CN107153825A - A kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs - Google Patents
A kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs Download PDFInfo
- Publication number
- CN107153825A CN107153825A CN201710368929.3A CN201710368929A CN107153825A CN 107153825 A CN107153825 A CN 107153825A CN 201710368929 A CN201710368929 A CN 201710368929A CN 107153825 A CN107153825 A CN 107153825A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- eeg
- mtd
- epileptic
- 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
- G06F2218/12—Classification; Matching
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Abstract
The present invention relates to a kind of epileptic electroencephalogram (eeg) classification field, and in particular to a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs.The problem of present invention is low to epileptic EEG Signal classification accuracy in the prior art, class categories are few, it is theoretical according to particle cluster algorithm and SVMs, it is proposed that a kind of Modulation recognition method based on particle cluster algorithm Support Vector Machines Optimized parameter.The epileptic electroencephalogram (eeg) sorting technique of the present invention includes step:Choose experimental data, chosen respectively in database Healthy People it is clear-headed when EEG signals, three groups of signals of EEG signals and the EEG signals of epileptic patient stage of attack of epileptics human hair interictal each 100 sections, and carry out decomposed and reconstituted, extract effective wave band;Solve characteristic vector;Experimental data is classified using SVMs;With the penalty factor and kernel functional parameter σ of particle cluster algorithm Support Vector Machines Optimized.The present invention is applied to normal, the eeg signal classification of epileptic attack and status epilepticus.
Description
Technical field
The present invention relates to a kind of epileptic electroencephalogram (eeg) classification field, and in particular to a kind of epileptic electroencephalogram (eeg) based on SVMs point
Class method.
Background technology
Epilepsy is the chronic disease of a kind of part being characterized with intracerebral neuron paradoxical discharge or full brain disorder, China
Possess the epileptic of substantial amounts, and with the speed of 600,000 in increase.The classification of epileptic EEG Signal and detection technique
Contribute to the workload of reduction medical personnel, the clinical practice with practical significance.Prior art utilizes approximate entropy, fluctuation system
Number, ExtremeLearningMachine carry out two to epileptic EEG Signal and classified, or utilize empirical mode decomposition and SVMs, to normal
EEG signals and epileptic EEG Signal are classified, or are calculated using WAVELET PACKET DECOMPOSITION coefficient matrix, Wavelet Packet Entropy, AdaBoost
Method, by eeg signal classification into normal condition and status epilepticus, and is waited with SVMs and recurrence quantification analysis point
Class EEG signals, or EEG signals are classified with Sample Entropy, AR parameters and adaptive extreme learning machine, or use wavelet transformation
Feature is extracted with the method that high level matrix is combined, EEG signals are classified using SVMs.The above method is by brain
Electric signal is categorized into normal signal and epilepsy signal, but does not account for the classification of epileptic attack intermittent phase signal, set forth herein
A kind of Modulation recognition detection technique of particle cluster algorithm Support Vector Machines Optimized, solves classification accuracy and class categories number
Part conflicts.On the basis of nicety of grading is ensured, EEG signals are classified as normally, epileptic attack and status epilepticus.
The content of the invention
The present invention is for the problem of epileptic EEG Signal classification accuracy in the prior art is low, class categories are few, according to grain
Swarm optimization and SVMs are theoretical, it is proposed that a kind of Modulation recognition based on particle cluster algorithm Support Vector Machines Optimized parameter
Detection method.
The object of the present invention is achieved like this:A kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs, specific step
Suddenly include:
Step a, selection experimental data:Choose EEG signals when Healthy People is regained consciousness, epileptic patient respectively in database
Three groups of signals of EEG signals and the EEG signals of epileptic patient stage of attack of hair interictal, each 100 sections, and carry out decomposing weight
Structure, extracts effective wave band;
Step b, solution characteristic vector;
Step c, using SVMs to experimental data classify;
Step d, the penalty factor with particle cluster algorithm Support Vector Machines Optimized and kernel functional parameter σ.
