CN113558643A - Multi-feature epilepsy signal classification method based on VMD and NLTWSVM - Google Patents
Multi-feature epilepsy signal classification method based on VMD and NLTWSVM Download PDFInfo
- Publication number
- CN113558643A CN113558643A CN202110771578.7A CN202110771578A CN113558643A CN 113558643 A CN113558643 A CN 113558643A CN 202110771578 A CN202110771578 A CN 202110771578A CN 113558643 A CN113558643 A CN 113558643A
- Authority
- CN
- China
- Prior art keywords
- electroencephalogram
- signal
- feature
- data set
- samples
- 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
Images
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]
- A61B5/372—Analysis of electroencephalograms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/40—Detecting, measuring or recording for evaluating the nervous system
- A61B5/4076—Diagnosing or monitoring particular conditions of the nervous system
- A61B5/4094—Diagnosing or monitoring seizure diseases, e.g. epilepsy
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Neurology (AREA)
- Heart & Thoracic Surgery (AREA)
- Molecular Biology (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Neurosurgery (AREA)
- Medical Informatics (AREA)
- Physics & Mathematics (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Psychology (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention belongs to a multi-feature epilepsy signal classification method based on VMD and NLTWSVM. The method comprises the following steps: preprocessing electroencephalogram signal data; the electroencephalogram signal data comprises an electroencephalogram data set of a healthy person, an epileptic seizure interval electroencephalogram data set and an epileptic seizure period electroencephalogram data set; carrying out variation modal decomposition on the electroencephalogram signals to obtain corresponding intrinsic mode functions; extracting multi-feature parameters from the obtained series of intrinsic mode functions; and inputting the extracted multi-feature parameters serving as feature combinations into a nonlinear dual-sub support vector machine for training and classification. Aiming at the characteristics of non-stability, non-linearity and the like of epilepsia electroencephalogram signals, a variational modal decomposition time-frequency analysis method is utilized to obtain signal components, multi-feature parameter extraction is carried out on the components, and the epilepsia electroencephalogram signals and the non-epilepsia electroencephalogram signals are correctly classified by combining a non-linear Gemini support vector machine learning method.
Description
Technical Field
The invention belongs to time-frequency analysis, pattern classification and machine learning of non-stationary nonlinear signals, belongs to the technical field of biomedical engineering signal processing and pattern recognition, and particularly relates to a multi-feature epilepsy signal classification method based on VMD and NLTWSVM.
Background
Epilepsy is a chronic non-infectious nervous system brain disease of transient cerebral dysfunction caused by sudden hypersynchronous abnormal discharge of cerebral neurons, and according to the statistics of world health organizations, about 5000 ten thousand epilepsy patients exist in the world at present, wherein about 80 percent of the epilepsy patients live in middle and low income countries, and the epilepsy disease has serious influence on the life, study, work and spirit of the epilepsy patients and even jeopardizes the life safety of the epilepsy patients. At present, when a neuroelectrophysiological examination is performed on an epileptic patient, an electroencephalogram (EEG) is used as a 'gold standard' for epileptic diagnosis, electroencephalograms of the patient are recorded through dry electrodes placed at a cerebral cortex, and a doctor uses naked eyes to observe electroencephalogram data to judge whether spikes, sharp waves, spines (spikes) -slow synthetic waves and other paroxysmal rhythmic waves representing cerebral overdischarge appear in the electroencephalogram. However, the judgment is very time-consuming and labor-consuming only by the expert's visual observation, and there may be cases of misdiagnosis caused by human subjective factors in the observation process. Therefore, the classification and identification of the epileptic brain electrical signals and the non-epileptic brain electrical signals are of great significance to the fields of epileptic seizure prediction and clinical diagnosis and treatment.
Electroencephalogram signals are typically nonlinear, non-stationary, biologically weak signals. The analysis method of the electroencephalogram signals comprises time domain, frequency domain, time-frequency domain and nonlinear dynamics. The Empirical Mode Decomposition (EMD) method proposed by Huang et al in the time-frequency domain method is widely applied in the field of electroencephalogram signal analysis, but in the method, because the envelope estimation error is gradually amplified after the envelope of the extreme point is subjected to multiple recursive decompositions, the similar frequency components cannot be correctly separated, and the problems of modal aliasing, end point effect and the like exist, so that part of detailed information of the signal is lost in the processing process, and the integrity of the signal and the accuracy of analysis are influenced. Aiming at the problem of modal aliasing of EMD, Wu and Huang propose an improved Ensemble Empirical Mode Decomposition (EEMD) method using noise to assist EMD, which utilizes the characteristic that white noise has uniform power spectrum density distribution in a frequency domain, eliminates modal aliasing phenomenon and improves the noise immunity of the algorithm, but the method increases the calculated amount and decomposes to obtain a plurality of components exceeding the real composition of signals.
