CN104000587A - Electroencephalogram (EEG) signal identifying system based on edge wavelet characteristics - Google Patents

Abstract

The embodiment of the invention discloses an electroencephalogram (EEG) signal identifying system based on edge wavelet characteristics. A method comprises the following steps of carrying out K layer wavelet decomposition on input EEG signals, summing low frequency components of a (K-1) layer, low frequency components of a K layer and high frequency components of the K layer, normalizing the low frequency components of the (K-1) layer, low frequency components of the K layer and high frequency components of the K layer to generate edge wavelet characteristics, choosing channels of the input EEG signals according to a Fisher criterion, choosing the channels which can better express brain thinking models, combining the channels to generate vectors of the edge wavelet characteristics, using a super-Dirichlet mixture model to simulate distribution of super vectors of the characteristics, working out parameters in the model, identifying the EEG signals to be identified, carrying out characteristic extraction and channel selection on the EEG signals to be identified to form super vectors, sending the super vectors into a probability model to calculate out probability values, and identifying the EEG signal through the prior probability and according to the principle of the maximum posterior probability. According to the EEG signal identifying system based on the edge wavelet characteristics, the identifying accuracy of the EEG signals is improved. Furthermore, the EEG signal identifying system based on the edge wavelet characteristics has larger practical value.

Description

A kind of brain wave (EEG) signal recognition system based on edge wavelet character

Technical field

The invention belongs to processing of biomedical signals field, described emphatically a kind of brain wave recognition system based on edge wavelet character and super Di Li Cray mixed model.

Background technology

Along with the development of computer technology and bio-medical technology, brain-machine switching technology has very important researching value and using value.By this technology, can help to suffer from the patient of nerve muscle disease, by brain, control computer and drive the various external equipments such as artificial limb to exchange with the external world.Eeg signal can be used for having recorded the active situation of human brain, when brain is being imagined different action, will express different eeg signals.How effectively eeg signal to be carried out to effective feature extraction, and to carry out Classification and Identification be an important subject in brain-machine exchange system.

Existing feature extracting method mainly comprises auto-regressive parameter feature, discrete wavelet coefficient characteristics etc., and discrete wavelet feature, owing to can carry out feature extraction in multiscale space, is therefore widely used at brain wave analysis field.The time sensitivity having due to eeg signal itself, therefore cause being mingled with in discrete wavelet coefficient characteristics more noise signal, be unfavorable for follow-up Classification and Identification work, so in this patent, adopted the edge wavelet character with lower temporal sensitivity as basic feature.According to the characteristic distributions of edge wavelet character itself, in the Classification and Identification stage, adopt the super Di Li of mixing Cray model to carry out modeling to extracting the distribution of feature simultaneously, adopt the method for maximum a posteriori probability to identify eeg signal.

Summary of the invention

In order to solve the existing defect of above-mentioned technology and to improve the discrimination of eeg signal, the invention provides a kind of brain wave recognition methods based on edge wavelet character and super Di Li Cray mixed model.

For achieving the above object, the brain wave recognition methods that the present invention proposes comprises the following steps:

One, characteristic extraction step:

A, wavelet transform step: to the M road eeg signal obtaining from sensor, utilize wavelet transformation to carry out the decomposition of K floor.

B, edge wavelet character extraction step: to the low frequency part in front K-1 layer wavelet coefficient, and low frequency and HFS in K layer wavelet coefficient are sued for peace, connection and normalized, generate the edge wavelet character of K+1 dimension.

Two, feature selection and model training step:

A, feature selection step:

To the M Road Edge wavelet character extracting, utilize Fisher criterion to carry out feature selection, pick out the feature that m road best embodies brain wave construction features.

B, model training step:

To obtain the feature super vector that K+1Wei edge, m road wavelet character carries out the dimension of combination producing m * (K+1).Use the distribution of super Di Li Cray mixed model (SDMM:super-Dirichlet Mixture Model) simulation edge wavelet character super vector, by gradient method, solve an equation and obtain the parameter in model, finally obtain a series of models, the brain wave that each model is corresponding a type.