Further, the specific method of the step a is:EEG signals when 100 sections of Healthy Peoples are regained consciousness are chosen respectively, it is insane
Epilepsy patient sends out the EEG signals and the EEG signals of epileptic patient stage of attack of interictal, and every section of duration 23.6s is adopted per segment signal
Collect at 4097 points, sample frequency is 173.6Hz, using wavelet transformation, 5 layers of decomposition and reconstruction are carried out to EEG signals, extracted
Imitate frequency range.
Yet further, decomposed and reconstituted using five layers of db4 wavelet transformations progress, third and fourth and five layers of reconstruction signal are concentrated
Between 0.5HZ~35HZ.
Further, the specific steps of the step b include:
Step b1, calculated into approximate entropy using innovatory algorithm;
Step b2, extraction coefficient of variation.
Yet further, the specific steps of the step b1 include:
Step b11, pending epileptic EEG Signal number are N, calculate N*N Distance matrix D, D is symmetrical matrix, D
The i-th row jth column element be designated as dij,
Step b12, using the element in Distance matrix D, to each vectorial XiCount dij<R number, and obtain the number
With distance sum N-m ratio
M increases by 1, are tried to achieve
It is step b13, rightTake the logarithm, then all i are averaged, be counted as φm(r):
M increases by 1, try to achieve φm+1(r):
Step b14, according to φmAnd φ (r)m+1(r) approximate entropy ApEn is tried to achieve:
ApEn (m, r)=φm(r)-φm+1(r)
Wherein, m=2, r=0.2*std (sig), std (sig) are signal sig standard deviation.
Yet further, the specific method of calculating coefficient of variation is in the step b2:Coefficient of variation can be expressed as:
Wherein, aniFor i-th layer after n-th section of eeg data wavelet transformation of amplitude, M is the length of signal, n spans 1-
100, i values are that 3,4 and 5, M take 4097.
Further, the specific steps of the step c include:
Step c1, hypothesis have two classes experimental data to be sorted:(x1,y1),(x2,y2),...,(xn,yn),x∈Rn,yi∈
{ 1, -1 }, xiIt is experimental data, yiIt is the label of experimental data, there is an optimal classification surface, its equation can be expressed as:
wTX+b=0
Wherein, wTIt is the normal vector of optimal classification surface, b is the constant term of optimal classification surface;
By this optimal classification surface, can classify two kinds of sample data, order:
F (x)=wTx+b
If f (x)=0, then x is the point being located on classifying face;
If f (x) < 0, the corresponding label y of x can be set as -1;F (x) > 0, then the corresponding label y of x are 1, will be real
Data are tested to be classified;
If all test samples can correctly classify, test sample meets condition:
yi[(wT·xi)+b] -1 >=0, i=1 ..., n
Two classifying face H1、H2The distance between be 2/ | | w | |, to find optimal classification line, that is, cause 2/ | | w | | it is maximum,
I.e.Minimize, optimal classification problem changes into a constrained optimization problem
s.t yi[w.x+b] -1 >=0, i=1 ..., N
The quadratic programming problem of above formula is solved, the w and b of optimal classification surface is obtained, optimal classification function is:
F (x)=sgn (wx+b)
Step c2, using kernel function the data of linearly inseparable are handled, in the number of lower dimensional space linearly inseparable
According to xi, higher dimensional space is arrived by Nonlinear Mapping Φ (x) so that data are found out in higher dimensional space linear separability in higher dimensional space
Optimal classification surface, non-linear conversion can be realized by meeting the kernel function of Mercer conditions,
The kernel function of the Mercer conditions of definition:
K(xi,xj)=Φ (xi).Φ(xj)
Radial basis kernel function:
Further, the specific steps of the step d include:
Step d1, initialization population:Particle populations number is set as 20, coordinates table of each particle in two-dimensional space
It is shown as xi=(xi1,xi2) (i=1,2 ..., 20), with speed vs=[vs1,vs2] (s=1,2 ..., 20) in two-dimentional solution space
Scan for;
Step d2, the fitness for calculating each particle:Particle needs to consider the optimal position of particle individual in more new position
Put piWith particle populations optimal location pg:
pi=(pi1,pi2), i=1,2 ..., 20,
pg=(pg1,pg2,…pgn), n=20;
Step d3, more new particle speed of searching optimization vector:
vs+1=vs+c1rand1()(pi-xi)
+c2rand2()(pg-xi)
Wherein, c1Take 1.5, c2Take 1.7, rand1() and rand2() represents 0 to 1 random number;
Step d4, renewal particle position:
xi+1=xi+vi+1
Step d5, judge whether to meet iterations 200, if:
It is unsatisfactory for, return to step d2;
Meet, then return to optimal solution.
Beneficial effect:
The present invention a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs, using particle cluster algorithm optimization support to
Amount machine parameter, is optimized to the parameter C and σ of SVMs, is found optimal parameter value, is further improved SVMs
Learning ability and convergence rate, improve the accuracy rate of data classification, and EEG signals be divided into normally, epileptic attack with
Status epilepticus, adds class categories number, and solve prior art solves classification accuracy and class categories number
Part conflicts.
Brief description of the drawings
The approximate entropy comparison diagram of the different group EEG signals of Fig. 1 this method;
The coefficient of variation of third layer after Fig. 2 each group EEG signals wavelet transformations;
4th layer of coefficient of variation after Fig. 3 each group EEG signals wavelet transformations;
The coefficient of variation of layer 5 after Fig. 4 each group EEG signals wavelet transformations;
The optimal classification surface schematic diagram obtained after Fig. 5 support vector cassifications;
Fig. 6 particle cluster algorithm flow charts.
Embodiment
Illustrate present embodiment, a kind of epileptic electroencephalogram (eeg) classification based on SVMs of present embodiment with reference to Fig. 1~6
Method, specific steps include:
Step a, selection experimental data:EEG signals when 100 sections of Healthy Peoples are regained consciousness, the breaking-out of epileptics human hair are chosen respectively
Between the phase EEG signals and the EEG signals of epileptic patient stage of attack, every section of duration 23.6s, per segment signal gather 4097 points, adopt
Sample frequency is 173.6Hz, using wavelet transformation, and db4 small echos carry out 5 layers of Wavelet decomposing and recomposing to three groups of EEG signals, third and fourth
And five layers of reconstruction signal are concentrated mainly between 0.5HZ~35HZ.
Step b, solution characteristic vector, specific steps include:
Step b1, calculated into approximate entropy using innovatory algorithm, concretely comprised the following steps:Step b11, pending epileptic electroencephalogram (eeg) letter
Number number is N, calculates N*N Distance matrix D, D is symmetrical matrix, and D the i-th row jth column element is designated as dij,
Step b12, using the element in Distance matrix D, to each vectorial XiCount dij<R number, and obtain the number
With distance sum N-m ratio
M increases by 1, are tried to achieve
It is step b13, rightTake the logarithm, then all i are averaged, be counted as φm(r):
M increases by 1, try to achieve φm+1(r):
Step b14, according to φmAnd φ (r)m+1(r) approximate entropy ApEn is tried to achieve:
ApEn (m, r)=φm(r)-φm+1(r)
Wherein, m=2, r=0.2*std (sig), std (sig) are signal sig standard deviation, difference group EEG signals
Approximate entropy comparison diagram as shown in figure 1, phase between health status, epileptic attack, the amplitude of the approximate entropy of epileptic attack phase have it is larger
Difference, different EEG signals can preferably be characterized by extracting the approximate entropy of EEG signals.