Disclosure of Invention
The invention aims to solve the technical problem of providing a multi-feature epilepsia signal classification method based on VMD and NLTWSVM, aiming at the characteristics of unsteady state, nonlinearity and the like of epilepsia electroencephalogram signals, a variational modal decomposition time-frequency analysis method is utilized to obtain signal components, multi-feature parameter extraction is carried out on the components, and the accurate classification of the epilepsia electroencephalogram signals and the non-epilepsia electroencephalogram signals is realized by combining a nonlinear two-element support vector machine learning method.
The present invention is achieved in such a way that,
a multi-feature epilepsy signal classification method based on VMD and NLTWSVM comprises the following steps:
preprocessing electroencephalogram signal data; the electroencephalogram signal data comprises an electroencephalogram data set of a healthy person, an epileptic seizure interval electroencephalogram data set and an epileptic seizure period electroencephalogram data set;
carrying out variation mode decomposition on the electroencephalogram signal x (t) to obtain a corresponding intrinsic mode function uk(t), K is the representation of a certain number of layers in K, wherein K is the order of the intrinsic mode function;
extracting multi-feature parameters from the obtained series of intrinsic mode functions;
and inputting the extracted multi-feature parameters serving as feature combinations into a nonlinear dual-sub support vector machine for training and classification.
Further, the pre-processing comprises: dividing the first 23s of electroencephalogram data of each group of electroencephalogram signals in the data set into 23 data fragment samples by taking 1s as a unit, randomly selecting 50 groups of electroencephalogram data from each data set, segmenting the electroencephalogram data to obtain data fragment samples as training samples of the classifier, and taking the remaining data fragment samples of each data set as test samples of the classifier.
Further, the performing variational modal decomposition on the electroencephalogram signal x (t) comprises:
establishing a variation model;
introducing a secondary penalty factor alpha and a Lagrangian multiplication operator lambda (t), converting a constraint variation problem into an unconstrained variation problem, and constructing an extended Lagrangian function;
solving the variation problem by using a multiplicative operator alternating direction method and alternately updatingAnd λn+1(t) seeking a 'saddle point' of the extended Lagrangian function, wherein the 'saddle point' is an optimal solution of the variation model;
wherein, establishing the variation model comprises the following steps:
let uk(t) is the IMF component, for each IMF component uk(t) performing Hilbert transform, constructing an analytic signal, and obtaining a single-side frequency spectrum:
mixing analytic signals of each IMF component to estimate center frequencyModulating the spectrum of each IMF component to a corresponding baseband:
calculating the square L of the gradient of the demodulation signal expressed by equation (2)2Norm ofEstimating the signal bandwidth of each IMF component, introducing constraint conditions, and constructing a variation model as shown in formula (3):
in the formula: δ (t) is the impulse function, { u ] is the convolution signk}={u1,u2,…,uKIs each IMF component obtained by decomposition, { omega }k}={ω1,ω2,…,ωK-the center frequency of each IMF component, K is the number of IMF components,
the expression of the extended lagrange function is as follows:
the iterative formula adopted for solving the variational problem by using the multiplicative operator alternating direction method is as follows:
the iteration stop conditions are as follows:
in the formula, tau is Lagrange multiplier updating parameter, epsilon is convergence tolerance.
