwidth can be divided by adjusting Q, the oscillation degree of a signal is controlled,rcontrols the excessive ringing phenomenon, plays the role of localizing the characteristics of the electroencephalogram signals,rsatisfies the following conditions:
the center frequency and the bandwidth of the decomposed sub-band component can be freely divided by adjusting the adjustable factor Q, and the flexibility is stronger.
Preferably, the number of subband components decomposed by the wavelet transform TQWT with adjustable Q factor is J +1, and in step S3, the number one decomposed by the wavelet transform TQWT with adjustable Q factor is setiThe sub-band components are denoted as s i [n],i=1,2, …, J +1, its energy is expressed as:
wherein, the first and the second end of the pipe are connected with each other,E i is shown asiThe energy of the subband component; total energy of J +1 subband componentsEComprises the following steps:
then it is firstiOf a sub-bandEnergy ratioQ i Comprises the following steps:
total energy in J +1 subband componentsEBased on 90%, sorting the energy of each subband component from large to small, and selecting the energy of the subband from J +1 subband components according to the sorting orderEThe P subband components of 90% of the signal are reconstructed, P < J +1.
Preferably, when the P subband components are subjected to signal reconstruction, the time sequence of the remaining (J + 1-P) subband components in the J +1 subband components is set to zero, that is, the (J + 1-P) subband components are set to be 0, and finally, the P subband components and the remaining (J + 1-P) subband components are reconstructed by using a synthesis filter bank of the TQWT to obtain a reconstructed electroencephalogram signal, so that the computational complexity is reduced.
Preferably, the reconstructed electroencephalogram signal is represented asf(t) To, forf(t) Continuously applying variation modal decomposition, sequentially searching modal components,f(t) Is decomposed into two signals, the expression satisfies:wherein, in the step (A),the L-th modal component is represented,represents a redundant signal satisfying:
comprising two parts, one being set to obtain the first L-1 modalitiesSum of componentsOne is the unprocessed part of the signalWhen searching the first modal component of SVMD, the sum of the first L-1 modal components is obtainedIs zero; the specific process of carrying out continuous variational modal decomposition (SVMD) comprises the following steps:
s41, setting up、、、A value of (1), whereinIn order to be a penalty factor,、、is a threshold parameter for judging whether the loop is finished;
s42, before the first iteration of continuous variation modal decomposition, the frequency domain response of the L-th modal component is obtained by initialization settingCenter frequency of the L-th modal componentAnd lagrange multiplier;
S43. Application、、Is updated by the iterative formula、、Wherein, in the step (A),the iterative formula of (a) is:
wherein the content of the first and second substances,is composed off(t) As a result of the fourier transform, the result,is a frequency domain variable;the iterative formula of (a) is:
wherein the content of the first and second substances,it is indicated that the parameters of the update,representing an intermediate parameter;
s44, according to the discriminant:
and (c) carrying out a judgment, wherein,Tis a signalf(t) If the discriminant is not satisfied, the process returns to step S42 to extract the next modality, and if the discriminant is satisfied, it indicates that all the modalities have been extracted, and SVMD decomposition in the continuous variation modality is completed.
The reconstructed signal is processed by using the SVMD, modal mixing existing in the traditional EMD can be inhibited, the operation speed can be greatly improved compared with that of a VMD, and the method has stronger robustness on a central frequency initial value.
Preferably, the method further comprises a process of classifying the plurality of modal components according to the spectral characteristics, and the process is used for subsequent feature extraction of the electroencephalogram signal.
Preferably, the modal components obtained after SVMD processing are arranged in the order of frequency from low to high, a plurality of modal components are classified according to the spectral characteristics and then reconstructed, and when the modal components are classified according to the spectral characteristics, if the modal quantity is consistent with the number of the spectral characteristics, each modal component is classified corresponding to each spectral characteristics category; if the number of the modes is larger than the number of the spectrum characteristic categories, the mode components are firstly combined with the mode components adjacent to each other in the sequence from back to front, and are correspondingly classified into the spectrum characteristic categories, so that the consistency of the length of the feature vector of the electroencephalogram signal is ensured, and the feature extraction of the subsequent electroencephalogram signal is facilitated.