Step b2, extraction coefficient of variation, coefficient of variation can be expressed as:
Wherein, aniFor i-th layer after n-th section of eeg data wavelet transformation of amplitude, M is the length of signal, n spans 1-
100, i spans 3,4 and 5, M take 4097, Fig. 2, Fig. 3 and Fig. 4 to be respectively third layer, after each group EEG signals wavelet transformation
Four layers and the coefficient of variation of layer 5, there is more obvious difference under different conditions.Epileptic attack phase coefficient of variation it is maximum and
And fluctuation is stronger;The coefficient of variation of epileptic attack intermittent phase is more steady, and value is in health status and status epilepticus
Between;Coefficient of variation under health status is the most steady and value is minimum.Therefore the coefficient of variation conduct under available different scale
EEG signals feature.
Step c, SVMs (Support Vector Machine, SVM) are widely used in pattern-recognition and classification
In detection, SVM solves high-dimensional, small sample well, the problems such as non-linear.The core concept of SVMs is to find
A kind of optimal classification surface that two class samples can be correctly classified, and different samples are bigger away from classifying face distance, classification effect
Fruit is better.
Experimental data is classified using SVMs, specific steps include:
Step c1, hypothesis have two classes experimental data to be sorted:(x1,y1),(x2,y2),…,(xn,yn),x∈Rn,yi∈
{ 1, -1 }, xiIt is experimental data, yiIt is the label of experimental data, there is an optimal classification surface, its equation can be expressed as:
wTX+b=0
Wherein, wTIt is the normal vector of optimal classification surface, b is the constant term of optimal classification surface;
By this optimal classification surface, can classify two kinds of sample data, order:
F (x)=wTx+b
If f (x)=0, then x is the point being located on classifying face;
If f (x) < 0, the corresponding label y of x can be set as -1;F (x) > 0, then the corresponding label y of x are 1, will be real
Data are tested to be classified;
If all test samples can correctly classify, test sample should meet condition:
yi[(wT·xi)+b] -1 >=0, i=1 ..., n
Two classifying face H1、H2The distance between be 2/ | | w | |, to find optimal classification line, that is, cause 2/ | | w | | it is maximum,
I.e.Minimize, optimal classification problem changes into a constrained optimization problem
s.t yi[w.x+b] -1 >=0, i=1 ..., N
The quadratic programming problem of above formula is solved, the w and b of optimal classification surface is obtained, optimal classification function is:
F (x)=sgn (wx+b)
The optimal classification surface that is obtained after support vector cassification as shown in figure 5, square and circle represent it is two kinds of
Data, H represents optimal classification surface, and H1, H2 are for nearest from classification line sample in all kinds of and parallel to the straight line of classifying face, distribution
It is referred to as supporting vector in the point of the straight line, H1 and H2 distance are exactly class interval.So-called optimal classification surface refers to line of classifying
Two classes can be properly separated, class interval can be maximized again.wTIt is the normal vector of optimal classification surface, b is normal for optimal classification surface
It is several.
Step c2, using kernel function the data of linearly inseparable are handled, in the number of lower dimensional space linearly inseparable
According to xi, higher dimensional space is arrived by Nonlinear Mapping Φ (x) so that data are found out in higher dimensional space linear separability in higher dimensional space
Optimal classification surface, non-linear conversion can be realized by meeting the kernel function of Mercer conditions,
The kernel function of the Mercer conditions of definition:
K(xi,xj)=Φ (xi).Φ(xj)
Kernel function has diversified forms, and what is be commonly used mainly includes RBF, polynomial function, Sigmoid functions
Deng.The difference of kernel function can constitute different SVMs, and the kernel function that present embodiment is chosen is RBF, radially
Base kernel function:
Step d, the penalty factor with particle cluster algorithm Support Vector Machines Optimized and kernel functional parameter σ, particle cluster algorithm stream
Journey figure is as shown in fig. 6, specific steps include:
Step d1, initialization population:Particle populations number is set as 20, coordinates table of each particle in two-dimensional space
It is shown as xi=(xi1,xi2) (i=1,2 ..., 20), with speed vs=[vs1,vs2] (s=1,2 ..., 20) in two-dimentional solution space
Scan for;
Step d2, the fitness for calculating each particle:Particle needs to consider the optimal position of particle individual in more new position
Put piWith particle populations optimal location pg:
pi=(pi1,pi2), i=1,2 ..., 20,
pg=(pg1,pg2,…pgn), n=20;
Step d3, more new particle speed of searching optimization vector:
vs+1=vs+c1rand1()(pi-xi)
+c2rand2()(pg-xi)
Wherein, c1Take 1.5, c2Take 1.7, rand1() and rand2() represents 0 to 1 random number;
Step d4, renewal particle position:
xi+1=xi+vi+1
Step d5, judge whether to meet iterations 200, if:
It is unsatisfactory for, return to step d2;
Meet, then return to optimal solution.