Further, when extracting multi-feature parameters for the eigenmode function, the electroencephalogram signal x (t) is decomposed by VMD to obtain a signal uk(t) is a time series of length L, the extracted features are solved by the following expression:
maximum value: max (u) Maxk(t)) (9)
Minimum value: min ═ Min (u)k(t)) (10)
further, the step of inputting the extracted multi-feature parameters into a nonlinear dual support vector machine for training and classification includes:
step 5.1: training samples are set, wherein p positive class samples and q negative class samples exist, and the expression is as follows:
T={(x1,+1),(x2,+1),…(xp,+1),(xp+1,-1),(xp+2,-1),…(xp+q,-1)} (17)
wherein xi∈Rn,i=1,2,…,p+q;
Step 5.2: for a nonlinear two-sub support vector machine, the original space linear indivisible samples are mapped into a linearly separable high-dimensional space by introducing a kernel function, and the nonlinear two-sub support vector machine still seeks two non-parallel decision hyperplanes as follows:
wherein C isT=[A;B]T∈Rn×l,A=(x1,x2,…,xp)T∈Rp×n,B=(xp+1,xp+2,…,xp+q)T∈Rq×n,l=p+q;
Step 5.3: the dual sub-support vector machine requires that each hyperplane satisfies the condition that the distance from the corresponding class sample point is minimum and the distance from another class sample point is enough, and accordingly two convex quadratic programming problems are constructed as follows:
and
wherein e+And e-Is a column vector of elements all one, ξ+And xi-Is a relaxation vector, ciI is 1,2 is a penalty parameter;
step 5.4: the dual problems of formula (19) and formula (20) are respectively
And
wherein S ═ K (A, C)T)e+],R=[K(B,CT)e-];
Step 5.5: by solving equations (21) and (22), the solution to the original problem is obtained by solving the following equation:
for a new input x ∈ RnIts category is judged as
Where | is a point x to a hyperplane K (x)T,CT)uk+bkA vertical distance of 0, k-, + is provided.
Compared with the prior art, the invention has the beneficial effects that:
the invention adopts a variational modal decomposition method, can analyze and decompose the multi-scale time frequency of the signal, effectively avoids the problem of modal aliasing, and has the advantages of better complex signal decomposition precision, better noise interference resistance and the like;
the statistics can be used for objectively representing the overall quantitative characteristics and the quantitative relation of the signals, 8 characteristic parameters are extracted from each IMF component by applying a statistical method and a box-type graph is constructed, the characteristic combination mode is effective, the difference between epileptic electroencephalogram signals and non-epileptic electroencephalogram signals can be accurately represented, and the characteristic value can be rapidly calculated when the data volume is large; the nonlinear dual-sub support vector machine is used for training and classifying, and the method has more advantages in classification effect and calculation speed. The biggest difference between the Support Vector Machine (SVM) used by the invention and the traditional SVM method is that the dual-sub SVM constructs a corresponding Support hyperplane for each class, and each hyperplane is close to the training point of the corresponding class as much as possible and is far away from the training point of the other class, without limiting whether the hyperplane is parallel or not and the interval size is not limited. The dual-sub support vector machine has more advantages in classification effect and calculation speed.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a flowchart of a solution of a variational modal decomposition model;
FIG. 3 is a diagram of raw electroencephalograms representing three object states in a database, normal, epileptic seizure interval, and epileptic seizure period;
FIG. 4 is a representative electroencephalogram data fragment sample of healthy electroencephalogram A6, inter-epileptic seizure electroencephalogram C3 and epileptic seizure electroencephalogram E10 and an IMF component diagram of each stage after VMD decomposition;
fig. 5 is a boxplot of maximum, minimum, mean, variance, skewness, kurtosis, coefficient of variation, and fluctuation index for IMF1, IMF2, IMF3, and IMF4 in 5 datasets for normal, inter-seizure, and seizure periods.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, a multi-feature epilepsy signal classification method based on VMD and NLTWSVM, wherein VMD is an abbreviation of variable Mode Decomposition, chinese is Variational modal Decomposition, NLTWSVM is Non-linear Twin Support Vector Machine, and the abbreviation is of nonlinear two-member Support Vector Machine, the method of the present invention includes the following steps:
step 1: selecting an electroencephalogram database clinically collected by a German Bonn epilepsy research laboratory, wherein the database comprises 5 groups of electroencephalogram data sets, each group comprises 100 sections of electroencephalogram signals with the duration of 23.6s, each section of signals is collected at 4096 points, and the sampling frequency is 173.6 Hz. The data set A, B is cerebral cortex electroencephalogram data acquired by healthy people when eyes are opened and closed, the data set C, D is intracranial electroencephalogram data acquired by an epileptic patient during the seizure period outside an epileptogenic focus and in the epileptogenic focus, and the data set E is intracranial electroencephalogram data acquired by the epileptic patient during the seizure period in the epileptogenic focus. The method comprises the steps of selecting 100 groups of 5 electroencephalogram data in an experiment, dividing the first 23s electroencephalogram data of each group into 23 data segment samples with 1s as a unit, enabling the length of each sample to be 173 points of data, randomly selecting 50 groups of electroencephalogram data of each data set, segmenting the electroencephalogram data to obtain 1150 data segment samples, using a characteristic value obtained by calculation of each data segment sample as a training sample of a classifier, and using the corresponding characteristic value of the data segment sample of each remaining data set for testing.