The application also provides an electroencephalogram signal denoising optimization system based on TQWT and SVMD, which comprises:
the electroencephalogram signal preprocessing module is used for preprocessing the acquired original electroencephalogram signal to obtain a preprocessed discrete electroencephalogram signal;
the TQWT decomposition module is used for performing wavelet transformation TQWT decomposition operation with adjustable Q factors on the preprocessed discrete electroencephalogram signals to obtain each sub-band component;
the signal analysis and reconstruction module is used for analyzing the energy of each sub-band component and selecting a plurality of sub-band components from each sub-band to carry out signal reconstruction so as to obtain a reconstructed electroencephalogram signal;
and the continuous variational modal decomposition module is used for carrying out SVMD optimization reprocessing on the reconstructed electroencephalogram signal to decompose a plurality of modal components.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
the invention provides an electroencephalogram signal denoising optimization method and system based on TQWT and SVMD, which comprises the steps of preprocessing an original electroencephalogram signal, performing wavelet transformation TQWT decomposition operation with adjustable Q factors on a preprocessed discrete electroencephalogram signal, adjusting the Q factors according to the oscillation degree of the electroencephalogram signal, freely dividing the frequency band and the bandwidth of a decomposed sub-band component, having high flexibility, selecting the sub-band component for reconstruction based on the energy of each sub-band component, reducing the calculation complexity, performing continuous variational modal decomposition SVMD optimization reprocessing on the reconstructed electroencephalogram signal, so as to inhibit the aliasing phenomenon caused by applying the traditional variational modal decomposition, having stronger robustness on the central frequency, optimizing the electroencephalogram signal, having less iteration times and accelerating the calculation speed.
Drawings
Fig. 1 shows a schematic flow diagram of an electroencephalogram signal denoising and optimizing method of TQWT and SVMD provided in embodiment 1 of the present invention;
FIG. 2 is a diagram of the decomposition of EEG signals based on the wavelet transformation TQWT with adjustable Q factor proposed in embodiment 1 of the present inventionx[n]Is decomposed into c j [n]A process diagram of (a);
FIG. 3 shows the decomposition of EEG signals based on a wavelet transformation TQWT decomposition operation with adjustable Q factor proposed in embodiment 1 of the present inventionx[n]Decomposition into electroencephalographic signals based on a wavelet transform TQWT decomposition operation with adjustable Q-factorx[n]Decomposed into signals c j [n]A process diagram of (1);
fig. 4 is a schematic diagram illustrating classification according to the consistency between the number of modes and the number of categories of spectral characteristics when the modes are classified according to the spectral characteristics according to embodiment 2 of the present invention;
fig. 5 shows a structural diagram of an electroencephalogram signal denoising optimization system based on TQWT and SVMD in embodiment 3 of the present invention.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
for better illustration of the present embodiment, some parts of the drawings may be omitted, enlarged or reduced, and do not represent actual sizes;
it will be understood by those skilled in the art that certain well-known descriptions of the figures may be omitted.
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
The positional relationships depicted in the drawings are for illustrative purposes only and are not to be construed as limiting the present patent;
example 1
The traditional empirical mode decomposition can generate a modal aliasing phenomenon after decomposing an electroencephalogram signal, the variational mode decomposition VMD can well inhibit the modal aliasing phenomenon generated when the EMD processes the electroencephalogram signal, but the calculation complexity is high, and the initial value of the central frequency has a large influence on the processing result, in order to solve the problems of poor flexibility, high calculation complexity and poor operation speed of the traditional mode, the embodiment combines wavelet transform (TQWT) with an adjustable Q factor and the continuous variational mode decomposition SVMD, as shown in FIG. 1, the embodiment provides an electroencephalogram signal denoising optimization method based on TQWT and SVMD, and the denoising optimization method comprises the following steps:
s1, collecting an original electroencephalogram signal, and preprocessing the original electroencephalogram signal to obtain a preprocessed discrete electroencephalogram signal;
the preprocessing operation of the original brain electrical signal is as follows: setting a signal extraction frequency segment, performing discrete Fourier transform processing on an original electroencephalogram signal, extracting the electroencephalogram signal which is subjected to the discrete Fourier transform processing and is positioned in the signal extraction frequency segment (0 Hz to 50Hz) by using a low-pass filter, and performing inverse discrete Fourier transform on the extracted electroencephalogram signal to obtain a preprocessed discrete electroencephalogram signalx[n],nRepresenting a time index of the discrete brain electrical signal.