Claims (8)
1. a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs, it is characterised in that specific steps include:
Step a, selection experimental data:Choose three groups of signals respectively in database, carrying out five layers using wavelet transformation decomposes weight
Structure, extracts effective wave band;Three groups of described signals are respectively EEG signals when 100 sections of Healthy Peoples are clear-headed, 100 sections of epileptic patients
The EEG signals of the EEG signals of interictal and 100 sections of epileptic patient stages of attack;
Step b, solution characteristic vector;
Step c, using SVMs to experimental data classify;
Step d, the penalty factor with particle cluster algorithm Support Vector Machines Optimized and kernel functional parameter σ.
2. a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs according to claim 1, it is characterised in that in step
In rapid a, EEG signals when 100 sections of Healthy Peoples are clear-headed, the EEG signals of 100 sections of epileptic patient interictals and 100 sections
The EEG signals of epileptic patient stage of attack, a length of 23.6s at every section gathers per segment signal at 4097 points, and sample frequency is
173.6Hz。
3. a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs according to claim 1 or 2, it is characterised in that
In step a, using db4 wavelet transformations carry out five layers it is decomposed and reconstituted, third and fourth and five layers of reconstruction signal concentrate on 0.5HZ~
Between 35HZ.
4. a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs according to claim 1, it is characterised in that described
Step b specific steps include:
Step b1, utilize innovatory algorithm calculate approximate entropy;
Step b2, extraction coefficient of variation.
5. a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs according to claim 4, it is characterised in that described
Step b1 specific steps include:
Step b11, pending epileptic EEG Signal number are N, calculate N*N Distance matrix D, D is symmetrical matrix, the i-th of D
Row jth column element is designated as dij,
<mrow>
<msub>
<mi>d</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mn>1</mn>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mo>|</mo>
<mi>u</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>u</mi>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>|</mo>
<mo><</mo>
<mi>r</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mn>0</mn>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mo>|</mo>
<mi>u</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>u</mi>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>|</mo>
<mo>&GreaterEqual;</mo>
<mi>r</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>,</mo>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
<mo>~</mo>
<mi>N</mi>
<mo>,</mo>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
<mo>~</mo>
<mi>N</mi>
</mrow>
Step b12, using the element in Distance matrix D, to each vectorial XiCount dij<R number, and obtain the number with away from
From total N-m ratio
<mrow>
<msubsup>
<mi>C</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>N</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<msub>
<mi>d</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>&cap;</mo>
<msub>
<mi>d</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
<mo>(</mo>
<mi>j</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</msub>
</mrow>
M increases by 1, are tried to achieve
<mrow>
<msubsup>
<mi>C</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</msubsup>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>N</mi>
<mo>-</mo>
<mn>2</mn>
</mrow>
</munderover>
<msub>
<mi>d</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>&cap;</mo>
<msub>
<mi>d</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
<mo>(</mo>
<mi>j</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</msub>
<mo>&cap;</mo>
<msub>
<mi>d</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>+</mo>
<mn>2</mn>
<mo>)</mo>
<mo>(</mo>
<mi>j</mi>
<mo>+</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</msub>
<mo>;</mo>
</mrow>
It is step b13, rightTake the logarithm, then all i are averaged, be counted as φm(r):
<mrow>
<msup>
<mi>&phi;</mi>
<mi>m</mi>
</msup>
<mrow>
<mo>(</mo>
<mi>r</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>N</mi>
<mo>-</mo>
<mi>m</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</munderover>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>C</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>N</mi>
<mo>-</mo>
<mi>m</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
M increases by 1, try to achieve φm+1(r):
<mrow>
<msup>
<mi>&phi;</mi>
<mrow>
<mi>m</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msup>
<mrow>
<mo>(</mo>
<mi>r</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>N</mi>
<mo>-</mo>
<mi>m</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</munderover>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>C</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>N</mi>
<mo>-</mo>
<mi>m</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
1
Step b14, according to φmAnd φ (r)m+1(r) approximate entropy ApEn is tried to achieve:
ApEn (m, r)=φm(r)-φm+1(r)
Wherein, m is the length of comparison window, and m=2, r=0.2*std (sig), std (sig) are signal sig standard deviation.
6. a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs according to claim 4, it is characterised in that described
The specific method of coefficient of variation is calculated in step b2 is:
<mrow>
<msub>
<mi>F</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>M</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<mo>|</mo>
<msub>
<mi>a</mi>
<mrow>
<mi>n</mi>
<mi>i</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>-</mo>
<msub>
<mi>a</mi>
<mrow>
<mi>n</mi>
<mi>i</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>|</mo>
</mrow>
Wherein, aniFor i-th layer after n-th section of eeg data wavelet transformation of amplitude, M is the length of signal, n span 1-100,
I values are that 3,4 and 5, M take 4097.
7. a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs according to claim 1, it is characterised in that described
Step c specific steps include:
Step c1, there is two classes experimental data to be sorted:(x1,y1),(x2,y2),...,(xn,yn),x∈Rn,yi∈ { 1, -1 },
xiIt is experimental data, yiIt is the label of experimental data, there is an optimal classification surface, its equation can be expressed as:
wTX+b=0
Wherein, wTIt is the normal vector of optimal classification surface, b is the constant term of optimal classification surface;
By this optimal classification surface, two kinds of sample data of classifying, order:
F (x)=wTx+b
If:
F (x)=0, then x is the point being located on classifying face;
F (x) < 0, set the corresponding label y of x as -1;
F (x) > 0, set the corresponding label y of x as 1;
According to label y, experimental data is classified;
Step c2, utilize Radial basis kernel functionData to linearly inseparable are handled, low
The data x of dimension space linearly inseparablei, higher dimensional space is arrived by Nonlinear Mapping Φ (x) so that data are linear in higher dimensional space
It can divide, optimal classification surface is found out in higher dimensional space.
8. a kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs according to claim 1, it is characterised in that described
Step d specific steps include:
Step d1, initialization population:Particle populations number is set as 20, and coordinate representation of each particle in two-dimensional space is
xi=(xi1,xi2) (i=1,2 ..., 20), with speed vs=[vs1,vs2] (s=1,2 ..., 20) carried out in two-dimentional solution space
Search;
Step d2, the fitness for calculating each particle:Particle needs to consider the optimal location p of particle individual in more new positioniWith
Particle populations optimal location pg:
pi=(pi1,pi2), i=1,2 ..., 20,
pg=(pg1,pg2,…pgn), n=20;
Step d3, more new particle speed of searching optimization vector:
vs+1=vs+c1rand1()(pi-xi)
+c2rand2()(pg-xi)
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>v</mi>
<mi>s</mi>
</msub>
<mo>=</mo>
<msub>
<mi>v</mi>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>i</mi>
<mi>f</mi>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>v</mi>
<mi>s</mi>
</msub>
<mo>></mo>
<msub>
<mi>v</mi>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>,</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>v</mi>
<mi>s</mi>
</msub>
<mo>=</mo>
<mo>-</mo>
<msub>
<mi>v</mi>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>i</mi>
<mi>f</mi>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>v</mi>
<mi>s</mi>
</msub>
<mo><</mo>
<mo>-</mo>
<msub>
<mi>v</mi>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>.</mo>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
Wherein, c1Take 1.5, c2Take 1.7, rand1() and rand2() represents 0 to 1 random number;
Step d4, renewal particle position:
xi+1=xi+vi+1
Step d5, judge whether to meet iterations 200, if:
It is unsatisfactory for, return to step d2;
Meet, then return to optimal solution.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710368929.3A CN107153825A (en) | 2017-05-23 | 2017-05-23 | A kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710368929.3A CN107153825A (en) | 2017-05-23 | 2017-05-23 | A kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107153825A true CN107153825A (en) | 2017-09-12 |
Family
ID=59793138
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710368929.3A Pending CN107153825A (en) | 2017-05-23 | 2017-05-23 | A kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107153825A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108280464A (en) * | 2017-12-22 | 2018-07-13 | 南京邮电大学 | The brain electrical feature extracting method of DWT and EMD fusion approximate entropies |
CN108320800A (en) * | 2018-04-16 | 2018-07-24 | 吉林大学 | Epileptic seizure detects and the preceding eeg data analysis system predicted of breaking-out |
CN108742603A (en) * | 2018-04-03 | 2018-11-06 | 山东大学 | It is a kind of using kernel function and dictionary to the brain electric detection method and device of learning model |
CN109431497A (en) * | 2018-10-23 | 2019-03-08 | 南京医科大学 | A kind of brain-electrical signal processing method and epilepsy detection system |
CN109934089A (en) * | 2018-10-31 | 2019-06-25 | 北京航空航天大学 | Multistage epileptic EEG Signal automatic identifying method based on supervision gradient lifter |
CN110477912A (en) * | 2019-09-20 | 2019-11-22 | 电子科技大学 | Epileptic electroencephalogram (eeg) amplitude irreversibility analysis method based on equivalent arrangement and standard deviation |
CN111914735A (en) * | 2020-07-29 | 2020-11-10 | 天津大学 | Epilepsia electroencephalogram signal feature extraction method based on TQWT and entropy features |
CN112790774A (en) * | 2020-11-10 | 2021-05-14 | 天津职业技术师范大学(中国职业培训指导教师进修中心) | Original electroencephalogram deep learning classification method and application |
CN113558643A (en) * | 2021-07-08 | 2021-10-29 | 吉林大学 | Multi-feature epilepsy signal classification method based on VMD and NLTWSVM |
CN113616161A (en) * | 2021-09-16 | 2021-11-09 | 山东中科先进技术研究院有限公司 | Epileptic seizure prediction system and method |
-
2017
- 2017-05-23 CN CN201710368929.3A patent/CN107153825A/en active Pending
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108280464A (en) * | 2017-12-22 | 2018-07-13 | 南京邮电大学 | The brain electrical feature extracting method of DWT and EMD fusion approximate entropies |
CN108742603A (en) * | 2018-04-03 | 2018-11-06 | 山东大学 | It is a kind of using kernel function and dictionary to the brain electric detection method and device of learning model |
CN108320800A (en) * | 2018-04-16 | 2018-07-24 | 吉林大学 | Epileptic seizure detects and the preceding eeg data analysis system predicted of breaking-out |
CN109431497A (en) * | 2018-10-23 | 2019-03-08 | 南京医科大学 | A kind of brain-electrical signal processing method and epilepsy detection system |
CN109431497B (en) * | 2018-10-23 | 2020-08-11 | 南京医科大学 | Electroencephalogram signal processing method and epilepsy detection system |
CN109934089A (en) * | 2018-10-31 | 2019-06-25 | 北京航空航天大学 | Multistage epileptic EEG Signal automatic identifying method based on supervision gradient lifter |
CN109934089B (en) * | 2018-10-31 | 2020-10-23 | 北京航空航天大学 | Automatic multi-stage epilepsia electroencephalogram signal identification method based on supervised gradient raiser |
CN110477912A (en) * | 2019-09-20 | 2019-11-22 | 电子科技大学 | Epileptic electroencephalogram (eeg) amplitude irreversibility analysis method based on equivalent arrangement and standard deviation |
CN111914735A (en) * | 2020-07-29 | 2020-11-10 | 天津大学 | Epilepsia electroencephalogram signal feature extraction method based on TQWT and entropy features |
CN112790774A (en) * | 2020-11-10 | 2021-05-14 | 天津职业技术师范大学(中国职业培训指导教师进修中心) | Original electroencephalogram deep learning classification method and application |
CN113558643A (en) * | 2021-07-08 | 2021-10-29 | 吉林大学 | Multi-feature epilepsy signal classification method based on VMD and NLTWSVM |
CN113616161A (en) * | 2021-09-16 | 2021-11-09 | 山东中科先进技术研究院有限公司 | Epileptic seizure prediction system and method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107153825A (en) | A kind of epileptic electroencephalogram (eeg) sorting technique based on SVMs | |
CN106886792B (en) | Electroencephalogram emotion recognition method for constructing multi-classifier fusion model based on layering mechanism | |
Diykh et al. | Complex networks approach for EEG signal sleep stages classification | |
Cui et al. | Automatic sleep stage classification based on convolutional neural network and fine-grained segments | |
CN104771163B (en) | EEG feature extraction method based on CSP and R CSP algorithms | |
Saha et al. | EEG analysis for olfactory perceptual-ability measurement using a recurrent neural classifier | |
Li et al. | Analysis and classification of EEG signals using a hybrid clustering technique | |
Yang et al. | Mlp with riemannian covariance for motor imagery based eeg analysis | |
Alvi et al. | Neurological abnormality detection from electroencephalography data: a review | |
Jiang et al. | Sleep stage classification using covariance features of multi-channel physiological signals on Riemannian manifolds | |
Chen et al. | Fusing frequency-domain features and brain connectivity features for cross-subject emotion recognition | |
Sameer et al. | CNN based framework for detection of epileptic seizures | |
Alvarez-Meza et al. | Kernel-based relevance analysis with enhanced interpretability for detection of brain activity patterns | |
Alvi et al. | Developing a deep learning based approach for anomalies detection from EEG data | |
Dai et al. | Classification of electroencephalogram signals using wavelet-CSP and projection extreme learning machine | |
CN113010013A (en) | Wasserstein distance-based motor imagery electroencephalogram migration learning method | |
CN110338760B (en) | Schizophrenia three-classification method based on electroencephalogram frequency domain data | |
Rajaguru et al. | A framework for epilepsy classification using modified sparse representation classifiers and naive Bayesian classifier from electroencephalogram signals | |
CN112884063A (en) | P300 signal detection and identification method based on multivariate space-time convolution neural network | |
Cai et al. | Brainsleepnet: Learning multivariate eeg representation for automatic sleep staging | |
Subasi et al. | The ensemble machine learning-based classification of motor imagery tasks in brain-computer interface | |
Liu et al. | Extracting multi-scale and salient features by MSE based U-structure and CBAM for sleep staging | |
Jiang et al. | Early seizure detection in childhood focal epilepsy with electroencephalogram feature fusion on deep autoencoder learning and channel correlations | |
Mahato et al. | Analysis of region of interest (RoI) of brain for detection of depression using EEG signal | |
Zhang et al. | Four-classes human emotion recognition via entropy characteristic and random Forest |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170912 |
|
WD01 | Invention patent application deemed withdrawn after publication |