Step 2: carrying out variation mode decomposition on the preprocessed electroencephalogram signal x (t) to obtain an intrinsic mode function uk(t), K is the order of the variational mode function;
step 2.1 the variational model is established by the following steps:
step 2.1.1, for each IMF component uk(t) performing Hilbert transform, constructing an analytic signal, and obtaining a single-side frequency spectrum:
step 2.1.2, mixing analytic signals of each IMF component into an estimated center frequencyModulating the spectrum of each IMF component to a corresponding baseband:
step 2.1.3, calculate the square L of the demodulated signal gradient represented by equation (2)2Norm, estimating the signal bandwidth of each IMF component, introducing constraint conditions, and constructing a variational model in the following form:
in the formula: δ (t) is the impulse function, { u ] is the convolution signk}={u1,u2,…,uKIs each IMF component obtained by decomposition, { omega }k}={ω1,ω2,…,ωKIs the center frequency of each IMF component,
and 2.2, introducing a secondary penalty factor alpha and a Lagrange multiplication operator lambda (t) to convert the constraint variation problem into the non-constraint variation problem. The extended Lagrangian function expression constructed is as follows:
step 2.3, solving the variation problem by using a multiplier orientation Method (ADMM), and alternately updatingAnd λn+1And (t) seeking a 'saddle point' of the extended Lagrangian function, wherein the 'saddle point' is the optimal solution of the variational model. The iterative formula is as follows:
the iteration stop conditions are as follows:
in the formula, tau is Lagrange multiplier updating parameter, epsilon is convergence tolerance.
The mode u in equation (5)kThe updating and solving method comprises the following steps:
1, rewriting formula (5) into the following equivalent expression:
2 converting equation (25) to the frequency domain based on the Parseval/Plancherel fourier equidistant transform:
3. using omega-omega as omega in the term with penalty factor in the formula (26)kInstead, one can obtain:
4. since the real signal has hermitian symmetry, equation (27) can be changed to a form of half-space integration on the non-negative frequency domain:
5. solving equation (28) with a quadratic optimization problem yields:
central frequency ω of equation (6)kThe updating and solving method comprises the following steps:
1. equation (6) is rewritten as the equivalent expression:
2. according to the same procedure, equation (30) is fourier transformed into the frequency domain and finally into a form of half-space integration on the non-negative frequency domain:
3. solving the quadratic optimization problem of equation (31), the update method for solving the center frequency is as follows:
therefore, as shown in fig. 2, the solution specifically using the multiplicative operator alternating direction method includes the following steps:
2. updating n according to n ← n + 1;
3. iteratively updating { u ] according to equations (29) and (32)kAnd { omega } andk};
5. And (5) stopping iteration if the iteration stop condition is met (8), and otherwise, returning to the step 2.
And step 3: extracting multi-feature parameters from the obtained intrinsic mode function, and assuming a signal u obtained after decomposition of electroencephalogram signals x (t) through VMDk(t) is a time series of length L, and the extracted features can be found by the following expression:
maximum value: max (u) Maxk(t)) (9)
Minimum value: min ═ Min (u)k(t)) (10)
and 4, step 4: inputting 8 feature combinations obtained by feature extraction into a nonlinear dual support vector machine for training, wherein the nonlinear dual support vector machine comprises the following specific implementation steps:
training samples are set, wherein p positive class samples and q negative class samples exist, and the expression is as follows:
T={(x1,+1),(x2,+1),…(xp,+1),(xp+1,-1),(xp+2,-1),…(xp+q,-1)} (17)
wherein xi∈Rn,i=1,2,…,p+q;
For a nonlinear two-sub support vector machine, the original space linear indivisible samples are mapped into a linearly separable high-dimensional space by introducing a kernel function, and the nonlinear two-sub support vector machine still seeks two non-parallel decision hyperplanes as follows:
wherein C isT=[A;B]T∈Rn×l,A=(x1,x2,…,xp)T∈Rp×n,B=(xp+1,xp+2,…,xp+q)T∈Rq×n,l=p+q;
The dual sub-support vector machine requires that each hyperplane satisfies the condition that the distance from the corresponding class sample point is minimum and the distance from another class sample point is enough, and accordingly two convex quadratic programming problems are constructed as follows:
and
wherein e+And e-Is a column vector of elements all one, ξ+And xi-Is the relaxation vector. c. CiI is 1,2 is a penalty parameter;
the dual problems of formula (19) and formula (20) are respectively
And
wherein S ═ K (A, C)T)e+],R=[K(B,CT)e-];
By solving equations (21) and (22), the solution to the original problem can be obtained by solving the following equation.
For a new input x ∈ RnIts category is judged as
Where | is a point x to a hyperplane K (x)T,CT)uk+bkA vertical distance of 0, k-, + is provided.
The experimental data of the invention come from Germany Bonn epilepsy electroencephalogram database. Fig. 3 shows representative raw electroencephalogram signals in data sets corresponding to three object states, namely, normal, inter-seizure and intra-seizure states of epileptic patients. The method carries out 4-layer VMD decomposition on a data fragment sample to obtain IMF components of each order. Representative electroencephalogram data segment samples of healthy electroencephalogram A6, epileptic seizure interval electroencephalogram C3 and epileptic seizure period electroencephalogram E10 and IMF component diagrams of each order after VMD decomposition are respectively shown in FIG. 4. As can be seen from FIG. 4, IMF components of each order obtained by VMD decomposition represent different characteristic components in the original EEG signal, and the EEG in epileptic seizure period has more drastic change than the EEG in healthy state and epileptic seizure interval.
After obtaining each order IMF component of 5 EEG data set EEG data fragment samples through VMD decomposition, 8 characteristics of corresponding maximum value, minimum value, average value, variance, skewness, kurtosis, variation coefficient and fluctuation index of each IMF component are respectively calculated. The box plots of 8 characteristic indices in 5 datasets for normal, inter-seizure and seizure periods for IMF1, IMF2, IMF3 and IMF4 from VMD decomposition are shown in FIG. 5. As can be seen from fig. 5, in each characteristic index extracted by the method, the IMF components of each order in the epileptic seizure period have substantially significant statistical differences compared with the IMF components in the normal state and the epileptic seizure interval, and the statistical differences lay a good foundation for accurately classifying epileptic brain electrical signals and non-epileptic brain electrical signals. The classification accuracy of the nonlinear Gemini support vector machine used in the method can reach 98.59%, and the method is suitable for classification detection of epileptic electroencephalogram signals and non-epileptic electroencephalogram signals.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (5)
1. A multi-feature epilepsy signal classification method based on VMD and NLTWSVM is characterized by comprising the following steps:
preprocessing electroencephalogram signal data; the electroencephalogram signal data comprises an electroencephalogram data set of a healthy person, an epileptic seizure interval electroencephalogram data set and an epileptic seizure period electroencephalogram data set;
carrying out variation mode decomposition on the electroencephalogram signal x (t) to obtain a corresponding intrinsic mode function uk(t), K is the representation of a certain number of layers in K, wherein K is the order of the intrinsic mode function;
extracting multi-feature parameters from the obtained series of intrinsic mode functions;
and inputting the extracted multi-feature parameters serving as feature combinations into a nonlinear dual-sub support vector machine for training and classification.
2. The method of claim 1, wherein the pre-processing comprises: dividing the first 23s of electroencephalogram data of each group of electroencephalogram signals in the data set into 23 data fragment samples by taking 1s as a unit, randomly selecting 50 groups of electroencephalogram data from each data set, segmenting the electroencephalogram data to obtain data fragment samples as training samples of the classifier, and taking the remaining data fragment samples of each data set as test samples of the classifier.
3. The method of claim 1, wherein performing variational modal decomposition of the brain electrical signal x (t) comprises:
establishing a variation model;
introducing a secondary penalty factor alpha and a Lagrangian multiplication operator lambda (t), converting a constraint variation problem into an unconstrained variation problem, and constructing an extended Lagrangian function;
solving the variation problem by using a multiplicative operator alternating direction method and alternately updatingAnd λn+1(t) seeking a 'saddle point' of the extended Lagrangian function, wherein the 'saddle point' is an optimal solution of the variation model;
wherein, establishing the variation model comprises the following steps:
let uk(t) is the IMF component, for each IMF component uk(t) performing Hilbert transform, constructing an analytic signal, and obtaining a single-side frequency spectrum:
mixing analytic signals of each IMF component to estimate center frequencyModulating the spectrum of each IMF component to a corresponding baseband:
calculating the square L of the gradient of the demodulation signal expressed by equation (2)2Norm, estimating the signal bandwidth of each IMF component, introducing constraint conditions, and constructing a variational model as shown in formula (3):
in the formula: δ (t) is the impulse function, { u ] is the convolution signk}={u1,u2,…,uKIs each IMF component obtained by decomposition, { omega }k}={ω1,ω2,…,ωK-the center frequency of each IMF component, K is the number of IMF components,
the expression of the extended lagrange function is as follows:
the iterative formula adopted for solving the variational problem by using the multiplicative operator alternating direction method is as follows:
the iteration stop conditions are as follows:
in the formula, tau is Lagrange multiplier updating parameter, epsilon is convergence tolerance.