S2, carrying out wavelet transformation TQWT decomposition operation with adjustable Q factors on the preprocessed discrete electroencephalogram signals to obtain sub-band components;
in this embodiment, it is assumed that the preprocessed discrete electroencephalogram signal is paired with a filter group with a tree structurex[n]Performing a wavelet transform TQWT decomposition operation with an adjustable Q factor, andx[n]the decomposition into J +1 subband components, J representing the total number of layers of TQWT decomposition, each level of decomposition in the tree-structured filter bank consisting of a two-channel filter bank,αandβare respectively a dual-channel filter bankMedium-low pass filterCoefficient and high-pass filterThe coefficient of (a) is determined,a frequency-domain variable is represented by,,and is made of;Andby a complementary function of powers of 2 pi in periodDefinition of, forAnd satisfies the following conditions:
is provided withAndare respectively the firstjThe equivalent low-pass filter and the equivalent high-pass filter for which the layer is being decomposed are:
and
wherein, the first and the second end of the pipe are connected with each other,is shown asjThe low-pass scaling factor of the layer low-pass filter,jindicating the layer that is currently being decomposed and,is shown asjThe low-pass scaling factor of the layer high-pass filter, in the J-th layer decomposition, see FIG. 2, the preprocessed discrete electroencephalogram signalx[n]Passing through an equivalent low-pass filterObtaining a signal c after low-pass scaling j [n]Referring to FIG. 3, the preprocessed discrete EEG signalx[n]Via an equivalent high-pass filterLow pass scaling, high pass filterIs high-pass scaled to obtain a signal d j [n](ii) a During TQWT decomposition operations, letAndthe sampling frequencies of the 0 th level decomposition and the 1 st level decomposition, respectively, wherein,the sampling frequency of the input original electroencephalogram signal is set as follows:
is provided withf c AndBWthe center frequency and bandwidth of the filter, respectively, and:
there are 3 parameters that TQWT needs to set, which are: quality factor Q, total number of layers J of decomposition, and redundancy r, where:
wherein, the first and the second end of the pipe are connected with each other,f c andBWthe center frequency and the bandwidth of the low-pass filter and the high-pass filter respectively, Q represents an adjustable factor and is the ratio of the center frequency to the bandwidth, the center frequency and the bandwidth can be divided by adjusting Q, the oscillation degree of a signal is controlled,rcontrols the excessive ringing phenomenon, plays the role of localizing the characteristics of the electroencephalogram signals,rsatisfies the following conditions:
since the electroencephalogram signal is a high oscillation signal, the present embodiment sets the value of Q to 3, the parameter r helps to localize the wavelet in the time domain without affecting the waveform shape of the electroencephalogram signal, and when performing TQWT, it is necessary to prevent unnecessary wavelet from excessively oscillating by appropriately selecting the value of r greater than or equal to 3, and therefore, in the present embodiment, the value of r is set to 3.
S3, analyzing the energy of each sub-band component, and selecting a plurality of sub-band components from each sub-band to reconstruct signals to obtain reconstructed electroencephalogram signals;
since TQWT conforms to the Pasteval theorem, which indicates that the total energy of the wavelet coefficients is equal to the energy of the signal, the total number of sub-band components decomposed by the wavelet transform TQWT with an adjustable Q factor is J +1, let the number of sub-band components decomposed by the wavelet transform TQWT with an adjustable Q factor beiThe individual subband components being denoted as s i [n],i=1,2, …, J +1, its energy is expressed as:
wherein the content of the first and second substances,E i is shown asiThe energy of the subband components; total energy of J +1 subband componentsEComprises the following steps:
then it is firstiEnergy ratio of individual sub-bandsQ i Comprises the following steps:
total energy in J +1 subband componentsEBased on 90%, sorting the energy of each subband component from large to small, and selecting the energy of the subband from J +1 subband components according to the sorting orderEAnd (3) performing signal reconstruction on the P subband components of 90%, wherein P is less than J +1.
When P subband components are subjected to signal reconstruction when a plurality of subband components are subjected to signal reconstruction, the time sequence of the residual (J + 1-P) subband components in the J +1 subband components is set to be zero, namely the (J + 1-P) subband components are set to be 0, finally, the P subband components and the residual (J + 1-P) subband components are reconstructed by utilizing a TQWT synthesis filter group to obtain a reconstructed electroencephalogram signal, and the calculation complexity is reduced. In this embodiment, 5 subband components with the largest energy ratio are selected for signal reconstruction, since the subband components with the high energy ratio of the electroencephalogram signal are uniformly distributed in the front part, if the number of decomposition layers is increased from the J layer to the J +1 layer, TQWT performs low-pass filtering on the last subband component only on the basis of the J-th layer decomposition, and does not cause any influence on the energy distribution of the front subband, so in this embodiment, the number of decomposition layers J is set to 8, and TQWT will perform preprocessing on the discrete electroencephalogram signalx[n]The method comprises the steps of decomposing the electroencephalogram signal into 9 subband components, selecting 5 subband components with the largest energy, and selecting the energy sum of the 5 subband components to be more than 90% of the total energy, wherein the 5 subband components contain most information of the electroencephalogram signal.
And S4, carrying out continuous variational modal decomposition (SVMD) optimization and reprocessing on the reconstructed electroencephalogram signal, and decomposing out a plurality of modal components.
