CN106419909A - Multi-class motion imagination EEG signal classification method based on characteristic recombination and wavelet transformation - Google Patents
Multi-class motion imagination EEG signal classification method based on characteristic recombination and wavelet transformation Download PDFInfo
- Publication number
- CN106419909A CN106419909A CN201610815762.6A CN201610815762A CN106419909A CN 106419909 A CN106419909 A CN 106419909A CN 201610815762 A CN201610815762 A CN 201610815762A CN 106419909 A CN106419909 A CN 106419909A
- Authority
- CN
- China
- Prior art keywords
- signal
- matrix
- tongue
- vector
- eeg signals
- 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
-
- 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
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/25—Bioelectric electrodes therefor
- A61B5/279—Bioelectric electrodes therefor specially adapted for particular uses
- A61B5/291—Bioelectric electrodes therefor specially adapted for particular uses for 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
- A61B5/7235—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
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Psychiatry (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Signal Processing (AREA)
- Psychology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a multi-class motion imagination EEG signal classification method based on characteristic recombination and wavelet transformation, mainly solving the problem of less classes and low classification accuracy of present technologies for classifying EEG signals. The method comprise following steps: 1) collecting motion imagination EEG signals and obtaining a training set and a test set; 2) training a two-grade classifier through characteristic combination, wavelet transformation and common space pattern algorithms; 3) extracting test characteristic classification vectors of the test set according to a method corresponding to step 2); 4) by means of the trained classifier, performing signal classification on test signals through the characteristic vectors of the test set to obtain the classifications of EEG signals of imagination left hand motion, imagination right hand motion, imagination feet motion and imagination tongue motion of the test signals. The method of the invention realizes classification on multi-class motion imagination signals and increases classification accuracy, and can be used in intelligent product control of on-line system containing motion imagination brain-computer interface BCI.
Description
Technical field
The invention belongs to areas of information technology, it is related to this four type games of left hand, the right hand, foot and tongue and imagines EEG signals
Sorting technique, can be used for handicap people medical treatment and life auxiliary, such as intelligent wheel chair, and the control of mechanical extremity is it can also be used to existing
For "smart" products, such as flying vehicles control, intelligent automobile driving etc. has the brain of Mental imagery brain-computer interface BCI on-line system
The electric control of product.
Background technology
Modern neuro is thought biology, and cerebral cortex, according to different features and function, can be divided into several areas, different
The different parts that body was administered and regulated and controled in region realize difference in functionality, and each position of body is realized each function and had in cerebral cortex
Corresponding relation, such as precentral gyruss mainly manage whole body skeleton motion, referred to as motor region;Postcentral gyruss mainly manage systemic somatic
Sensation, referred to as sensory region;The major contributing lattice in frontal lobe area, Control emotion and distinguish to mistake;Occipital lobe area is mainly used to administer vision;
In addition to the maincenter of some specific functions, most of region is referred to as association region to cerebral cortex, and they receive multichannel sensation
Information, the neural activity of each functional areas of Correspondent.According to locus, mainly there is body in wherein related to people's limb motion region
Body motor region and somathetic area, we call motor sensory area these regions.
Mental imagery, refers to that experimenter imagines that corresponding actions are made and actual limbs portion in certain position of limbs in the brain
Position is not carried out the imagination process of action, the experimenter's imagination both hands applause action such as sat quietly, but both hands are protected in this process
Hold static.Neuroanatomy research shows, the corticocerebral feature connection that can make one during Mental imagery becomes
Change, thus leading to Energy distribution in brain to change, the change of this energy and actual carry out basic during corresponding limb action
It is consistent.It is exactly specifically, when imagining left hand motion or actual execution left hand action, the motor sensory area of brain offside
The mu rhythm and pace of moving things (8-12Hz) EEG signals and the beta rhythm and pace of moving things (18-24Hz) EEG signals energy dropoff, this phenomenon is referred to as event
Correlation desynchronizes phenomenon ERD;Meanwhile, in ipsilateral movement sensory region mu rhythm and pace of moving things EEG signals and the beta rhythm and pace of moving things brain electricity of brain
Signal energy strengthens, and this phenomenon is referred to as event-related design phenomenon ERS;When imagination right hand motion or the actual execution right hand move
When making, also in corresponding area, ERD and ERS phenomenon can occur, therefore ERD and ERS phenomenon is to differentiate right-hand man's Mental imagery brain at present
The most basic feature of the signal of telecommunication.And imagine that the athletic meeting of foot produces ERD phenomenon in the top area of central authorities of brain, and in motor sensory area
Produce ERS phenomenon;Think that languet motion then can all produce ERS phenomenon in the top area of central authorities of brain and motor sensory area.
Cospace pattern CSP algorithm is the method that Mental imagery EEG signals feature is commonly used of extracting at present, and this algorithm utilizes
On algebraically, matrix simultaneous diagonalization is theoretical, the spatial filter finding a specific direction signal is filtered process so that
In signal after after filtering, the variance of a class reaches maximum, and simultaneously another kind of variance reaches minimum, thus reaching extraction feature
Purpose.But this cospace pattern CSP algorithm has the following disadvantages:1. can only be classified for two class EEG signals,
Cannot be classified using traditional C/S P method for four class signals;2.CSP algorithm characteristics extraction process combines all leading
Dependency, each lead signals is not made after analysis comprehensive, people's EEG signals individual diversity is not larger in addition, so passing
CSP is relatively low to some experimenter's accuracys rate for system.
Content of the invention
Present invention aims to the deficiency of above-mentioned prior art, a kind of feature based restructuring and wavelet transformation are proposed
Mental imagery Method of EEG signals classification, to improve kind number and the classification accuracy rate of classification.