4. The method of claim 1, wherein, in extracting the multi-feature parameters for the eigenmode function, the electroencephalogram signal x (t) is decomposed by VMD to obtain a signal uk(t) is a time series of length L, the extracted features are solved by the following expression:
maximum value: max (u) Maxk(t)) (9)
Minimum value: min ═ Min (u)k(t)) (10)
5. the method of claim 1, wherein inputting the extracted multi-feature parameters as feature combinations into a nonlinear binary support vector machine for training and classification specifically comprises:
step 5.1: training samples are set, wherein p positive class samples and q negative class samples exist, and the expression is as follows:
T={(x1,+1),(x2,+1),…(xp,+1),(xp+1,-1),(xp+2,-1),…(xp+q,-1)} (17)
wherein xi∈Rn,i=1,2,…,p+q;
Step 5.2: for a nonlinear two-sub support vector machine, the original space linear indivisible samples are mapped into a linearly separable high-dimensional space by introducing a kernel function, and the nonlinear two-sub support vector machine still seeks two non-parallel decision hyperplanes as follows:
wherein C isT=[A;B]T∈Rn×l,A=(x1,x2,…,xp)T∈Rp×n,B=(xp+1,xp+2,…,xp+q)T∈Rq×n,l=p+q;
Step 5.3: the dual sub-support vector machine requires that each hyperplane satisfies the condition that the distance from the corresponding class sample point is minimum and the distance from another class sample point is enough, and accordingly two convex quadratic programming problems are constructed as follows:
and
wherein e+And e-Is a column vector of elements all one, ξ+And xi-Is a relaxation vector, ciI is 1,2 is a penalty parameter;
step 5.4: the dual problems of formula (19) and formula (20) are respectively
And
wherein S ═ K (A, C)T)e+],R=[K(B,CT)e-];
Step 5.5: by solving equations (21) and (22), the solution to the original problem is obtained by solving the following equation:
for a new input x ∈ RnIts category is judged as
Where | is a point x to a hyperplane K (x)T,CT)uk+bkA vertical distance of 0, k-, + is provided.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110771578.7A CN113558643A (en) | 2021-07-08 | 2021-07-08 | Multi-feature epilepsy signal classification method based on VMD and NLTWSVM |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110771578.7A CN113558643A (en) | 2021-07-08 | 2021-07-08 | Multi-feature epilepsy signal classification method based on VMD and NLTWSVM |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113558643A true CN113558643A (en) | 2021-10-29 |
Family
ID=78164050
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110771578.7A Pending CN113558643A (en) | 2021-07-08 | 2021-07-08 | Multi-feature epilepsy signal classification method based on VMD and NLTWSVM |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113558643A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114159070A (en) * | 2021-12-20 | 2022-03-11 | 武汉大学 | Real-time cardiac arrest risk prediction method and system of convolutional neural network |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104573630A (en) * | 2014-12-05 | 2015-04-29 | 杭州电子科技大学 | Multiclass brain electrical mode online identification method based on probability output of twin support vector machine |
CN107153825A (en) * | 2017-05-23 | 2017-09-12 | 哈尔滨理工大学 | A kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs |
CN110728231A (en) * | 2019-10-10 | 2020-01-24 | 华东理工大学 | Sleep staging method based on improved particle swarm algorithm and twin support vector machine |
CN111383710A (en) * | 2020-03-13 | 2020-07-07 | 闽江学院 | Gene splice site recognition model construction method based on particle swarm optimization gemini support vector machine |
CN111543946A (en) * | 2020-05-08 | 2020-08-18 | 南京邮电大学 | Epilepsia electroencephalogram signal automatic detection method based on improved variational modal decomposition algorithm |
US10743809B1 (en) * | 2019-09-20 | 2020-08-18 | CeriBell, Inc. | Systems and methods for seizure prediction and detection |
CN112450885A (en) * | 2020-11-27 | 2021-03-09 | 湖南师范大学 | Epileptic electroencephalogram-oriented identification method |
-
2021
- 2021-07-08 CN CN202110771578.7A patent/CN113558643A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104573630A (en) * | 2014-12-05 | 2015-04-29 | 杭州电子科技大学 | Multiclass brain electrical mode online identification method based on probability output of twin support vector machine |
CN107153825A (en) * | 2017-05-23 | 2017-09-12 | 哈尔滨理工大学 | A kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs |
US10743809B1 (en) * | 2019-09-20 | 2020-08-18 | CeriBell, Inc. | Systems and methods for seizure prediction and detection |
CN110728231A (en) * | 2019-10-10 | 2020-01-24 | 华东理工大学 | Sleep staging method based on improved particle swarm algorithm and twin support vector machine |
CN111383710A (en) * | 2020-03-13 | 2020-07-07 | 闽江学院 | Gene splice site recognition model construction method based on particle swarm optimization gemini support vector machine |
CN111543946A (en) * | 2020-05-08 | 2020-08-18 | 南京邮电大学 | Epilepsia electroencephalogram signal automatic detection method based on improved variational modal decomposition algorithm |
CN112450885A (en) * | 2020-11-27 | 2021-03-09 | 湖南师范大学 | Epileptic electroencephalogram-oriented identification method |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114159070A (en) * | 2021-12-20 | 2022-03-11 | 武汉大学 | Real-time cardiac arrest risk prediction method and system of convolutional neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Satapathy et al. | EEG signal classification using PSO trained RBF neural network for epilepsy identification | |
García-Martínez et al. | A review on nonlinear methods using electroencephalographic recordings for emotion recognition | |
Mandhouj et al. | An automated classification of EEG signals based on spectrogram and CNN for epilepsy diagnosis | |
Li et al. | FFT-based deep feature learning method for EEG classification | |
CN106886792B (en) | Electroencephalogram emotion recognition method for constructing multi-classifier fusion model based on layering mechanism | |
CN110070105B (en) | Electroencephalogram emotion recognition method and system based on meta-learning example rapid screening | |
CN111310570B (en) | Electroencephalogram signal emotion recognition method and system based on VMD and WPD | |
Alotaibi et al. | Ensemble Machine Learning Based Identification of Pediatric Epilepsy. | |
Mustafa et al. | The analysis of eeg spectrogram image for brainwave balancing application using ann | |
Geng et al. | Epileptic seizure detection based on improved wavelet neural networks in long-term intracranial EEG | |
CN114366124A (en) | Epilepsia electroencephalogram identification method based on semi-supervised deep convolution channel attention single classification network | |
Thanaraj et al. | Multichannel interictal spike activity detection using time–frequency entropy measure | |
Goshvarpour et al. | Gender and age classification using a new Poincare section-based feature set of ECG | |
CN114305452B (en) | Cross-task cognitive load identification method based on electroencephalogram and field adaptation | |
Wang et al. | Automated rest eeg-based diagnosis of depression and schizophrenia using a deep convolutional neural network | |
Tang et al. | Epileptic seizure detection based on path signature and bi-LSTM network with attention mechanism | |
Siuly et al. | Exploring Rhythms and Channels-Based EEG Biomarkers for Early Detection of Alzheimer's Disease | |
Tripathi et al. | Automatic seizure detection and classification using super-resolution superlet transform and deep neural network-A preprocessing-less method | |
CN110338760A (en) | A kind of three classification method of schizophrenia based on electroencephalogram frequency domain data | |
CN113558643A (en) | Multi-feature epilepsy signal classification method based on VMD and NLTWSVM | |
Anugraha et al. | A machine learning application for epileptic seizure detection | |
Aung et al. | Analysis of EEG signals contaminated with motion artifacts using multiscale modified-distribution entropy | |
Sharma | Localization of epileptic surgical area using automated hybrid approach based on higher-order statistics with sensitivity analysis and residual wavelet transform | |
CN116898454B (en) | Epileptic classification method and system based on electroencephalogram feature fusion deep learning model | |
Goshvarpour et al. | An innovative information-based strategy for epileptic EEG classification |
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 |