Continuous variational modal decomposition is accomplished by continuously applying Variational Modal Extraction (VME) on the signal, with some constraints added to avoid convergence to previously extracted modalities. This process will know that all modal components are extracted or will terminate when the reconstruction error (the error of the sum of the input signal and the modulus) is less than a threshold, and the reconstructed EEG signal is represented asf(t) To, forf(t) The decomposition of the variation mode is continuously applied,f(t) Is decomposed into two signals, the expression satisfies:wherein, in the step (A),the L-th modal component is represented,representing a residual, satisfying:
comprising two parts, one being the sum of the first L-1 modal components set to be acquiredOne is the unprocessed part of the signalThe sum of the first L-1 modal components obtained when looking for the first modal component of SVMDIs zero; the specific process of carrying out continuous variational modal decomposition (SVMD) comprises the following steps:
s41, setting、、、A value of (1), whereinIn order to be a penalty factor,、、is a threshold parameter for determining whether a cycle is over;
s42, before the first iteration of continuous variation modal decomposition, the frequency domain response of the L-th modal component is obtained by initialization settingCenter frequency of L-th modal componentAnd lagrange multiplier;
S43. Application、、Is updated by the iterative formula、、Wherein, in the step (A),the iterative formula of (a) is:
wherein the content of the first and second substances,is composed off(t) As a result of the fourier transform, the result,is a frequency domain variable;the iterative formula of (a) is:
wherein the content of the first and second substances,it is indicated that the parameters of the update,representing an intermediate parameter;
s44, according to the discriminant:
a determination is made, wherein,Tis a signalf(t) If the discriminant is not satisfied, the process returns to step S42 to extract the next modality, and if the discriminant is satisfied, it indicates that all the modalities have been extracted, and SVMD decomposition in the continuous variation modality is completed. And obtaining a plurality of modes after SVMD processing through continuous variational mode decomposition through a plurality of iterations. The reconstructed signal is processed by using the SVMD, modal mixing existing in the traditional EMD can be inhibited, the operation speed can be greatly improved compared with that of a VMD, and the method has stronger robustness on a central frequency initial value.
Example 2
In this embodiment, in addition to the method provided in embodiment 1 being used for denoising and optimizing an electroencephalogram signal, the method further includes a process of classifying a plurality of modal components according to spectral characteristics, and the process is used for subsequent feature extraction of the electroencephalogram signal.
The modal quantity obtained after the embodiment 1 cannot be determined, one electroencephalogram signal can generate 4~8 modal components after being subjected to continuous variational modal decomposition processing, the decomposed intrinsic modal functions are arranged from low to high according to the frequency, as the characteristics extracted from the signal in the same frequency range have more commonalities, the decomposed modal components correspond to a plurality of categories according to the characteristics of the frequency spectrum, the modal components obtained after the continuous variational modal decomposition VMD processing are arranged from low to high according to the frequency sequence, the plurality of modal components are classified according to the spectral characteristics and then reconstructed, when the modal components are classified according to the spectral characteristics, as shown in fig. 4, the plurality of modal components are classified according to the spectral characteristics and then reconstructed, and when the spectral characteristics are classified, if the modal quantity is consistent with the number of the spectral characteristics, each modal component is correspondingly classified with each spectral characteristics category; if the number of the modes is larger than the number of the frequency spectrum characteristic categories, the mode components are firstly combined with the mode components adjacent to each other from back to front in sequence, and are correspondingly classified into one frequency spectrum characteristic category, so that the consistency of the length of the feature vector of the electroencephalogram signal is ensured, and the feature extraction of the subsequent electroencephalogram signal is facilitated.
For example, assume that Ck (t), k =1,2,.., 8 represents the mode decomposed after embodiment 1, and I1, I2, I3, I4 represent 4 types of functions combined according to the characteristics of the mode spectrum. When the SVMD decomposes 4 modes of electroencephalogram signals, each mode is used as a function, namely C1 (t) belongs to I1, C2 (t) belongs to I2, C3 (t) belongs to I3, and C4 (t) belongs to I4; when the continuous VMD decomposes the EEG signal into 5 modes, the 4 th mode and the 5 th mode are combined into a class of functions, and other three mode functions are respectively used as a class of functions, namely C1 (t) belongs to I1, C2 (t) belongs to I2, C3 (t) belongs to I3, and C4 (t) + C5 (t)), belongs to I4; when the SVMD decomposes the electroencephalogram signal into 6 modes, the 3 rd mode and the 4 th mode are combined into a class of functions, the 5 th mode and the 6 th mode are combined into a class of functions, and the other two modes are respectively used as a class of functions, namely C1 (t) belongs to I1, C2 (t) belongs to I2, (C3 (t) + C4 (t)) belongsto I3, (C5 (t) + C6 (t)) belongsto I4; when the SVMD decomposes the electroencephalogram signal into 8 modes, the 1 st mode and the 2 nd mode are taken as a class function, the 3 rd mode and the 4 th mode are taken as a class function, the 5 th mode and the 6 th mode are taken as a class function, and the 7 th mode and the eighth mode are taken as a class function.