The technical scheme is that be achieved in that:
One. know-why
During according to carrying out this four type games of left hand, the right hand, both feet and tongue imagination, the EEG signals table in sensation of movement region
Reveal ERD and ERS phenomenon, left hand and right hand Mental imagery EEG signals can be carried out combinations of features, as a class signal, will be double
Foot and tongue movements imagination EEG signals carry out combinations of features, as another kind of signal, then two class signals after combination are entered
Row classification;Good time and frequency domain resolution are had according to orthogonal wavelet transformation, with the Mental imagery EEG signals mu rhythm and pace of moving things and
The beta rhythm and pace of moving things has the characteristic of lower frequency, using classical Mallat small echo QMF compression algorithm, the brain electricity to each passage
Signal is decomposed and is reconstructed, and then reuses classical CSP algorithm and carries out characteristic of division extraction, has thus taken into full account entirety
Energy and each channel energy relation, can lift the suitability of classification.
Two. technical scheme
According to above-mentioned principle, technical scheme includes as follows:
(1) EEG signals are obtained:By the left electrode C3 on the electrode cap of subject wears, right electrode C4 and middle electrode CZ
And these three electrodes each 22 electrodes around, with 256HZ sample rate fsCollection experimenter is on imagination left hand, the right side respectively
The EEG signals of multigroup experiment when handss, both feet and the tongue four type games imagination, and the original EEG signals warp successively by collection
Cross amplification, analog/digital conversion, after low-pass filtering, obtain imagining left chirokinesthetic EEG signals El, imagine right chirokinesthetic brain telecommunications
Number Er, EEG signals E of imagination both feet motionfSignal E with imagination tongue movementst;
(2) obtain in (1) four class EEG signals mean random are divided into training set T1 and test set T2, in wherein T1
All include four described class EEG signals El、Er、EfAnd Et, the four class EEG signals Uniform Name comprising in T2 are Ex;
(3) to four class EEG signals in training set T1, according to the row of matrix, ordered arrangement carries out feature group from top to bottom
Close, obtain two groups of signal X after feature restructuring1, X2:
(4) to two groups of signal X after combinations of features1, X2Carry out the tower wavelet decomposition of Mallat and reconstruct respectively, obtain
One group of reconstruction signal X1' and second group of reconstruction signal X'2;
(5) by two groups of reconstruction signal X1', X'2As the input signal of cospace pattern CSP algorithm, obtain recombination signal X1
And X2Corresponding first projection matrix W1And this two groups of reconstruction signals corresponding combinations of features vector F respectivelyX1And FX2, and by this two
Individual combinations of features vector FX1And FX2, it is input in the first grader SVM1, support vector machine classifier SVM1 is entered with line parameter instruction
Practice;
(6) to the imagination left hand motor message E in training set T1lWith imagination right hand motor message ErCarry out and (4) identical
The tower wavelet decomposition of Mallat and reconstructed operation, obtain left hand reconstruction signal El' and right hand reconstruction signal Er';
(7) by left hand reconstruction signal El' and right hand reconstruction signal E'rAs the input signal of cospace pattern CSP algorithm,
Obtain right-hand man reconstruction signal El' and E'rCorresponding second projection matrix W2And this two recombination signals corresponding left hand respectively
Characteristic vector FlWith right hand characteristic vector Fr, and by this two characteristic vectors Fl, FrBe input in the second grader SVM2, to
Hold vector machine classifier SVM2 and carry out parameter training;
(8) to the imagination both feet motor message E in training set T1fWith imagination tongue movements signal EtCarry out and (4) identical
The tower wavelet decomposition of Mallat and reconstructed operation, obtain double-legged reconstruction signal E'fWith tongue reconstruction signal Et';
(9) by double-legged reconstruction signal E'fWith tongue reconstruction signal Et' as cospace pattern CSP algorithm input signal,
Obtain double-legged reconstruction signal E'fWith tongue reconstruction signal Et' corresponding 3rd projection matrix W3And this two recombination signals are respectively
Corresponding left hand characteristic vector FfWith right hand characteristic vector Ft;And by this two characteristic vectors Ff, FtIt is input to the 3rd grader
In SVM3, parameter training is carried out to support vector machine classifier SVM3;
(10) to four class EEG signals in test set T2, feature reinforcement is carried out by the duplication of its data section, obtains
Feature strengthens signal ex:
(11) this feature is strengthened signal exCarry out and the tower wavelet decomposition of (4) identical Mallat and reconstructed operation, obtain
To test set reconstruction signal e'x;
(12) by test set reconstruction signal e'xWith the first projection matrix W1Convolution, extracts testing feature vector f of signalx,
And this testing feature vector is input to the first grader SVM1 train in (5) is classified, identify exBe with regard to left,
Brain electrical test data E of the right handr_lBrain electrical test data E again with respect to both feet, tonguef_t;
(13) will identify in (12) with regard to left hand and right hand brain electrical test data Er_lWith the second projection matrix W2Convolution, carries
Take characteristic vector fl_r, and by this feature vector fl_rIt is input to the second grader SVM2 train in (7) to be classified, identification
Go out exIt is belonging to left hand EEG signals ElOr right hand EEG signals Er;
(14) by brain electrical test data E with regard to foot, tongue of identification in (12)f_t3rd projection matrix W3Convolution, extracts
Characteristic vector ff_t, and by characteristic vector ff_tIt is input to the 3rd grader SVM3 train in (9) to be classified, identify ex
It is belonging to double-legged EEG signals EfOr tongue EEG signals Et.
The present invention compared with prior art has the advantage that:
1. the present invention passes through left hand, right hand EEG signals, and both feet, the combinations of features of tongue EEG signals are it is achieved that to many
Type games imagine the classification of EEG signals;
2., invention introduces the tower wavelet transformation of Mallat is processed to EEG signals, improve dividing of EEG signals
Class accuracy rate, extends the suitability of algorithm.
Brief description
Fig. 1 is the flowchart of the present invention;
The Mallat small echo QMF compression structure that Fig. 2 uses for the present invention;
Fig. 3 is the EEG signals distribution of electrodes schematic diagram of collection in the present invention;
Fig. 4 is signals collecting sequential chart in the present invention;
Fig. 5 is the sub-process figure solving projection matrix in the present invention using CSP method;
Fig. 6 is the sub-process figure in the present invention, test set signal classified.
Specific embodiment
Below in conjunction with the accompanying drawings the present invention is described in further detail.
With reference to Fig. 1, the present invention is implemented as follows:
Step 1. obtains EEG signals.
(1a) 22 electrodes of brain wave acquisition equipment are installed, and signal sampling frequencies are set for 256Hz:
(1a1) subject wears' electrode cap, installs left electrode C3, the middle electrode Cz of electrode cap according to Fig. 3 distribution of electrodes figure
19 electrodes with right electrode C4 and these three surrounding them;
(1a2) sample frequency of setting brain wave acquisition equipment is 256Hz, carries out during Mental imagery for gathering experimenter
EEG signals;
(1b) experimenter is sitting in the display looking squarely front on chair apart from its 1m, according to shown in signals collecting sequential Fig. 4
Sequential carry out Mental imagery test, by brain wave acquisition equipment obtain four type games imagine EEG signals:
(1b1) between the 0th second to the 2nd second starting, screen display black, experimenter rests;When starting within the 2nd second,
Main frame sends the prompt tone of Bee, simultaneously center Screen occur white crosses fork prompting experimenter ready, white crosses fork hold
The continuous display time is 1 second, and the prompt tone persistent period of Bee is 200 milliseconds;
(1b2) the 3rd second starts, spider disappear, center Screen occur at random white to upward arrow, down arrow,
One of arrow and right-hand arrow to the left, the wherein experimenter of arrow prompting to the left carry out left hand Mental imagery, and right-hand arrow carries
Show that experimenter carries out right hand Mental imagery, prompting experimenter carries out the tongue movements imagination to upward arrow, down arrow prompting is tested
Person's both feet Mental imagery;The arrow persistent period is 4 seconds, and when the 7th second starts, arrow disappears, and represents this off-test, experimenter
Stop motion is imagined, enters resting state.
(1b3) each type games imagination carries out 40 experiments, and the four type games imaginations carry out 160 experiments, altogether by electricity
Polar cap collects 160 groups of data of the original EEG signals of experimenter, by this original EEG signals of four classes sequentially pass through amplification, mould/
After number conversion, low-pass filtering, obtain imagining left chirokinesthetic EEG signals El, imagine right chirokinesthetic EEG signals Er, the imagination pair
EEG signals E of foot motionfSignal E with imagination tongue movementst;
The 160 groups of EEG signals mean random obtaining in step 1 are divided into training set T1 and test set T2 by step 2., its
Four described class EEG signals E are all included in middle T1l、Er、EfAnd Et, the four class EEG signals Uniform Name comprising in T2 are Ex;
To four class EEG signals in training set T1, according to the row of matrix, ordered arrangement carries out feature to step 3. from top to bottom
Combination, obtains two groups of signal X after combinations of features1, X2:
Wherein ElAnd ErRepresent imagination left hand, data during right hand motion, E respectivelyfAnd EtImagine both feet respectively, tongue is transported
Data when dynamic.
Step 4. is to two groups of signal X after combinations of features1, X2Carry out the tower wavelet decomposition of Mallat and weight according to Fig. 2 respectively
Structure, obtains first group of reconstruction signal X1' and second group of reconstruction signal X'2:
(4a) select Daubechies function as the basic function of wavelet decomposition process, to first group of letter after combinations of features
Number X1With second group of signal X after feature restructuring2Carry out the 4 layers of decomposition of the tower small echo of Mallat respectively, obtain X15 little wavelength-divisions
Amount A4, D4, D3, D2, D1 and X25 Wavelet Component A4', D4', D3', D2', D1';
(4b) with first group of signal X15 Wavelet Component in the 4th Wavelet Component D4 and the 3rd Wavelet Component D3 to this
First group of signal X1It is reconstructed, obtain first group of reconstruction signal X'1=D4+D3;With in 5 wavelet packet of second group of signal
The 4th Wavelet Component D4' and the 3rd Wavelet Component D3' to this second group of signal X2It is reconstructed, obtain second group of reconstruction signal
X2'=D4'+D3'.
Step 5. is by two groups of reconstruction signal X1', X'2As the input signal of cospace pattern CSP algorithm, obtain restructuring letter
Number X1And X2Corresponding first projection matrix W1;
(5a) according to shown in Fig. 5, calculating first group of reconstruction signal X respectively1' left hand and right hand average spatial covariance matrix
Rl_rWith second group of reconstruction signal X'2Both feet, tongue average spatial covariance matrix Rf_t:
Wherein,Represent first group of reconstruction signal X respectively1' and second group of reconstruction signal X'2Transposition,WithRepresenting matrix respectivelyAnd matrixMark, N1Represent first group of reconstruction signal X1'
Total replicated experimental unitses, N in training set T12Represent second group of reconstruction signal X2' total in training set T1 repeat test time
Number;
(5b) left hand and right hand average spatial covariance matrix R step (5a) being calculatedl_r, both feet, tongue mean space
Covariance matrix Rf_tSummation, obtains total mixing average covariance matrices Rc:
Rc=Rl_r+Rf_t,
(5c) to total mixing average covariance matrices RcCarry out following Eigenvalues Decomposition:
Rc=U λ UT,
Wherein, U represents mixing average covariance matrices RcEigenvectors matrix after decomposition, UTRepresent eigenvectors matrix U
Transposition, λ represent mixing average covariance matrices RcEigenvalue diagonal matrix after decomposition;
(5d) according to mixing average covariance matrices RcFeature value vector matrix U after decomposition and eigenvalue diagonal matrix λ,
Calculate whitening matrix P:
(5e) with the whitening matrix P that obtains in step (4d) respectively to left hand and right hand average spatial covariance matrix Rl_rWith double
Foot, tongue average spatial covariance matrix Rf_tCarry out albefaction, calculate left hand and right hand albefaction covariance matrix Sl_rWith both feet, tongue
Albefaction covariance matrix Sf_t:
Sl_r=PRl_rPT
Sf_t=PRf_tPT
Wherein, PTRepresent the transposed matrix of whitening matrix P;
(5f) to left hand and right hand albefaction covariance matrix Sl_rWith both feet, tongue albefaction covariance matrix Sf_tDivided as follows
Solution:
Sl_r=Usλl_rUs T
Sf_t=Usλf_tUs T
λl_r+λf_t=E
Wherein, UsIt is albefaction characteristic vector, λl_rIt is left hand and right hand albefaction covariance matrix Sl_rAlbefaction eigenvalue after decomposition
Diagonal matrix, λf_tIt is both feet, tongue albefaction covariance matrix Sf_tAlbefaction eigenvalue diagonal matrix after decomposition, E represents unit
Battle array;
(5g) albefaction characteristic vector U being obtained according to step (5f)sThe whitening matrix P obtaining with step (5d), calculates the
One projection matrix W1:
W1=Us TP
Wherein, Us TIt is albefaction characteristic vector UsTransposed matrix;
Step 6. calculates the first recombination signal X1' corresponding left hand and right hand tagsort vector FX1With the second recombination signal X'2
Corresponding both feet, tongue tagsort vector FX2, and by this two special class vector FX1And FX2, it is input to the first grader
In SVM1, parameter training is carried out to support vector machine classifier SVM1.
(6a) the first projection matrix W being obtained using step (5g)1To the first recombination signal X1' and the second recombination signal X'2
Carry out convolution, obtain left hand and right hand convolution eigenmatrix Zl_fWith both feet, tongue convolution eigenmatrix Zf_t:
Zl_r=W1X1′
Zf_t=W1X2′;
(6b) from left hand and right hand convolution eigenmatrix Zl_rIn, extract the tagsort vector F with regard to left hand and right handX1:
Wherein, vjRepresent tagsort vector FX1J-th vector element, var (Zl_r) represent special to left hand and right hand convolution
Levy matrix Zl_rCarry out variance computing,Represent to left hand and right hand convolution eigenmatrix Zl_rJth row carry out variance computing;
m1Represent and extract left hand and right hand convolution eigenmatrix Zl_rBefore and after line number, value be m1=6;
(6c) from both feet, tongue convolution eigenmatrix Zf_tIn, extract the tagsort vector F with regard to both feet, tongueX2:
Wherein, yjRepresent tagsort vector FX1J-th people's vector element, var (Zf_t) represent to both feet, tongue volume
Long-pending eigenmatrix Zf_tCarry out variance computing,Represent to both feet, tongue convolution eigenmatrix Zf_tThe jth row side of carrying out
Difference operation;m2Represent and extract both feet, tongue convolution eigenmatrix Zf_tBefore and after line number, value m2=6.
(6d) the tagsort vector F with regard to left hand and right hand that (6b), (6c) are obtainedX1With the feature with regard to both feet, tongue
Class vector FX2As the input of grader SVM1, the core letter of Training Support Vector Machines grader, wherein support vector machines 1
Number is Radial basis kernel function, obtains optimum kernel function penalty coefficient p and kernel function radius g using grid data service.
Step 7. is to the imagination left hand motor message E in training set T1lWith imagination right hand motor message ErCarry out with step 4
The tower wavelet decomposition of identical Mallat and reconstructed operation, obtain left hand reconstruction signal El' and right hand reconstruction signal E ' r.
Step 8. calculates the second projection matrix W2, train the second grader SVM2.
(8a) the left hand reconstruction signal E that step 7 is obtainedl' and right hand reconstruction signal Er' as cospace pattern CSP algorithm
Input signal, with obtaining left hand reconstruction signal E with step 5 identical methodl' and right hand reconstruction signal Er' corresponding second throwing
Shadow matrix W2;
(8b) use the second projection matrix W2, left hand reconstruction signal El' and right hand reconstruction signal Er', according to same with step 6
The method of sample calculates left hand reconstruction signal El' corresponding left hand tagsort vector FlWith right hand reconstruction signal Er' the corresponding right hand
Tagsort vector Fr, and by this two tagsorts vector Fl, Fr, it is input in the second grader SVM2, to support vector machine
Grader SVM2 carries out parameter training.
Step 9. is to the imagination both feet motor message E in training set T1fWith imagination tongue movements signal EtCarry out with step 4
The tower wavelet decomposition of identical Mallat and reconstructed operation, obtain double-legged reconstruction signal E'fWith tongue reconstruction signal E 't.
Step 10. calculates the second projection matrix W3, train the 3rd grader SVM3.
(8a) the double-legged reconstruction signal E' that step 9 is obtainedfWith tongue reconstruction signal Et' as cospace pattern CSP algorithm
Input signal, with obtaining double-legged reconstruction signal E' with step 5 identical calculating processfWith tongue reconstruction signal Et' corresponding
Three projection matrix W3;
(8b) use the 3rd projection matrix W3, double-legged reconstruction signal E'fWith tongue reconstruction signal Et', according to same with step 6
The method of sample calculates double-legged reconstruction signal E'fCorresponding both feet tagsort vector FfWith tongue reconstruction signal Et' corresponding tongue
Tagsort vector Ft, and by this two tagsorts vector FfAnd Ft, it is input in the 3rd grader SVM3, to supporting vector
Machine grader SVM3 carries out parameter training;
Step 11. is to four class EEG signals E in test set T2x, feature reinforcement is carried out by the duplication of its data section,
Obtain feature and strengthen signal ex:
And this feature is strengthened signal exCarry out tower wavelet decomposition and reconstructed operation with step 4 identical Mallat, obtain
To test set reconstruction signal e'x.
Step 12. is according to as shown in fig. 6, by test set reconstruction signal e'xWith the first projection matrix W1Convolution, according to step
Rapid 6 identical methods, extract test set reconstruction signal e'xTest set first order tagsort vector fx, and this feature is classified
Vector fxIt is input to the first grader SVM1 train in (6d) to be classified, identify exIt is the brain electrical measurement with regard to left hand and right hand
Examination data Er_lBrain electrical test data E again with respect to both feet, tonguef_t.
Step 13. is according to as shown in fig. 6, will identify in (12) with regard to left hand and right hand brain electrical test data Er_lWith second
Projection matrix W2Convolution, according to step 6 identical method, extract test set second level tagsort vector fl_r, and by this spy
Levy class vector fl_rIt is input to the second grader SVM2 trained in (8b) to be classified, identify exIt is belonging to left hand
EEG signals ElOr right hand EEG signals Er;
Step 14. is according to as shown in fig. 6, brain electrical test data E with regard to foot, tongue by identification in step (12)f_tWith
3rd projection matrix W3Convolution, according to step 6 identical method, extract test set second level tagsort vector ff_t, and will
This feature class vector ff_tIt is input to the 3rd grader SVM3 trained in (9) to be classified, identify exIt is belonging to double
Foot EEG signals EfOr tongue EEG signals Et.
Can be to test signal E by above stepxClassified, identified test signal ExIt is belonging to following four classes signal
In which species concrete signal:Imagine left chirokinesthetic EEG signals, imagine right chirokinesthetic EEG signals, the imagination is double
The EEG signals of foot motion and the signal of imagination tongue movements.
In above method, the four type games imagination EEG signals first passing through training set train three specific directions
Projection matrix, the first projection matrix W1, the second projection matrix W2, the 3rd projection matrix W3With three support vector machine classifiers,
One grader SVM1, the second grader SVM2, the 3rd grader SVM3;Then by three specific directions training above
Projection matrix and three graders are classified to the signal in test set, identify that test signal belongs to the four type games imaginations
Which kind of.This makes it possible to the Mental imagery EEG signals by identifying people, realize to such as wheelchair, the smart machine such as mechanical arm
Control.
Claims (3)
1. the multiclass Mental imagery Method of EEG signals classification of feature restructuring and wavelet transformation, including:
(1) EEG signals are obtained:By the left electrode C3 on the electrode cap of subject wears, right electrode C4 and middle electrode CZ and
These three electrodes each 22 electrodes around, with 256HZ sample rate fsCollection experimenter is in the imagination left hand, right hand, double respectively
The EEG signals of multigroup experiment when foot and the tongue four type games imagination, and the original EEG signals of collection are sequentially passed through and put
Greatly, analog/digital conversion, after low-pass filtering, obtain imagining left chirokinesthetic EEG signals El, imagine right chirokinesthetic EEG signals Er,
EEG signals E of imagination both feet motionfSignal E with imagination tongue movementst;
(2) obtain in (1) four class EEG signals mean random are divided into training set T1 and test set T2, all wrap in wherein T1
Include four described class EEG signals El、Er、EfAnd Et, the four class EEG signals Uniform Name comprising in T2 are Ex;
(3) to four class EEG signals in training set T1, according to the row of matrix, ordered arrangement carries out combinations of features from top to bottom, obtains
To feature group and after two groups of signal X1, X2:
(4) to two groups of signal X after combinations of features1, X2Carry out the tower wavelet decomposition of Mallat and reconstruct respectively, obtain first group
Reconstruction signal X '1With second group of reconstruction signal X'2;
(5) by two groups of reconstruction signal X '1, X'2As the input signal of cospace pattern CSP algorithm, obtain recombination signal X1And X2
Corresponding first projection matrix W1And this two groups of reconstruction signals corresponding combinations of features vector F respectivelyX1And FX2, and by this two spies
Levy mix vector FX1And FX2, it is input in the first grader SVM1, parameter training is carried out to support vector machine classifier SVM1;
(6) to the imagination left hand motor message E in training set T1lWith imagination right hand motor message ErCarry out and (4) identical
The tower wavelet decomposition of Mallat and reconstructed operation, obtain left hand reconstruction signal E 'lWith right hand reconstruction signal E 'r;
(7) by left hand reconstruction signal E 'lWith right hand reconstruction signal E'rAs the input signal of cospace pattern CSP algorithm, obtain
Right-hand man reconstruction signal E 'lAnd E'rCorresponding second projection matrix W2And this two recombination signals corresponding left hand feature respectively
Vectorial FlWith right hand characteristic vector Fr, and by this two characteristic vectors Fl, FrBe input in the second grader SVM2, to support to
Amount machine grader SVM2 carries out parameter training;
(8) to the imagination both feet motor message E in training set T1fWith imagination tongue movements signal EtCarry out and (4) identical
The tower wavelet decomposition of Mallat and reconstructed operation, obtain double-legged reconstruction signal E'fWith tongue reconstruction signal E 't;
(9) by double-legged reconstruction signal E'fWith tongue reconstruction signal E 'tAs the input signal of cospace pattern CSP algorithm, obtain
Double-legged reconstruction signal E'fWith tongue reconstruction signal E 'tCorresponding 3rd projection matrix W3And this two recombination signals correspond to respectively
Left hand characteristic vector FfWith right hand characteristic vector Ft;And by this two characteristic vectors Ff, FtIt is input to the 3rd grader SVM3
In, parameter training is carried out to support vector machine classifier SVM3;
(10) to four class EEG signals Uniform Name E in test set T2x, feature reinforcement is carried out by the duplication of its data section,
Obtain feature and strengthen signal ex:
(11) this feature is strengthened signal exCarry out and the tower wavelet decomposition of (4) identical Mallat and reconstructed operation, tested
Collection reconstruction signal e'x;
(12) by test set reconstruction signal e'xWith the first projection matrix W1Convolution, extracts reconstruction signal e'xThe test set first order
Tagsort vector fx, and this testing feature vector is input to the first grader SVM1 train in (5) and is classified, know
Do not go out exIt is brain electrical test data E with regard to left hand and right handr_lBrain electrical test data E again with respect to both feet, tonguef_t;
(13) will identify in (12) with regard to left hand and right hand brain electrical test data Er_lWith the second projection matrix W2Convolution, extracts and surveys
Examination collection second level tagsort vector fl_r, and by this feature vector fl_rIt is input to the second grader SVM2 train in (7)
Classified, identified exIt is belonging to left hand EEG signals ElOr right hand EEG signals Er;
(14) by brain electrical test data E with regard to foot, tongue of identification in (12)f_t3rd projection matrix W3Convolution, extracts test
Collection second level tagsort vector ff_t, and by characteristic vector ff_tIt is input to the 3rd grader SVM3 train in (9) to carry out
Classification, identifies exIt is belonging to double-legged EEG signals EfOr tongue EEG signals Et.
2. to two groups of signal X after combinations of features in method according to claim 1, wherein step (4)1, X2Carry out small echo
Decompose and reconstruct, carry out as follows:
(4a) select Daubechies function as the basic function of wavelet decomposition process, to first group of signal X after combinations of features1
With second group of signal X after feature restructuring2Carry out the 4 layers of decomposition of the tower small echo of Mallat respectively, obtain X15 Wavelet Component A4,
D4, D3, D2, D1 and X25 Wavelet Component A4', D4', D3', D2', D1';
(4b) with first group of signal X15 Wavelet Component in the 4th Wavelet Component D4 and the 3rd Wavelet Component D3 to this first
Group signal X1It is reconstructed, obtain first group of reconstruction signal X'1=D4+D3;With in 5 wavelet packet of second group of signal
Four Wavelet Component D4' and the 3rd Wavelet Component D3' are to this second group of signal X2It is reconstructed, obtain second group of reconstruction signal X2'=
D4'+D3'.
3. the multiclass Mental imagery Method of EEG signals classification of feature restructuring according to claim 1 and wavelet transformation, its
It is characterised by:Recombination signal X is obtained in described step (5)1And X2Corresponding first projection matrix W1And two groups of reconstruction signals are respectively
Corresponding combinations of features vector FX1And FX2, carry out as follows:
(5a) calculate first group of reconstruction signal X ' respectively1Left hand and right hand average spatial covariance matrix Rl_rWith second group of reconstruct letter
Number X'2Both feet, tongue average spatial covariance matrix Rf_t:
Wherein,Represent first group of reconstruction signal X ' respectively1With second group of reconstruction signal X'2Transposition,WithRepresenting matrix respectivelyAnd matrixMark, N1Represent first group of reconstruction signal X1' in training set T1
Total replicated experimental unitses, N2Represent second group of reconstruction signal X2' replicated experimental unitses total in training set T1;
(5b) left hand and right hand average spatial covariance matrix R step (5a) being calculatedl_r, both feet, tongue mean space association side
Difference matrix Rf_tSummation, obtains total mixing average covariance matrices Rc:
Rc=Rl_r+Rf_t,
(5c) to total mixing average covariance matrices RcCarry out following Eigenvalues Decomposition:
Rc=U λ UT,
Wherein, U represents mixing average covariance matrices RcEigenvectors matrix after decomposition, UTRepresent characteristic vector
The transposition of matrix U, λ represents mixing average covariance matrices RcEigenvalue diagonal matrix after decomposition;
(5d) according to mixing average covariance matrices RcFeature value vector matrix U after decomposition and eigenvalue diagonal matrix λ, calculate
Whitening matrix P:
(5e) with the whitening matrix P that obtains in step (4d) respectively to left hand and right hand average spatial covariance matrix Rl_rWith both feet,
Tongue average spatial covariance matrix Rf_tCarry out albefaction, calculate left hand and right hand albefaction covariance matrix Sl_rWith both feet, tongue albefaction
Covariance matrix Sf_t:
Sl_r=PRl_rPT
Sf_t=PRf_tPT
Wherein, PTRepresent the transposed matrix of whitening matrix P;
(5f) to left hand and right hand albefaction covariance matrix Sl_rWith both feet, tongue albefaction covariance matrix Sf_tDecomposed as follows:
Sl_r=Usλl_rUs T
Sf_t=Usλf_tUs T
λl_r+λf_t=E
Wherein, UsIt is albefaction characteristic vector, λl_rIt is left hand and right hand albefaction covariance matrix Sl_rAlbefaction eigenvalue after decomposition is diagonal
Matrix, λf_tIt is both feet, tongue albefaction covariance matrix Sf_tAlbefaction eigenvalue diagonal matrix after decomposition, E represents unit matrix;
(5g) albefaction characteristic vector U being obtained according to step (5f)sThe whitening matrix P obtaining with step (5d), calculates the first projection
Matrix W1:
W1=Us TP
Wherein, Us TIt is albefaction characteristic vector UsTransposed matrix;
(5h) the first projection matrix W being obtained using step (5g)1The first recombination signal X ' to training set T11With the second restructuring
Signal X'2Carry out convolution, obtain left hand and right hand convolution eigenmatrix Zl_fWith both feet, tongue convolution eigenmatrix Zf_t:
Zl_r=W1X1′
Zf_t=W1X2′;
(5i) from left hand and right hand convolution eigenmatrix Zl_rIn, extract the tagsort vector F with regard to left hand and right handX1:
Wherein, vjRepresent tagsort vector FX1J-th vector element, var (Zl_r) represent to left hand and right hand convolution feature square
Battle array Zl_rCarry out variance computing,Represent to left hand and right hand convolution eigenmatrix Zl_rJth row carry out variance computing;m1Table
Show extraction left hand and right hand convolution eigenmatrix Zl_rBefore and after line number, value be m1=6;
(5j) from both feet, tongue convolution eigenmatrix Zf_tIn, extract the tagsort vector F with regard to both feet, tongueX2:
Wherein, yjRepresent tagsort vector FX1J-th people's vector element, var (Zf_t) represent special to both feet, tongue convolution
Levy matrix Zf_tCarry out variance computing,Represent to both feet, tongue convolution eigenmatrix Zf_tJth row carry out variance fortune
Calculate;m2Represent and extract both feet, tongue convolution eigenmatrix Zf_tBefore and after line number, value m2=6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610815762.6A CN106419909A (en) | 2016-09-12 | 2016-09-12 | Multi-class motion imagination EEG signal classification method based on characteristic recombination and wavelet transformation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610815762.6A CN106419909A (en) | 2016-09-12 | 2016-09-12 | Multi-class motion imagination EEG signal classification method based on characteristic recombination and wavelet transformation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106419909A true CN106419909A (en) | 2017-02-22 |
Family
ID=58168932
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610815762.6A Pending CN106419909A (en) | 2016-09-12 | 2016-09-12 | Multi-class motion imagination EEG signal classification method based on characteristic recombination and wavelet transformation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106419909A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106943140A (en) * | 2017-03-08 | 2017-07-14 | 重庆邮电大学 | A kind of Mental imagery EEG feature extraction method based on RandomSelect RCSP |
CN106963370A (en) * | 2017-03-27 | 2017-07-21 | 广州视源电子科技股份有限公司 | A kind of electric allowance recognition methods of the brain based on SVMs and device |
CN107168313A (en) * | 2017-05-17 | 2017-09-15 | 北京汽车集团有限公司 | Control the method and device of vehicle drive |
CN107169462A (en) * | 2017-05-19 | 2017-09-15 | 山东建筑大学 | A kind of two sorting techniques of the EEG signals tagsort based on step analysis |
CN107316532A (en) * | 2017-05-27 | 2017-11-03 | 西南交通大学 | The method of testing and system of dispatcher's inferential capability |
CN108960299A (en) * | 2018-06-15 | 2018-12-07 | 东华大学 | A kind of recognition methods of multiclass Mental imagery EEG signals |
CN109078262A (en) * | 2018-08-15 | 2018-12-25 | 北京机械设备研究所 | A kind of MI-BCI training method based on peripheral nerve electro photoluminescence |
CN109308458A (en) * | 2018-08-31 | 2019-02-05 | 电子科技大学 | A method of small target deteection precision is promoted based on characteristic spectrum change of scale |
CN111931646A (en) * | 2020-08-10 | 2020-11-13 | 华东理工大学 | Motor imagery electroencephalogram signal identification system based on empirical mode decomposition algorithm |
CN112232177A (en) * | 2020-10-13 | 2021-01-15 | 燕山大学 | CSP-UMAP-based motor imagery electroencephalogram data dimension reduction method and experimental device thereof |
CN112270255A (en) * | 2020-10-27 | 2021-01-26 | 广州大学 | Electroencephalogram signal identification method and device, electronic equipment and storage medium |
CN116994697A (en) * | 2023-08-04 | 2023-11-03 | 首都医科大学宣武医院 | Brain-computer interaction method based on complete spinal cord injury patient training evaluation |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101558997A (en) * | 2009-05-27 | 2009-10-21 | 江西蓝天学院 | Electroencephalograph signal identification method based on second-order blind identification |
CN104571505A (en) * | 2014-12-24 | 2015-04-29 | 天津大学 | Brain-machine interface method based on sequence composite limb imaginary movement |
CN104814734A (en) * | 2015-05-13 | 2015-08-05 | 西安电子科技大学 | CSP (Common Spatial Patterns) and cross-correlation based motor imagery electroencephalogram classification method |
-
2016
- 2016-09-12 CN CN201610815762.6A patent/CN106419909A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101558997A (en) * | 2009-05-27 | 2009-10-21 | 江西蓝天学院 | Electroencephalograph signal identification method based on second-order blind identification |
CN104571505A (en) * | 2014-12-24 | 2015-04-29 | 天津大学 | Brain-machine interface method based on sequence composite limb imaginary movement |
CN104814734A (en) * | 2015-05-13 | 2015-08-05 | 西安电子科技大学 | CSP (Common Spatial Patterns) and cross-correlation based motor imagery electroencephalogram classification method |
Non-Patent Citations (2)
Title |
---|
BAI XIAOPING ET AL: "The Offline Feature Extraction of Four-class Motor Imagery EEG Based on ICA and Wavelet-CSP", 《PROCEEDINGS OF THE 33RD CHINESE CONTROL CONFERENCE》 * |
张焕等: "基于小波-共空间模式的多类想象运动诱发脑电特征提取", 《测试技术学报》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106943140A (en) * | 2017-03-08 | 2017-07-14 | 重庆邮电大学 | A kind of Mental imagery EEG feature extraction method based on RandomSelect RCSP |
CN106963370A (en) * | 2017-03-27 | 2017-07-21 | 广州视源电子科技股份有限公司 | A kind of electric allowance recognition methods of the brain based on SVMs and device |
CN107168313A (en) * | 2017-05-17 | 2017-09-15 | 北京汽车集团有限公司 | Control the method and device of vehicle drive |
CN107169462A (en) * | 2017-05-19 | 2017-09-15 | 山东建筑大学 | A kind of two sorting techniques of the EEG signals tagsort based on step analysis |
CN107316532A (en) * | 2017-05-27 | 2017-11-03 | 西南交通大学 | The method of testing and system of dispatcher's inferential capability |
CN108960299B (en) * | 2018-06-15 | 2021-05-14 | 东华大学 | Method for identifying multi-class motor imagery electroencephalogram signals |
CN108960299A (en) * | 2018-06-15 | 2018-12-07 | 东华大学 | A kind of recognition methods of multiclass Mental imagery EEG signals |
CN109078262A (en) * | 2018-08-15 | 2018-12-25 | 北京机械设备研究所 | A kind of MI-BCI training method based on peripheral nerve electro photoluminescence |
CN109078262B (en) * | 2018-08-15 | 2022-11-01 | 北京机械设备研究所 | MI-BCI training method based on peripheral nerve electrical stimulation |
CN109308458B (en) * | 2018-08-31 | 2022-03-15 | 电子科技大学 | Method for improving small target detection precision based on characteristic spectrum scale transformation |
CN109308458A (en) * | 2018-08-31 | 2019-02-05 | 电子科技大学 | A method of small target deteection precision is promoted based on characteristic spectrum change of scale |
CN111931646A (en) * | 2020-08-10 | 2020-11-13 | 华东理工大学 | Motor imagery electroencephalogram signal identification system based on empirical mode decomposition algorithm |
CN112232177A (en) * | 2020-10-13 | 2021-01-15 | 燕山大学 | CSP-UMAP-based motor imagery electroencephalogram data dimension reduction method and experimental device thereof |
CN112270255A (en) * | 2020-10-27 | 2021-01-26 | 广州大学 | Electroencephalogram signal identification method and device, electronic equipment and storage medium |
CN112270255B (en) * | 2020-10-27 | 2023-09-26 | 广州大学 | Electroencephalogram signal identification method and device, electronic equipment and storage medium |
CN116994697A (en) * | 2023-08-04 | 2023-11-03 | 首都医科大学宣武医院 | Brain-computer interaction method based on complete spinal cord injury patient training evaluation |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106419909A (en) | Multi-class motion imagination EEG signal classification method based on characteristic recombination and wavelet transformation | |
CN104091172B (en) | A kind of feature extracting method of Mental imagery EEG signals | |
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 | |
CN107550491B (en) | A kind of multi-class Mental imagery classifying identification method | |
CN107168524A (en) | A kind of Steady State Visual Evoked Potential sorting technique based on deep learning mixed model | |
CN106502405A (en) | Based on the compound limbs Imaginary Movement multi-mode brain-computer interface method and system of synchronicity | |
CN106362287A (en) | Novel MI-SSSEP mixed brain-computer interface method and system thereof | |
CN102200833B (en) | Speller brain-computer interface (SCI) system and control method thereof | |
CN102613972A (en) | Extraction method of characteristics of electroencephalogram signals based on motor imagery | |
CN106529476A (en) | Deep stack network-based electroencephalogram signal feature extraction and classification method | |
CN101488189B (en) | Brain-electrical signal processing method based on isolated component automatic clustering process | |
CN105824418A (en) | Brain-computer interface communication system based on asymmetric visual evoked potential | |
CN104548347A (en) | Pure idea nerve muscle electrical stimulation control and nerve function evaluation system | |
CN111329474A (en) | Electroencephalogram identity recognition method and system based on deep learning and information updating method | |
CN109858537A (en) | EEG feature extraction method of the improved EEMD in conjunction with CSP | |
CN110163128A (en) | The Method of EEG signals classification of improved EMD algorithm combination wavelet package transforms and CSP algorithm | |
Li et al. | Analysis and classification of EEG signals using a hybrid clustering technique | |
CN109247917A (en) | A kind of spatial hearing induces P300 EEG signal identification method and device | |
CN108042132A (en) | Brain electrical feature extracting method based on DWT and EMD fusions CSP | |
CN110263606A (en) | Scalp brain electrical feature based on end-to-end convolutional neural networks extracts classification method | |
CN110013248A (en) | Brain electricity tensor mode identification technology and brain-machine interaction rehabilitation system | |
CN105266804B (en) | A kind of brain-electrical signal processing method based on low-rank and sparse matrix decomposition | |
CN108523883A (en) | A kind of continuous Mental imagery identifying system of left and right index finger based on actual act modeling | |
CN111631908A (en) | Active hand training system and method based on brain-computer interaction and deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170222 |
|
WD01 | Invention patent application deemed withdrawn after publication |