Three, differentiate coupling step: extract after certain eeg signal data, adopt the method for step 1 to carry out being input in a series of probabilistic models that train after feature extraction, the effective passage obtaining according to training in step 2 A carries out characteristics combination and forms super vector, and utilize training pattern in step 2 B to calculate the probability for this model, and in conjunction with the prior distribution of eeg signal, adopt the method for maximum a posteriori to identify eeg signal.

According in the wavelet transform step described in a kind of brain wave recognition methods step 1 A based on edge wavelet character and super Di Li Cray mixed model of an embodiment of the invention, adopt the coefficient of its high and low rank wave filter of the western wavelet basis of the 2 many shellfishes in rank to be respectively

H＝[-0.129,-0.224,0.837,-0.483] ^T

L＝[-0.129,0.224,0.837,0.483] ^T

As follows according to the edge wavelet character extraction step described in a kind of brain wave recognition methods step 1 B based on edge wavelet character and super Di Li Cray mixed model of an embodiment of the invention

(1) to the low frequency part in front K-1 layer wavelet coefficient, and the low frequency in K layer wavelet coefficient and HFS carry out absolute value summation and process and obtain edge wavelet coefficient c _k

$c_k\left\{\begin{array}{cc}\underset{j=0}{\overset{\frac{T}{2^k}-1}{\mathrm{\Σ}}}\left|w_L(k,j)\right|& k=1,...,K-1\\ \underset{j=0}{\overset{\frac{L}{2^K}-1}{\mathrm{\Σ}}}\left|w_H(k,j)\right|& k=K\\ \underset{j=0}{\overset{\frac{L}{2^K}-1}{\mathrm{\Σ}}}\left|w_L(k,j)\right|& k=K+1\end{array}\right.$

Wherein L is the length of eeg signal data, and w (k, j) is j the wavelet coefficient that k layer decomposes

(2) the edge wavelet coefficient obtaining is normalized and is combined into the edge wavelet character x of K+1 dimension.1. this feature has and is distributed in (0,1) open interval, 2. adds and is 1 feature, and concrete methods of realizing is as follows:

$x_k=\frac{c_k}{\underset{k=1}{\overset{K+1}{\mathrm{\Σ}}}{c}_k},$ k＝1,...,K+1

x＝[x ₁,...，x _K+1] ^T

As follows according to the concrete steps of feature selection in a kind of brain wave recognition methods step 2 A based on edge wavelet character and super Di Li Cray mixed model of an embodiment of the invention:

Feature brain wave data is divided into positive sample x ⁺with negative sample x ^-, positive sample is brain thinking pattern to be trained, negative sample is other brain thinking patterns.The mean vector μ (i) and covariance matrix Σ (i) that calculate respectively positive negative sample in the feature of i road, utilize Fisher decision criteria to calculate the Fisher Ratio value of this road signal.The Fisher Ratio value of i road signal is defined as:

$\mathrm{FR}\left(i\right)=\frac{d^T[{\mathrm{\μ}}^{+}\left(i\right)-{\mathrm{\μ}}^{-}\left(i\right)][{\mathrm{\μ}}^{+}\left(i\right)-{\mathrm{\μ}}^{-}\left(i\right){]}^{T}d}{d^T[{\mathrm{\Σ}}^{+}\left(i\right)+{\mathrm{\Σ}}^{-}\left(i\right)]d}$

d＝[Σ ^-(i)+Σ ^-(i)] ^-1[μ ⁺(i)-μ ^-(i)]

Select there is maximum FR value m road signal as final training characteristics.And constitutive characteristic super vector

As follows according to model training concrete steps in a kind of brain wave recognition methods step 2 B based on edge wavelet character and super Di Li Cray mixed model of an embodiment of the invention:

(1) the m Road Edge wavelet character that obtains from feature selection step is separate and meet Dirichlet distribute, super vector x _supmeet super Di Li Cray probability density distribution:

$\mathrm{SDir}(x_{\sup };\mathrm{\α})=\underset{n=1}{\overset{m}{\mathrm{\Π}}}\frac{\mathrm{\Γ}\left(\underset{k=1}{\overset{K+1}{\mathrm{\Σ}}}{\mathrm{\α}}_{n,k}\right)}{\underset{k=1}{\overset{K+1}{\mathrm{\Π}}}\mathrm{\Γ}\left({\mathrm{\α}}_{n,k}\right)}\underset{k=1}{\overset{K+1}{\mathrm{\Π}}}{\left(x_{n,k}\right)}^{{\mathrm{\α}}_{n,k}-1}$

(2) for brain wave edge wavelet character super vector sequence X=[x _sup(1) ..., x _sup(T)] can simulate its probability distribution with the super Di Li Cray mixed model (SDMM) that contains c composition.

$f\left(X\right)=\underset{t=1}{\overset{T}{\mathrm{\Π}}}\underset{c=1}{\overset{C}{\mathrm{\Σ}}}{\mathrm{\π}}_c\mathrm{SDir}(x_{\sup }\left(t\right);\mathrm{\α}\left(c\right))$

Weight factor wherein ${\mathrm{\π}}_c=\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\Σ}}}{\stackrel{\‾}{z}}_{\mathrm{tc}}=\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\Σ}}}\frac{{\mathrm{\π}}_c\mathrm{SDir}(x_{\sup }\left(t\right);\mathrm{\α}\left(c\right))}{{\mathrm{\π}}_c\mathrm{SDir}(x_{\sup }\left(t\right);\mathrm{\α}\left(c\right))}$

(3) computation model parameter, for c mixed components, parameter vector α _cbe divided into m subvector, the corresponding x of each parameter subvector _supin a subvector.So we can obtain all parameters by solution equation below:

Beneficial effect of the present invention is, in terms of existing technologies, the present invention's application edge discrete wavelet feature super vector is extracted as the feature of brain wave, by super Di Li Cray mixed distribution training pattern, provide again complete implementation system for application, result of the test has been verified high efficiency of the present invention, has very strong practicality.

Accompanying drawing explanation

Fig. 1 is the flow chart of steps of method provided by the invention;

Fig. 2 is edge wavelet character leaching process schematic diagram;

The specific embodiment

Below in conjunction with accompanying drawing, specific embodiments of the present invention is described in detail.

Fig. 1 is flow chart of the present invention, and wherein dotted line represents training department's minute flow process trend, and solid line represents identification division flow process trend, comprises the following steps:

The first step: characteristic extraction step, the N road eeg signal for the treatment of training carries out feature extraction

Step S1: eeg signal is carried out to wavelet transform and obtain discrete wavelet feature;

Step S2: discrete wavelet feature is processed and obtained edge wavelet character;

Second step: special feature selection step

Step S3: use Fisher decision criteria to carry out feature selection, choose m road and can reflect the feature of brain thinking pattern and the feature of choosing is combined to form to feature super vector.

The 3rd step: model training step

Step S4: use the distribution of super Di Li Cray mixed model simulation feature super vector, and solve the parameter in model;

The 4th step: identification step

Step S5: step S1 and step S2 that N road eeg signal to be identified is repeated in the first step generate edge wavelet character, input step S3 and step S4 train in the model obtaining, calculate the probit of each probabilistic model, and utilize maximum posteriori criterion, brain wave content is identified.

To be specifically described each step below:

Step S1 carries out wavelet transform to the eeg signal of input, obtains discrete wavelet feature, and conversion process as shown in Figure 2.In this patent, adopt the 2 rank western wavelet basiss of many shellfishes to decompose signal, its high and low pass filter coefficient is respectively:

H＝[-0.129,-0.224,0.837,-0.483] ^T

L＝[-0.129,0.224,0.837,0.483] ^T

Utilize wave filter successively to decompose and obtain discrete wavelet feature the low-frequency component of input signal.