Example 3
As shown in fig. 5, this embodiment provides an electroencephalogram signal denoising optimization system based on TQWT and SVMD, which includes:
the electroencephalogram signal preprocessing module is used for preprocessing the acquired original electroencephalogram signals to obtain preprocessed discrete electroencephalogram signals;
the TQWT decomposition module is used for performing wavelet transformation TQWT decomposition operation with adjustable Q factors on the preprocessed discrete electroencephalogram signals to obtain each sub-band component;
the signal analysis and reconstruction module is used for analyzing the energy of each sub-band component and selecting a plurality of sub-band components from each sub-band to carry out signal reconstruction so as to obtain a reconstructed electroencephalogram signal;
and the continuous variational modal decomposition module is used for carrying out SVMD optimization reprocessing on the reconstructed electroencephalogram signal to decompose a plurality of modal components.
It should be understood that the above-described 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. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (6)
1. A TQWT and SVMD based electroencephalogram signal denoising optimization method is characterized by comprising the following steps:
s1, collecting original electroencephalogram signals, and preprocessing the original electroencephalogram signals to obtain preprocessed discrete electroencephalogram signals;
s2, performing wavelet transformation TQWT decomposition operation with adjustable Q factors on the preprocessed discrete electroencephalogram signals to obtain sub-band components;
s3, analyzing the energy of each sub-band component, and selecting a plurality of sub-band components from each sub-band to reconstruct signals to obtain reconstructed electroencephalogram signals;
the total number of sub-band components decomposed by the wavelet transform TQWT with adjustable Q factor is J +1, and in step S3, the number one decomposed by the wavelet transform TQWT with adjustable Q factor is setiThe sub-band components are denoted as s i [n],i=1,2, …, J +1, its energy is expressed as:
wherein the content of the first and second substances,E i is shown asiThe energy of the subband components; total energy of J +1 subband componentsEComprises the following steps:
then it is firstiEnergy ratio of individual sub-bandsQ i Comprises the following steps:
total energy in J +1 subband componentsEBased on 90%, sorting the energy of each subband component from large to small, and selecting the energy of the subband from J +1 subband components according to the sorting orderEPerforming signal reconstruction on P subband components of 90%, P < J +1; when P subband components are subjected to signal reconstruction, setting the time sequence of the remaining (J + 1-P) subband components in the J +1 subband components to zero, namely setting the (J + 1-P) subband components to be 0, and finally reconstructing the P subband components and the remaining (J + 1-P) subband components by utilizing a synthesis filter group of TQWT to obtain a reconstructed electroencephalogram signal; when P subband components are subjected to signal reconstruction, the time sequence of the remaining (J + 1-P) subband components in the J +1 subband components is set to zero, namely the (J + 1-P) subband components are set to be 0, the components with the same number and the same point number are used for reconstruction through a synthesis filter of TQWT, the unselected components are set to be 0, but the zero-set components are added into the synthesis filter of the TQWT for signal reconstruction;
s4, carrying out continuous variational modal decomposition (SVMD) optimization reprocessing on the reconstructed electroencephalogram signal, and decomposing out a plurality of modal components;
classifying the plurality of modal components according to the spectral characteristics for subsequent feature extraction of the electroencephalogram signals; the modal components obtained after SVMD processing are arranged in sequence from low frequency to high frequency, a plurality of modal components are classified according to the spectral characteristics and then reconstructed, and when the modal components are classified according to the spectral characteristics, if the modal quantity is consistent with the number of the spectral characteristics, each modal component is correspondingly classified with each spectral characteristics category; if the number of the modes is larger than the number of the spectral characteristic categories, the mode components are firstly combined with the mode components adjacent to each other in the order from back to front, and the combination is correspondingly classified into a spectral characteristic category.
2. The method for de-noising and optimizing electroencephalogram signals based on TQWT and SVMD as claimed in claim 1, wherein in step S1, the preprocessing operation performed on the original electroencephalogram signals is:
setting a signal extraction frequency band, carrying out discrete Fourier transform processing on an original electroencephalogram signal, then extracting the electroencephalogram signal which is subjected to discrete Fourier transform processing and is positioned in the signal extraction frequency band by using a low-pass filter, then carrying out discrete Fourier inverse transform on the extracted electroencephalogram signal, and obtaining a preprocessed discrete electroencephalogram signalx[n],nRepresenting a time index of the discrete brain electrical signal.
3. The method for de-noising and optimizing electroencephalogram signals based on TQWT and SVMD as claimed in claim 2, wherein in step S2, it is assumed that the preprocessed discrete electroencephalogram signals are grouped by using a filter with a tree structurex[n]Performing a wavelet transform TQWT decomposition operation with an adjustable Q factor, andx[n]the decomposition is J +1 subband component, J represents the total number of layers of TQWT decomposition, each level of decomposition in the filter bank with tree structure is composed of a dual-channel filter bank,αandβlow pass filters in dual channel filter bankCoefficient and high-pass filterThe coefficient of (a) is calculated,a frequency-domain variable is represented by,,and is made of;Andby a power-complementary function of period 2 piDefinition of, forSatisfies the following conditions:
4. the method for de-noising and optimizing electroencephalogram signals based on TQWT and SVMD as claimed in claim 3, wherein the method comprisesAndare respectively the firstjThe equivalent low-pass filter and the equivalent high-pass filter for which the layer is being decomposed are:
and
wherein the content of the first and second substances,is shown asjThe low-pass scaling factor of the layer low-pass filter,jindicating the layer that is currently being decomposed and,denotes the firstjLow-pass scaling factor of the layer high-pass filter, in the second placejDiscrete electroencephalogram signal after pretreatment in layer decompositionx[n]Via an equivalent low-pass filterObtaining a signal c after low-pass scaling j [n]Pre-processed discrete electroencephalogram signalsx[n]Via an equivalent high-pass filterLow pass scaling, high pass filterIs subjected to high-pass scaling to obtain a signal d j [n](ii) a During TQWT decomposition operations, three parameters are set, respectively: quality factor Q, total number of layers decomposed J, and redundancyrWherein:
wherein the content of the first and second substances,andBWthe center frequency and the bandwidth of the low-pass filter and the high-pass filter respectively, Q represents an adjustable factor and is the ratio of the center frequency to the bandwidth, the center frequency and the bandwidth can be divided by adjusting Q, the oscillation degree of a signal is controlled,rcontrols the excessive ringing phenomenon, plays the role of localizing the characteristics of the electroencephalogram signals,rsatisfies the following conditions:
5. the method for de-noising and optimizing electroencephalogram signals based on TQWT and SVMD as claimed in claim 4, wherein the reconstructed electroencephalogram signal is represented asf(t) To, forf(t) Continuously applying variation modal decomposition, sequentially searching modal components,f(t) Is decomposed into two signals, the expression satisfies:wherein, in the step (A),the L-th modal component is represented,represents a redundant signal satisfying:
comprises two parts, one of which is set to obtain the sum of the first L-1 modal componentsOne is the unprocessed part of the signalThe sum of the first L-1 modal components obtained when looking for the first modal component of SVMDIs zero; the specific process of carrying out continuous variational modal decomposition (SVMD) comprises the following steps:
s41, setting up、、、A value of (1), whereinIn order to be a penalty factor,、、is a threshold parameter for determining whether a cycle is over;
s42, before the first continuous variation modal decomposition iteration, the frequency domain response of the L-th modal component is obtained through initialization settingCenter frequency of L-th modal componentAnd lagrange multiplier;
S43. Use、、Is updated by the iterative formula、、Wherein, in the process,the iterative formula of (c) is:
wherein the content of the first and second substances,is composed off(t) As a result of the fourier transform, the result,is a frequency domain variable;the iterative formula of (a) is:
wherein the content of the first and second substances,it is indicated that the parameters of the update,representing an intermediate parameter;
s44, according to the discriminant:
a determination is made, wherein,Tis a signalf(t) If the discriminant is not satisfied, the process returns to step S42 to extract the next modality, and if the discriminant is satisfied, it indicates that all the modalities have been extracted, and SVMD decomposition in the continuous variation modality is completed.
6. A system for de-noising and optimizing electroencephalogram signals based on TQWT and SVMD is characterized by comprising:
the electroencephalogram signal preprocessing module is used for preprocessing the acquired original electroencephalogram signals to obtain preprocessed discrete electroencephalogram signals;
the TQWT decomposition module is used for performing wavelet transformation TQWT decomposition operation with adjustable Q factors on the preprocessed discrete electroencephalogram signals to obtain each sub-band component;
the signal analysis and reconstruction module is used for analyzing the energy of each sub-band component and selecting a plurality of sub-band components from each sub-band to carry out signal reconstruction so as to obtain a reconstructed electroencephalogram signal;
the total number of sub-band components decomposed by the wavelet transform TQWT with adjustable Q factor is J +1, and the number of sub-band components decomposed by the wavelet transform TQWT with adjustable Q factor is setiThe individual subband components being denoted as s i [n],i=1,2, …, J +1, its energy is expressed as:
wherein the content of the first and second substances,E i is shown asiThe energy of the subband component; total energy of J +1 subband componentsEComprises the following steps:
then it is firstiEnergy ratio of individual sub-bandsQ i Comprises the following steps:
total energy in J +1 subband componentsEBased on 90%, sorting the energy of each subband component from large to small, and selecting the energy of the subband from J +1 subband components according to the sorting orderEPerforming signal reconstruction on the P subband components of 90 percent, wherein P is less than J +1; when P subband components are subjected to signal reconstruction, setting the time sequence of the remaining (J + 1-P) subband components in the J +1 subband components to zero, namely setting the (J + 1-P) subband components to be 0, and finally reconstructing the P subband components and the remaining (J + 1-P) subband components by utilizing a synthetic filter group of TQWT to obtain a reconstructed electroencephalogram signal; when P subband components are subjected to signal reconstruction, the time sequence of the remaining (J + 1-P) subband components in the J +1 subband components is set to zero, namely the (J + 1-P) subband components are set to be 0, the components with the same number and the same point number are used for reconstruction through a synthesis filter of TQWT, the unselected components are set to be 0, but the zero-set components are added into the synthesis filter of the TQWT for signal reconstruction;
the continuous variational modal decomposition module is used for carrying out SVMD optimization reprocessing on the reconstructed electroencephalogram signal to decompose a plurality of modal components;
classifying the plurality of modal components according to the spectral characteristics for subsequent feature extraction of the electroencephalogram signals; the modal components obtained after SVMD processing are arranged in sequence from low frequency to high frequency, a plurality of modal components are classified according to the spectral characteristics and then reconstructed, and when the modal components are classified according to the spectral characteristics, if the modal quantity is consistent with the number of the spectral characteristics, each modal component is correspondingly classified with each spectral characteristics category; if the number of the modes is larger than the number of the spectral characteristic categories, the mode components are firstly combined with the mode components adjacent to each other in the order from back to front, and the combination is correspondingly classified into a spectral characteristic category.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210659050.5A CN114757236B (en) | 2022-06-13 | 2022-06-13 | Electroencephalogram signal denoising optimization method and system based on TQWT and SVMD |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210659050.5A CN114757236B (en) | 2022-06-13 | 2022-06-13 | Electroencephalogram signal denoising optimization method and system based on TQWT and SVMD |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114757236A CN114757236A (en) | 2022-07-15 |
CN114757236B true CN114757236B (en) | 2022-10-11 |
Family
ID=82336098
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210659050.5A Active CN114757236B (en) | 2022-06-13 | 2022-06-13 | Electroencephalogram signal denoising optimization method and system based on TQWT and SVMD |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114757236B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116687406B (en) * | 2023-05-06 | 2024-01-02 | 粤港澳大湾区精准医学研究院(广州) | Emotion recognition method and device, electronic equipment and storage medium |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103777237A (en) * | 2012-10-17 | 2014-05-07 | 中国石油化工股份有限公司 | Surface elevation smoothing method based on space-variant weighted banding wave-number domain filtering |
CN106951835A (en) * | 2017-03-03 | 2017-07-14 | 东华大学 | A kind of EEG signals noise remove method |
CN107392364A (en) * | 2017-07-12 | 2017-11-24 | 河海大学 | The short-term load forecasting method of variation mode decomposition and depth belief network |
CN110916716A (en) * | 2019-12-30 | 2020-03-27 | 龙岩学院 | Wearable heart sound monitoring facilities |
CN111307277A (en) * | 2020-03-20 | 2020-06-19 | 北京工业大学 | Single-mode sub-signal selection method based on variational modal decomposition and predictive performance |
CN111914735A (en) * | 2020-07-29 | 2020-11-10 | 天津大学 | Epilepsia electroencephalogram signal feature extraction method based on TQWT and entropy features |
CN112287796A (en) * | 2020-10-23 | 2021-01-29 | 电子科技大学 | Radiation source identification method based on VMD-Teager energy operator |
CN112362756A (en) * | 2020-11-24 | 2021-02-12 | 长沙理工大学 | Concrete structure damage monitoring method and system based on deep learning |
CN112784686A (en) * | 2020-12-29 | 2021-05-11 | 聊城大学 | Adaptive signal analysis method based on continuous multivariable variational modal decomposition |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110865357B (en) * | 2019-10-25 | 2023-03-31 | 南京理工大学 | Laser radar echo signal noise reduction method based on parameter optimization VMD |
-
2022
- 2022-06-13 CN CN202210659050.5A patent/CN114757236B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103777237A (en) * | 2012-10-17 | 2014-05-07 | 中国石油化工股份有限公司 | Surface elevation smoothing method based on space-variant weighted banding wave-number domain filtering |
CN106951835A (en) * | 2017-03-03 | 2017-07-14 | 东华大学 | A kind of EEG signals noise remove method |
CN107392364A (en) * | 2017-07-12 | 2017-11-24 | 河海大学 | The short-term load forecasting method of variation mode decomposition and depth belief network |
CN110916716A (en) * | 2019-12-30 | 2020-03-27 | 龙岩学院 | Wearable heart sound monitoring facilities |
CN111307277A (en) * | 2020-03-20 | 2020-06-19 | 北京工业大学 | Single-mode sub-signal selection method based on variational modal decomposition and predictive performance |
CN111914735A (en) * | 2020-07-29 | 2020-11-10 | 天津大学 | Epilepsia electroencephalogram signal feature extraction method based on TQWT and entropy features |
CN112287796A (en) * | 2020-10-23 | 2021-01-29 | 电子科技大学 | Radiation source identification method based on VMD-Teager energy operator |
CN112362756A (en) * | 2020-11-24 | 2021-02-12 | 长沙理工大学 | Concrete structure damage monitoring method and system based on deep learning |
CN112784686A (en) * | 2020-12-29 | 2021-05-11 | 聊城大学 | Adaptive signal analysis method based on continuous multivariable variational modal decomposition |
Non-Patent Citations (4)
Title |
---|
Successive variational mode decomposition;MojtabaNazari et.al;《Signal Processing》;20200505;第174卷;第1-10页 * |
基于信号共振稀疏分解的风电机组齿轮箱故障诊断;刘一鸣;《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》;20200115(第01期);第C042-504页 * |
基于可调Q因子小波变换和迁移学习的癫痫脑电信号检测;罗婷瑞 等;《计算机科学》;20200731;第47卷(第7期);第1-7页 * |
基于小波分解和IIR滤波器的最优均衡器设计;凌永权 等;《电视技术》;20210331;第45卷(第3期);第1-2页 * |
Also Published As
Publication number | Publication date |
---|---|
CN114757236A (en) | 2022-07-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109784242B (en) | Electroencephalogram signal denoising method based on one-dimensional residual convolution neural network | |
Islam et al. | A wavelet-based artifact reduction from scalp EEG for epileptic seizure detection | |
CN106709469B (en) | Automatic sleep staging method based on electroencephalogram and myoelectricity multiple characteristics | |
CN110353672B (en) | System and method for removing eye artifacts in electroencephalogram signals | |
CN111310570B (en) | Electroencephalogram signal emotion recognition method and system based on VMD and WPD | |
CN107260166A (en) | A kind of electric artefact elimination method of practical online brain | |
CN107411739A (en) | EEG signals Emotion identification feature extracting method based on dual-tree complex wavelet | |
Patil et al. | Quality advancement of EEG by wavelet denoising for biomedical analysis | |
CN114266276A (en) | Motor imagery electroencephalogram signal classification method based on channel attention and multi-scale time domain convolution | |
CN111543984B (en) | Method for removing ocular artifacts of electroencephalogram signals based on SSDA | |
CN114757236B (en) | Electroencephalogram signal denoising optimization method and system based on TQWT and SVMD | |
Abdullah et al. | Seizure detection by means of hidden Markov model and stationary wavelet transform of electroencephalograph signals | |
Zachariah et al. | Automatic EEG artifact removal by independent component analysis using critical EEG rhythms | |
CN113842115A (en) | Improved EEG signal feature extraction method | |
CN112861625B (en) | Determination method for stacked denoising self-encoder model | |
CN114081503A (en) | Method for removing ocular artifacts in electroencephalogram signals | |
Gao et al. | EEG reconstruction with a dual-scale CNN-LSTM model for deep artifact removal | |
CN117064405A (en) | Single-channel electroencephalogram signal artifact removal method, equipment and medium | |
CN113180659A (en) | Electroencephalogram emotion recognition system based on three-dimensional features and cavity full convolution network | |
CN111493864B (en) | EEG signal mixed noise processing method, equipment and storage medium | |
CN115462803A (en) | BG-Attention-based electroencephalogram signal denoising method, device and storage medium | |
CN115270847A (en) | Design decision electroencephalogram recognition method based on wavelet packet decomposition and convolutional neural network | |
CN114699078A (en) | Emotion recognition method and system based on small number of channel EEG signals | |
Islam et al. | Wavelet-based Artifact Removal Algorithm for EEG Data by Optimizing Mother Wavelet and Threshold Parameters | |
Mir et al. | Adaptive data analysis methods for biomedical signal processing applications |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |