CN108573207A - EMD and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method - Google Patents
EMD and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method Download PDFInfo
- Publication number
- CN108573207A CN108573207A CN201711402734.2A CN201711402734A CN108573207A CN 108573207 A CN108573207 A CN 108573207A CN 201711402734 A CN201711402734 A CN 201711402734A CN 108573207 A CN108573207 A CN 108573207A
- Authority
- CN
- China
- Prior art keywords
- matrix
- carried out
- signal
- feature
- csp
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2134—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on separation criteria, e.g. independent component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses EMD and CSP to merge most optimum wavelengths space filtering brain electrical feature extracting method, this method carries out empirical mode (EMD) to pretreated signal and decomposes, obtain intrinsic mode function (Intrinsic Mode Functions, IMFs), observe and calculate the energy spectrum of each IMF components, screen effective IMF frequency ranges (5 28Hz), form new signal matrix, most optimum wavelengths calculating is carried out to it, it reuses CSP filters and is filtered acquisition feature, finally use support vector machines (Support Vector Machine, SVM) classify.It is 95% or more that classification results, which obtain 9 tested imagination movement average correct classification rates, ensure that the feasibility and validity of the present invention.
Description
Technical field
The present invention relates to EMD and CSP to merge most optimum wavelengths space filtering brain electrical feature extracting method, belongs to intelligent information
Processing technology field.
Background technology
Traditional moving conduit is made of cerebral nerve and muscle, and nerve conduction impulsion, muscle cooperation is completed corresponding
Action, and brain-computer interface (Brain-Computer Interface, BCI) then provides another moving conduit, does not depend on
Traditional moving conduit is directly connected with external equipment by brain mind, establishes moving conduit, the brain mind control of employment
External equipment needs not move through nerve conduction and muscular movement, and one kind is provided newly for the patient of neurotrosis or muscle damage
Motion mode, it is no longer necessary to rely on others' treatment, movement oneself can be completed.The development of brain-computer interface technical field is not
Can only help paralytic using the electronic equipments such as computer, nerve prosthesis, mechanical arm, also achieve including:It moves extensive
Other functions such as multiple, communication, environmental Kuznets Curves even amusement.
Brain-computer interface technology includes mainly signal acquisition, pretreatment, feature extraction, tagsort and interface equipment control
Deng five steps.Wherein, the characteristic signal obtained by feature extraction can identify the differentiation letter of different imagination movement EEG signals
Breath, has a great impact to subsequent Classification and Identification, therefore feature extraction is widely paid close attention in BCI research circle.
Effective feature extracting method is the key that improve accuracy of identification, currently, T/F method is widely used
In the research of EEG signals.Traditional T/F method includes：Short Time Fourier Transform (Short-timeFourier
Transform, STFT), wavelet transformation (WaveletTransform, WT) etc., but the essence of these methods is all based on Fu
In leaf transformation, according to Heisenberg's uncertainty principle, this method can not possibly obtain the fine resolution of T/F simultaneously.Closely
Year, Hilbert-Huang transform (Hilbert-Huang Transform, HHT) has been used as another T/F analytic approach
Become to become more and more popular, while it is also very suitable for analyzing non-linear and non-stationary signal.Original signal is by empirical mode point
Solution (Empirical Mode Decomposition, EMD) is broken down into a series of intrinsic mode function (Intrinsic
Mode Functions, IMFs), Hilbert-Huang transform then is carried out to each intrinsic mode function, seeks its corresponding energy
Spectrum and marginal spectrum, classify as feature.HHT, which is not related to Heisenberg's uncertainty principle, can obtain time domain and frequency domain
High-resolution.It is widely used in many field of signal processing at present, such as radar detection, seismic signal and biomedicine signals
Deng.
Furthermore due to EEG signal low spatial resolution, the BCI systems that EEG signal is constituted need to carry out effective space
Filtering, so that it is guaranteed that extracting characteristic information from tested related brain domain.In this regard, common algorithm has：Cospace mould
Formula (Common Spatial Pattern, CSP), independent principal component analysis (Independent Component
Analysis, ICA) and co-domain space spectral model (Common Spatial Spectral Pattern, CSSP), filter CSP
(Filter BankCommon Spatial Pattern, FBCSP), differentiate filtering CSP (Discriminant Filtering
Common Spatial Pattern, DFBCSP) etc. a variety of CSP innovatory algorithms.
However, traditional CSP needs a large amount of input channel, while lacking frequency information.CSP algorithms mainly pass through
Between the judgement of the spatial information of brain power supply come the very different realizing different classifications, algorithm essence is to utilize square on algebraically
The theory of battle array simultaneous diagonalization, finds one group of spatial filter, and EEG signals are obtained more apparent by the projection of this group of filter
Feature vector.Since EEG signals are extremely complicated non-linear, unstable signals, in collected signal, there is brain electric incessantly
Signal, also other impurity signals, even if still there are other aliasing signals in the frequency range of EEG signals, this
A little impurity and aliasing signal will effect characteristics extraction validity, and simple CSP filtering comes simply by spatial information
The validity for judging EEG signals, there are certain deficiencies, may not remove impurity and aliasing signal from frequency range dry
Only, if many frequency signals unrelated with Mental imagery are mixed in the validity for wherein having seriously affected feature vector.And this
Invention can be good at solving the problems, such as above.
Invention content
The present invention is in view of the above shortcomings of the prior art, it is proposed that a kind of EMD and CSP fusions most optimum wavelengths space filtering
Brain electrical feature extracting method, provides the accuracy rate of pattern-recognition significantly.This method is to be based on following content：1) it is based on experience
The EEG signal of Mode Decomposition is handled.2) intrinsic mode function is screened according to spectrum analysis, forms new signal matrix.3) to letter
Number matrix carries out most optimum wavelengths calculating, then carries out public space pattern decomposition, solves CSP multi inputs, lacks asking for frequency domain information
Topic.4) support vector cassification.
The technical scheme adopted by the invention to solve the technical problem is that：The present invention first carries out signal after pretreatment
It is solid to filter out 2 ranks before concentrating on imagination movement frequency range according to the spectrum analysis of each rank intrinsic mode function for empirical mode decomposition
There is mode function, most optimum wavelengths calculating is carried out to it, is further carried out CSP space filterings, is carried out by support vector machines special
Sign selection obtains final classification result.
The present invention to the extraction of channel signal improve：
Left hand is imagined and is moved, the data in two channels extraction C3 and C4 are as original signal；The right hand is imagined and is transported
It is dynamic, the data in the channels C3 are only extracted as original signal.The channels C3 of right hand imagination movement are moved with the left hand imagination respectively
The channel C3, C4 is compared, and feature vector is extracted, and carries out Classification and Identification.
The present invention improves traditional most optimum wavelengths calculating, including：
Most optimum wavelengths calculating is carried out to signal, it is as follows
2：N 1：N-1
Δxi=xi-xi
Wherein N is the sampling number of signal, Xi:jRepresentation signal matrix is arranged from the i-th row to jth, while retaining first row
Initial data；Traditional is converted with row matrix.
Method flow：
Step 1：9 tested EEG signals are chosen as training set and test set, respectively to single tested C3, C4 two
Signal in a channel is pre-processed；
Step 2：Empirical mode decomposition is carried out to pretreated EEG signal x (t)；Obtain a series of intrinsic mode functions
IMFi(i is the exponent number of intrinsic mode function) simultaneously draws all intrinsic mode function energy spectrum diagrams；
EEG signal carries out empirical mode decomposition and is as follows：
(1) local extremum for judging each x (t), is carried out curve fitting with cubic spline curve, and local maximum is formed
Coenvelope emax(t), local minimum forms lower envelope emin(t)。
(2) e is soughtmax(t) and emin(t) mean value：
(3) difference of input signal x (t) and m (t) are calculated:
C (t)=x (t)-m (t) (2)
If c (t) cannot meet the definition of IMF by condition, (1)-(3) are repeated the above process, otherwise, extraction c (t)
As intrinsic mode function, surplus r (t) calculates as follows：
R (t)=x (t)-c (t) (3)
(4) the surplus data new as one are next more low-frequency solid to obtain by identical screening process
There is mode function.Until survival function r (t) be a monotonic function or only there are one it is ultimate attainment when, decomposable process stop.It is false
If original signal x (t) is broken down into n intrinsic mode function and a survival function amount r (t), reconstruction signal can be obtained：
Step 3：By the C3 of single test, 2 rank IMF components merge before the channels C4, constitute the matrix X of a 4*2000i(i
=L indicates that imagination left hand movement, i=R indicate the movement of the imagination right hand), wherein 4 be IMF number, can regard port number as, 2000
For the sampled point number once tested, i.e. length of window.
Step 4：Most optimum wavelengths calculating is carried out to above-mentioned IMF matrixes：
2：N 1：N-1
Δxi=xi -xi (5)
Wherein N is the sampling number of signal, Xi:jRepresentation signal matrix is arranged from the i-th row to jth；To the matrix after transformation into
Row public space pattern decomposes；
Public space pattern algorithm detailed process is as follows：
Movement A and B carries out T respectively to be imagined to two classesA,TBSecondary experiment, TA,TBFor positive integer；
(1) covariance of blending space is calculated；
First, the covariance that two type games imaginary signals are tested every time is calculated, formula is as follows：
Wherein, trace (XXT) it is matrix XXTMark, i.e. matrix XXTThe sum of diagonal entry；
Then, the average covariance of the two type games imagination is calculated separately：
Wherein, CA,i、CB,iThe covariance of the ith experiment of Mental imagery A and B is indicated respectively；
And then acquire the covariance of blending space：
CM=CA+CB (8)
(2) Eigenvalues Decomposition is carried out to blending space covariance, formula is as follows：
Wherein, UMIt is characterized vector matrix, ΛMIt is characterized value diagonal matrix；
(3) whitening processing is carried out；
To ΛMIt carries out descending sort and obtains ΛMd, and to UMIt does same row-column transform and obtains UMd；It enables
To CA、CBWhitening processing is carried out respectively, and formula is as follows：
SA=PCAPT
SB=PCBPT (10)
Utilize SA、SBThe characteristics of feature vector having the same, can obtain after Eigenvalues Decomposition：
SA=B ΛABT
SB=B ΛBBT (11)
Wherein, B SAWith SBCommon trait vector, ΛA、ΛBRespectively SAAnd SBFeature diagonal matrix, and ΛA+
ΛB=I, I are unit matrix；
Therefore acquiring spatial filter matrices is：
W=BTP (12)
W is carried out to X to filter：
Z0=WX (13)
(4) feature vector f is sought；
Extract Z0Preceding m rows and rear m rows, constitute Z=[z1,z2,…,z2m]T, then carry out feature extraction, calculation formula
It is as follows：
Wherein, var () indicates variance, i=1,2 ..., 2m, then feature vector f=[f1,f2,…,f2m]T；Step 5：
Tagsort is carried out using support vector machines, is as follows：
(1) it for non-linear EEG problems, converts by nonlinear transformation feature set to linear in another space
Problem constructs optimal classification surface.Optimal decision function is accordingly：
Wherein N is supporting vector number, and α i are Lagrangue multipliers.To which object function becomes that following formula is made to minimize：
Nuclear parameter γ and error penalty factor are the major parameters for influencing SVM performances.The value of γ influences spatial alternation
Data distribution afterwards, penalty factor then determine the convergence rate and Generalization Ability of support vector machines；Therefore, to γ's and C
Selection largely affects the discrimination of EEG signals.
(2) gridding cross validation method is used to carry out the selection of γ and C optimized parameters, using training set as original
Data set changes the value of kernel function and penalty factor, classifies with cross validation method in a certain range, selection point
Class accuracy rate highest γ and C is as optimal parameter.
(3) after determining γ and C, test set is inputted into trained support vector machines and carries out tagsort.
Advantageous effect：
1, the intrinsic mode function after the present invention is decomposed using three channel EMD carries out CSP filtering, on the basis of CSP
The frequency domain information of EMD is added, well solves the problem of CSP lacks frequency domain information.
2, the present invention regards the multistage intrinsic mode function after empirical mode decomposition as multiple input signals progress public space
Mode Decomposition in the case of tri- channels C4, Cz, obtains preferable tagsort as a result, solving general CSP using only C3
Algorithm needs a large amount of input channel problems.
3, improvement is made for most optimum wavelengths algorithm：Since matrix line number is less, row calculating is changed to row meter by the present invention
It calculates, while retaining the initial data of first row, prepare for subsequent algorithm.
Description of the drawings
Fig. 1 is the method for the present invention flow chart.
Specific implementation mode
The present invention is described in further detail with reference to the accompanying drawings of the specification.
As shown in Figure 1, the method for the invention includes the following steps：
Step 1：Acquire everybody tested EEG signals.9 tested EEG signals are chosen as training set and test
Collection respectively pre-processes the signal in single two channels tested C3, C4；
Step 2：Empirical mode decomposition is carried out to pretreated EEG signal x (t)；Obtain a series of intrinsic mode functions
IMFi(i is the exponent number of intrinsic mode function) simultaneously draws all intrinsic mode function energy spectrum diagrams；
EEG signal carries out empirical mode decomposition and is as follows：
(1) local extremum for judging each x (t), is carried out curve fitting with cubic spline curve, and local maximum is formed
Coenvelope emax(t), local minimum forms lower envelope emin(t)。
(2) e is soughtmax(t) and emin(t) mean value：
(3) difference of input signal x (t) and m (t) are calculated:
C (t)=x (t)-m (t) (2)
If c (t) cannot meet the definition of IMF by condition, (1)-(3) are repeated the above process, otherwise, extraction c (t)
As intrinsic mode function, surplus r (t) calculates as follows：
R (t)=x (t)-c (t) (3)
(4) the surplus data new as one are next more low-frequency solid to obtain by identical screening process
There is mode function.Until survival function r (t) be a monotonic function or only there are one it is ultimate attainment when, decomposable process stop.It is false
If original signal x (t) is broken down into n intrinsic mode function and a survival function amount r (t), reconstruction signal can be obtained：
Step 3：By the C3 of secondary experiment, 2 rank IMF components merge before the channels C4, constitute the matrix X of a 4*2000i(i=
L indicates that imagination left hand movement, i=R indicate the movement of the imagination right hand), wherein 4 be IMF number, can regard port number as, 2000 are
The sampled point number once tested, i.e. length of window.
Step 4：Most optimum wavelengths calculating is carried out to above-mentioned IMF matrixes：
2：N 1：N-1
Δxi=xi-xi (5)
Wherein N is the sampling number of signal, Xi:jRepresentation signal matrix is arranged from the i-th row to jth；To the matrix after transformation into
Row public space pattern decomposes；
Public space pattern algorithm detailed process is as follows：
Movement A and B carries out T respectively to be imagined to two classesA,TBSecondary experiment, TA,TBFor positive integer；
(1) covariance of blending space is calculated；
First, the covariance that two type games imaginary signals are tested every time is calculated, formula is as follows：
Wherein, trace (XXT) it is matrix XXTMark, i.e. matrix XXTThe sum of diagonal entry；
Then, the average covariance of the two type games imagination is calculated separately：
Wherein, CA,i、CB,iThe covariance of the ith experiment of Mental imagery A and B is indicated respectively；
And then acquire the covariance of blending space：
CM=CA+CB (8)
(2) Eigenvalues Decomposition is carried out to blending space covariance, formula is as follows：
Wherein, UMIt is characterized vector matrix, ΛMIt is characterized value diagonal matrix；
(3) whitening processing is carried out；
To ΛMIt carries out descending sort and obtains ΛMd, and to UMIt does same row-column transform and obtains UMd；It enables
To CA、CBWhitening processing is carried out respectively, and formula is as follows：
SA=PCAPT
SB=PCBPT (10)
Utilize SA、SBThe characteristics of feature vector having the same, can obtain after Eigenvalues Decomposition：
SA=B ΛABT
SB=B ΛBBT (11)
Wherein, B SAWith SBCommon trait vector, ΛA、ΛBRespectively SAAnd SBFeature diagonal matrix, and ΛA+
ΛB=I, I are unit matrix；
Therefore acquiring spatial filter matrices is：
W=BTP (12)
W is carried out to X to filter：
Z0=WX (13)
(4) feature vector f is sought；
Extract Z0Preceding m rows and rear m rows, constitute Z=[z1,z2,…,z2m]T, then carry out feature extraction, calculation formula
It is as follows：
Wherein, var () indicates variance, i=1,2 ..., 2m, then feature vector f=[f1,f2,…,f2m]T；Step 5：
Tagsort is carried out using support vector machines, is as follows：
(1) it for non-linear EEG problems, converts by nonlinear transformation feature set to linear in another space
Problem constructs optimal classification surface.Optimal decision function is accordingly：
Wherein N is supporting vector number, and α i are Lagrangue multipliers.To which object function becomes that following formula is made to minimize：
Nuclear parameter γ and error penalty factor are the major parameters for influencing SVM performances.The value of γ influences spatial alternation
Data distribution afterwards, penalty factor then determine the convergence rate and Generalization Ability of support vector machines；Therefore, to γ's and C
Selection largely affects the discrimination of EEG signals.
(2) gridding cross validation method is used to carry out the selection of γ and C optimized parameters, using training set as original
Data set changes the value of kernel function and penalty factor, classifies with cross validation method in a certain range, selection point
Class accuracy rate highest γ and C is as optimal parameter.
(3) after determining γ and C, test set is inputted into trained support vector machines and carries out tagsort.
In conclusion be merely preferred embodiments of the present invention, but protection scope of the present invention is not limited to
This.In the disclosed technical scope of invention, the change or replacement that can be readily occurred in, should all cover disclosed herein
Within technical scope.Therefore, protection scope of the present invention should be subject to the scope of protection of the claims.
Claims (4)
1.EMD and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method, which is characterized in that the method includes such as
Lower step：
Step 1：9 tested EEG signals are chosen as training set and test set, respectively to single two channels tested C3, C4
In signal pre-processed；
Step 2：Empirical mode decomposition is carried out to pretreated EEG signal x (t)；Obtain a series of intrinsic mode function IMFi(i
For the exponent number of intrinsic mode function) and draw all intrinsic mode function energy spectrum diagrams；
It is as follows that EEG signal carries out the step of empirical mode decomposition：
(1) local extremum for judging each x (t), is carried out curve fitting with cubic spline curve, and local maximum forms coenvelope
emax(t), local minimum forms lower envelope emin(t)；
(2) e is soughtmax(t) and emin(t) mean value：
(3) difference of input signal x (t) and m (t) are calculated：
C (t)=x (t)-m (t) (2)
If c (t) cannot meet the definition of IMF by condition, (1)-(3) are repeated the above process, otherwise, extraction c (t) is as solid
There are mode function, surplus r (t) to calculate as follows：
R (t)=x (t)-c (t) (3)
(4) the surplus data new as one pass through identical screening process to obtain next more low-frequency natural mode of vibration
Function, until survival function r (t) be a monotonic function or only there are one it is ultimate attainment when, decomposable process stop, it is assumed that original letter
Number x (t) is broken down into n intrinsic mode function and a survival function amount r (t), can obtain reconstruction signal：
Step 3：By the C3 of single test, 2 rank IMF components merge before the channels C4, constitute the matrix X of a 4*2000i(i=L tables
Show that imagination left hand movement, i=R indicate the movement of the imagination right hand), wherein 4 be IMF number, can regard port number as, 2000 be primary
The sampled point number of experiment, i.e. length of window；
Step 4：Most optimum wavelengths calculating is carried out to above-mentioned IMF matrixes：
Wherein N is the sampling number of signal, Xi:jRepresentation signal matrix is from the i-th row to jth row；Matrix after transformation is carried out public
Cospace Mode Decomposition.
2. EMD according to claim 1 and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method, feature
It is：The method changes most optimum wavelengths computational algorithm, includes the following steps：
Wherein N is the sampling number of signal, Xi:jRepresentation signal matrix is arranged from the i-th row to jth, while retaining the original of first row
Signal data；Public space pattern decomposition is carried out to the matrix after transformation.
3. EMD according to claim 1 and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method, feature
It is：The public space pattern algorithm of the method includes the following steps：
Movement A and B carries out T respectively to be imagined to two classesA,TBSecondary experiment, TA,TBFor positive integer；
(1) covariance of blending space is calculated；
First, the covariance that two type games imaginary signals are tested every time is calculated, formula is as follows：
Wherein, trace (XXT) it is matrix XXTMark, i.e. matrix XXTThe sum of diagonal entry；
Then, the average covariance of the two type games imagination is calculated separately：
Wherein, CA,i、CB,iThe covariance of the ith experiment of Mental imagery A and B is indicated respectively；
And then acquire the covariance of blending space：
CM=CA+CB (9)
(2) Eigenvalues Decomposition is carried out to blending space covariance, formula is as follows：
Wherein, UMIt is characterized vector matrix, ΛMIt is characterized value diagonal matrix；
(3) whitening processing is carried out；
To ΛMIt carries out descending sort and obtains ΛMd, and to UMIt does same row-column transform and obtains UMd；It enablesTo CA、CB
Whitening processing is carried out respectively, and formula is as follows：
SA=PCAPT
SB=PCBPT (11)
Utilize SA、SBThe characteristics of feature vector having the same, can obtain after Eigenvalues Decomposition：
SA=B ΛABT
SB=B ΛBBT (12)
Wherein, B SAWith SBCommon trait vector, ΛA、ΛBRespectively SAAnd SBFeature diagonal matrix, and ΛA+ΛB=I,
I is unit matrix；
Therefore acquiring spatial filter matrices is：
W=BTP (13)
W is carried out to X to filter：
Z0=WX (14)
(4) feature vector f is sought；
Extract Z0Preceding m rows and rear m rows, constitute Z=[z1,z2,…,z2m]T, feature extraction is then carried out, calculation formula is as follows：
Wherein, var () indicates variance, i=1,2 ..., 2m, then feature vector f=[f1,f2,…,f2m]T。
4. EMD according to claim 1 and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method, feature
It is：The method carries out tagsort using support vector machines, includes the following steps：
(1) for non-linear EEG problems, feature set is converted to the linear problem in another space by nonlinear transformation,
Optimal classification surface is constructed, corresponding optimal decision function is：
Wherein N is supporting vector number, and α i are Lagrangue multipliers, to which object function becomes that following formula is made to minimize：
Nuclear parameter γ and error penalty factor are the major parameters for influencing SVM performances, and the value of γ influences the number after spatial alternation
According to distribution, penalty factor then determines the convergence rate and Generalization Ability of support vector machines；Therefore, very big to the selection of γ and C
The discrimination of EEG signals is affected in degree；
(2) gridding cross validation method is used to carry out the selection of γ and C optimized parameters, using training set as original data
Collection changes the value of kernel function and penalty factor, classifies with cross validation method, selection sort is accurate in a certain range
Rate highest γ and C is as optimal parameter；
(3) after determining γ and C, test set is inputted into trained support vector machines and carries out tagsort.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711402734.2A CN108573207A (en) | 2017-12-22 | 2017-12-22 | EMD and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711402734.2A CN108573207A (en) | 2017-12-22 | 2017-12-22 | EMD and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108573207A true CN108573207A (en) | 2018-09-25 |
Family
ID=63575984
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711402734.2A Pending CN108573207A (en) | 2017-12-22 | 2017-12-22 | EMD and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108573207A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109480834A (en) * | 2018-12-20 | 2019-03-19 | 浙江大学 | A kind of Method of EEG signals classification based on quick multiple dimension empirical mode decomposition |
| CN111259741A (en) * | 2020-01-09 | 2020-06-09 | 燕山大学 | Electroencephalogram signal classification method and system |
| CN113536882A (en) * | 2021-03-08 | 2021-10-22 | 东北电力大学 | Multi-class motor imagery electroencephalogram signal feature extraction and classification method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104771163A (en) * | 2015-01-30 | 2015-07-15 | 杭州电子科技大学 | Electroencephalogram feature extraction method based on CSP and R-CSP algorithms |
| CN107239142A (en) * | 2017-06-01 | 2017-10-10 | 南京邮电大学 | A kind of EEG feature extraction method of combination public space pattern algorithm and EMD |
-
2017
- 2017-12-22 CN CN201711402734.2A patent/CN108573207A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104771163A (en) * | 2015-01-30 | 2015-07-15 | 杭州电子科技大学 | Electroencephalogram feature extraction method based on CSP and R-CSP algorithms |
| CN107239142A (en) * | 2017-06-01 | 2017-10-10 | 南京邮电大学 | A kind of EEG feature extraction method of combination public space pattern algorithm and EMD |
Non-Patent Citations (1)
| Title |
|---|
| SOUMYA SEN GUPTA等: "Improved Classification of Motor Imagery Datasets for BCI by using Approximate Entropy and WOSF Features", 《 2014 INTERNATIONAL CONFERENCE ON SIGNAL PROCESSING AND INTEGRATED NETWORKS (SPIN)》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109480834A (en) * | 2018-12-20 | 2019-03-19 | 浙江大学 | A kind of Method of EEG signals classification based on quick multiple dimension empirical mode decomposition |
| CN111259741A (en) * | 2020-01-09 | 2020-06-09 | 燕山大学 | Electroencephalogram signal classification method and system |
| CN111259741B (en) * | 2020-01-09 | 2023-04-07 | 燕山大学 | Electroencephalogram signal classification method and system |
| CN113536882A (en) * | 2021-03-08 | 2021-10-22 | 东北电力大学 | Multi-class motor imagery electroencephalogram signal feature extraction and classification method |
| CN113536882B (en) * | 2021-03-08 | 2023-04-07 | 东北电力大学 | Multi-class motor imagery electroencephalogram signal feature extraction and classification method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104771163B (en) | EEG feature extraction method based on CSP and R CSP algorithms | |
| CN107239142A (en) | A kind of EEG feature extraction method of combination public space pattern algorithm and EMD | |
| CN108888264A (en) | EMD and CSP merges power spectral density brain electrical feature extracting method | |
| CN108042132A (en) | Brain electrical feature extracting method based on DWT and EMD fusions CSP | |
| CN105426842B (en) | Multiclass hand motion recognition method based on support vector machines and surface electromyogram signal | |
| CN110163180A (en) | Mental imagery eeg data classification method and system | |
| CN105956624B (en) | Mental imagery brain electricity classification method based on empty time-frequency optimization feature rarefaction representation | |
| CN106909784A (en) | Epileptic electroencephalogram (eeg) recognition methods based on two-dimentional time-frequency image depth convolutional neural networks | |
| CN108280414A (en) | A kind of recognition methods of the Mental imagery EEG signals based on energy feature | |
| CN107157477A (en) | EEG signals Feature Recognition System and method | |
| CN109726751A (en) | Method based on depth convolutional neural networks identification brain Electrical imaging figure | |
| CN107361766A (en) | A kind of mood EEG signal identification method based on EMD domains multidimensional information | |
| CN110163128A (en) | The Method of EEG signals classification of improved EMD algorithm combination wavelet package transforms and CSP algorithm | |
| CN103425249A (en) | Electroencephalogram signal classifying and recognizing method based on regularized CSP and regularized SRC and electroencephalogram signal remote control system | |
| CN108573207A (en) | EMD and CSP merges most optimum wavelengths space filtering brain electrical feature extracting method | |
| CN102930284A (en) | Surface electromyogram signal pattern recognition method based on empirical mode decomposition and fractal | |
| CN109858537A (en) | EEG feature extraction method of the improved EEMD in conjunction with CSP | |
| CN108387887A (en) | A kind of mixing noise-reduction method of underwater sound signal | |
| CN109271887A (en) | A kind of composite space filtering and template matching method for the identification of brain power mode | |
| CN102306303A (en) | Electroencephalography signal characteristic extraction method based on small training samples | |
| CN108280464A (en) | The brain electrical feature extracting method of DWT and EMD fusion approximate entropies | |
| CN109685071A (en) | Brain electricity classification method based on the study of common space pattern feature width | |
| CN111310656A (en) | Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis | |
| CN108852348A (en) | The collection point sort method and system of scalp brain electricity | |
| CN111000555A (en) | Training data generation method, automatic recognition model modeling method and automatic recognition method for epilepsia electroencephalogram signals |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180925 |
|
| RJ01 | Rejection of invention patent application after publication |