Step S2 processes and obtains edge wavelet character the K layer small echo signal obtaining.Implementing procedure is as follows

(1), as shown in dash area in Fig. 2, the low-frequency component of K-1 layer before extracting, and the low frequency of K layer, radio-frequency component, carry out absolute value and ask.

$c_k\left\{\begin{array}{cc}\underset{j=0}{\overset{\frac{T}{2^k}-1}{\mathrm{\Σ}}}\left|w_L(k,j)\right|& k=1,...,K-1\\ \underset{j=0}{\overset{\frac{L}{2^K}-1}{\mathrm{\Σ}}}\left|w_H(k,j)\right|& k=K\\ \underset{j=0}{\overset{\frac{L}{2^K}-1}{\mathrm{\Σ}}}\left|w_L(k,j)\right|& k=K+1\end{array}\right.$

(2), by result normalization connection, obtain the edge wavelet character x of K+1 dimension.

$x_k=\frac{c_k}{\underset{k=1}{\overset{K+1}{\mathrm{\Σ}}}{c}_k},$ k＝1,...,K+1

x＝[x ₁,...,x _K+1] ^T

Step S3 utilizes Fisher decision criteria to carry out feature selection to the N Road Edge wavelet character obtaining, and chooses the feature that m road can be reflected brain thinking pattern.Performing step is as follows:

(1) feature of thinking pattern to be trained is set to positive sample, and other pattern features are set to negative sample, calculates respectively the mean vector μ of positive each circuit-switched data of negative sample ⁺(i), μ ^-(i) with covariance matrix Σ ⁺(i), Σ ^-(i)

(2) utilize Fisher decision criteria to calculate respectively the Fisher Ratio value of each road signal

$\mathrm{FR}\left(i\right)=\frac{d^T[{\mathrm{\μ}}^{+}\left(i\right)-{\mathrm{\μ}}^{-}\left(i\right)][{\mathrm{\μ}}^{+}\left(i\right)-{\mathrm{\μ}}^{-}\left(i\right){]}^{T}d}{d^T[{\mathrm{\Σ}}^{+}\left(i\right)+{\mathrm{\Σ}}^{-}\left(i\right)]d}$

d＝[Σ ⁺(i)+Σ ^-(i)] ^-1[μ ⁺(i)-μ ^-(i)]

Select there is maximum FR value m road signal as final training characteristics.And constitutive characteristic super vector

Step S4 is used the distribution of super Di Li Cray mixed model simulation feature super vector, and solves the parameter in model.Detailed step is:

(1) the m Road Edge wavelet character that obtains from feature selection step is separate and meet Dirichlet distribute, super vector x _supmeet super Di Li Cray probability density distribution:

$\mathrm{SDir}(x_{\sup };\mathrm{\α})=\underset{n=1}{\overset{m}{\mathrm{\Π}}}\frac{\mathrm{\Γ}\left(\underset{k=1}{\overset{K+1}{\mathrm{\Σ}}}{\mathrm{\α}}_{n,k}\right)}{\underset{k=1}{\overset{K+1}{\mathrm{\Π}}}\mathrm{\Γ}\left({\mathrm{\α}}_{n,k}\right)}\underset{k=1}{\overset{K+1}{\mathrm{\Π}}}{\left(x_{n,k}\right)}^{{\mathrm{\α}}_{n,k}-1}$

(2) for edge wavelet character super vector sequence X=[x _sup(1) ..., x _sup(T)] can simulate its probability distribution with the super Di Li Cray mixed model (SDMM) that contains c composition.

$f\left(X\right)=\underset{t=1}{\overset{T}{\mathrm{\Π}}}\underset{c=1}{\overset{C}{\mathrm{\Σ}}}{\mathrm{\π}}_c\mathrm{SDir}(x_{\sup }\left(t\right);\mathrm{\α}\left(c\right))$

Weight factor wherein ${\mathrm{\π}}_c=\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\Σ}}}{\stackrel{\‾}{z}}_{\mathrm{tc}}=\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\Σ}}}\frac{{\mathrm{\π}}_c\mathrm{SDir}(x_{\sup }\left(t\right);\mathrm{\α}\left(c\right))}{{\mathrm{\π}}_c\mathrm{SDir}(x_{\sup }\left(t\right);\mathrm{\α}\left(c\right))}$

(3) computation model parameter, for c mixed components, parameter vector α _cbe divided into m subvector, the corresponding x of each parameter subvector _supin a subvector.So we can obtain all parameters by solution equation below:

Step S5 has realized the identifying of brain wave, and idiographic flow is as follows

(1) the N road eeg signal of input is carried out to feature extraction by step S1, S2, obtain edge wavelet character

(2) according to the training result of step S3, carry out feature selection, and form super vector.

(3) in the probabilistic model of super vector input step S4 being trained, calculate probit,

(4) in conjunction with prior probability, calculate maximum a posteriori probability and obtain, the model that produces maximum a posteriori probability is recognition result.

Below to proposed, based on edge wavelet character and the brain wave recognition methods of super Di Li Cray mixed model and the specific embodiment of each module, set forth by reference to the accompanying drawings.By the description of above embodiment, one of ordinary skill in the art can clearly recognize that the mode that the present invention can add essential general hardware platform by software realizes, and can certainly realize by hardware, but the former is better embodiment.Understanding based on such, the part that technical scheme of the present invention contributes to prior art in essence in other words can embody with the form of computer software product, this software product is stored in a storage medium, comprises that some instructions are used so that the method described in each embodiment of one or more computer equipment execution the present invention.

According to thought of the present invention, all will change in specific embodiments and applications.In sum, this description should not be construed as limitation of the present invention.

Above-described embodiment of the present invention, does not form the restriction to invention protection domain.Any modification of doing within the spirit and principles in the present invention, be equal to and replace and improvement etc., within all should being included in protection scope of the present invention.

Claims (5)

1. the brain wave based on edge wavelet character (EEG) signal recognition system, is characterized in that, comprises the following steps:

One. characteristic extraction step:

A, wavelet transform step: to the M road eeg signal obtaining from sensor, utilize wavelet transformation to carry out the decomposition of K floor;

B, edge wavelet character extraction step: to the low frequency part in front K-1 layer wavelet coefficient, and low frequency and HFS in K layer wavelet coefficient are sued for peace, connection and normalized, generate the edge wavelet character of K+1 dimension;

Two, feature selection and model training step:

A, feature selection step:

To the M Road Edge wavelet character extracting, utilize Fisher criterion to carry out feature selection, pick out the feature that m road best embodies brain wave construction features;

B, model training step:

To obtain the feature super vector that K+1Wei edge, m road wavelet character carries out the dimension of combination producing m * (K+1); Use the distribution of super Di Li Cray mixed model (SDMM:super-Dirichlet Mixture Model) simulation edge wavelet character super vector, by gradient method, solve an equation and obtain the parameter alpha in model, finally obtain a series of models, the brain wave that each model is corresponding a type;

Three. identification step: extract after certain eeg signal data, the method of employing step 1 is extracted the feature of each passage, and carry out channel selecting and form super characteristic vector according to the training result of step 2 A, in the probabilistic model that super characteristic vector input step two B are trained, calculate probit, and in conjunction with prior probability distribution, adopt maximum a posteriori criterion to carry out the identification of brain wave type.

2. a kind of brain wave recognition system based on edge wavelet character as claimed in claim 1, is characterized in that, the wavelet basis that the wavelet transformation described in step 1 A adopts is the western wavelet basiss of the many shellfishes of second order, and its higher order filter coefficient is

h＝[-0.129,-0.224,0.837,-0.483] ^T

Lower order filter coefficient is

L＝[-0.129,0.224,0.837,0.483] ^T

Decompose number of plies K=4.

3. a kind of eeg signal recognition system based on edge wavelet character as claimed in claim 1, is characterized in that, the edge wavelet character extraction step described in step 1 B is as follows:

(1) to the low frequency part in front K-1 layer wavelet coefficient, and the wavelet coefficient of the low frequency in K layer wavelet coefficient and HFS carries out absolute value summation and processes and obtain edge wavelet coefficient;

(2) the edge wavelet coefficient obtaining is normalized and is combined into the edge wavelet character x of K+1 dimension.

4. a kind of eeg signal recognition system based on edge wavelet character as claimed in claim 1, is characterized in that, the feature selection step described in step 2 A is as follows:

For from by extracting the passage that can express brain wave feature the M Road Edge wavelet character of extraction, feature brain wave data is divided into positive sample x ⁺with negative sample x ^-, calculate respectively in the feature of i road mean vector and the covariance matrix of positive negative sample, utilize Fisher decision criteria to calculate the Fisher Ratio value of this road signal.The Fisher Ratio value of i road signal is defined as follows:

$\mathrm{FR}\left(i\right)=\frac{d^T[{\mathrm{\μ}}^{+}\left(i\right)-{\mathrm{\μ}}^{-}\left(i\right)][{\mathrm{\μ}}^{+}\left(i\right)-{\mathrm{\μ}}^{-}\left(i\right){]}^{T}d}{d^T[{\mathrm{\Σ}}^{+}\left(i\right)+{\mathrm{\Σ}}^{-}\left(i\right)]d}$

D bis-[Σ ⁺(i)+Σ ^-(i)] ^-1[μ ⁺(i)-μ ^-(i)]

Wherein μ (i) and Σ (i) are respectively mean vector and the covariance matrix of m road feature samples.Select there is maximum FR value m road signal as final training characteristics.And form super vector

5. a kind of eeg signal recognition system based on edge wavelet character as claimed in claim 1, is characterized in that, the model training step described in step 2 B is as follows:

(1) the m Road Edge wavelet character that obtains from feature selection step is separate and meet Dirichlet distribute, super vector x _supmeet super Di Li Cray probability density distribution:

$\mathrm{SDir}(x_{\sup };\mathrm{\α})=\underset{n=1}{\overset{m}{\mathrm{\Π}}}\frac{\mathrm{\Γ}\left(\underset{k=1}{\overset{K+1}{\mathrm{\Σ}}}{\mathrm{\α}}_{n,k}\right)}{\underset{k=1}{\overset{K+1}{\mathrm{\Π}}}\mathrm{\Γ}\left({\mathrm{\α}}_{n,k}\right)}\underset{k=1}{\overset{K+1}{\mathrm{\Π}}}{\left(x_{n,k}\right)}^{{\mathrm{\α}}_{n,k}-1};$

(2) for brain wave edge wavelet character super vector sequence X=[x _sup(1) ..., x _sup(T)] can simulate its probability distribution with the super Di Li Cray mixed model (SDMM) that contains c composition:

$f\left(X\right)=\underset{t=1}{\overset{T}{\mathrm{\Π}}}\underset{c=1}{\overset{C}{\mathrm{\Σ}}}{\mathrm{\π}}_c\mathrm{SDir}(x_{\sup }\left(t\right);\mathrm{\α}\left(c\right))$

Weight factor wherein ${\mathrm{\π}}_c=\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\Σ}}}{\stackrel{\‾}{z}}_{\mathrm{tc}}=\frac{1}{T}\underset{t=1}{\overset{T}{\mathrm{\Σ}}}\frac{{\mathrm{\π}}_c\mathrm{SDir}(x_{\sup }\left(t\right);\mathrm{\α}\left(c\right))}{{\mathrm{\π}}_c\mathrm{SDir}(x_{\sup }\left(t\right);\mathrm{\α}\left(c\right))};$

(3) computation model parameter, for c mixed components, parameter vector α _cbe divided into m subvector, the corresponding x of each parameter subvector _supin a subvector.So we can obtain all parameters by solution equation below: