CN106419911A - Emotional detection method based on brain electric wave analysis - Google Patents

Abstract

The invention discloses an emotion detection method based on brain electric wave analysis. The method includes the following steps: firstly, extracting features by using the original electroencephalo-graph data, wherein the features include fractal dimension feature, energy feature, statistical feature and high-order cross feature; evaluating the feature parameters of the extracted features by an intraclass correlation coefficient-like method so as to obtain the most stable feature parameter; allowing a support vector machine to use the obtained feature parameters to train a classification model; finally, conducting the real-time detection of the emotion with allowing the trained classification model. The emotion detection method has the remarkable advantages that the most stable feature parameter can be obtained by the intraclass correlation coefficient-like method, and the stable and accurate classification model is thus successfully trained; compared with the traditional method, there is no need to re-train the classification model, the operation is simpler and the classification accuracy is higher.

Description

Emotion detection method based on brain wave analysis

Technical Field

The invention relates to the technical field of digital signal processing, in particular to an emotion detection method based on brain wave analysis.

Background

The brain wave is formed by summing up the postsynaptic potentials generated synchronously by a large number of neurons when the brain is active. It records the electrical wave changes during brain activity, which is a general reflection of the electrophysiological activity of brain neurons on the surface of the cerebral cortex or scalp. With the advancement of science and technology, the application of brain waves is also developing from the medical field to the engineering application field. At present, emotion detection methods based on electroencephalogram analysis are highly concerned.

In recent years, in the course of studying emotion detection methods based on brain wave analysis, researchers have applied different brain wave features and corresponding classifiers. Calibo et al apply energy signatures and classify in conjunction with neural networks. Lin et al apply differential asymmetric energy signatures and classify in conjunction with a support vector machine.

Although the above methods have shown some accuracy in each particular test, they have significant disadvantages. The method needs to retrain the classification model when testing each target, and has complex operation process and low detection precision.

Therefore, a method with simple operation process and high classification precision is needed.

Disclosure of Invention

Aiming at the defects of the prior art, the invention aims to provide an emotion detection method based on brain wave analysis, which obtains the most stable characteristic parameters by utilizing an intra-class relation method so as to realize real-time stable and accurate detection of emotion.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows:

the emotion detection method based on brain wave analysis is characterized by comprising the following steps of:

step 1: performing feature extraction on the acquired brain wave sample signal;

step 2: evaluating the characteristic parameters of the extracted characteristics by adopting an intra-class relation numerical method to obtain the most stable characteristic parameters;

and step 3: training a classification model through a support vector machine according to the most stable characteristic parameters obtained in the step 2;

and 4, step 4: and (4) performing real-time emotion detection by adopting the classification model trained in the step (3).

The further technical scheme is that the features extracted in the step 1 are fractal dimension features, energy features, statistical features and high-order cross features respectively.

The method further adopts the technical scheme that the fractal dimension characteristics are extracted after being processed by a Higuchi algorithm, and the calculation formula is as follows:

wherein FD is a value of the fractal dimension,<L_n(k)>is composed ofDeviation mean value L of_n(k) The average number of (a) is,the new time-series samples of the brain wave sample signal x (n) are n ∈ (1-k) as the initial time, and k as the interval time.

The further technical scheme is that the energy characteristics are obtained by processing the brain wave signals by using discrete Fourier transform, and the processing formula is as follows:

wherein,is the energy characteristic of the brain wave sample signal, X (e)^jω) Is a frequency spectrum of the brain wave sample signal x (n),n is the number of input samples, N ∈ (1-N).

The further technical proposal is that the statistical characteristics are respectively mean values mu of brain wave signals_XStandard deviation σ_XMean of absolute values of first deviation_XMean value of absolute value of first deviation of standardized brain wave signalSecond deviation ofMean value gamma of the pair values_XAnd the mean value of the absolute value of the second deviation of the normalized brain wave signalThe calculation formula is as follows:

wherein, X (N), X (N +1) and X (N +2) are three different sequences of brain wave sample signals, N is the number of input samples, and N belongs to (1-N).

The further technical scheme is that the high-order cross features are obtained by processing the weighted brain wave sample signals through a filter, and the calculation formula is as follows:

wherein D is_lIs a cross feature of order l, X_n(l)、X_n-1(l) Respectively, two characteristic functions under the order of l, wherein l is the order of the cross characteristic, and N ∈ (1-N).

The further technical scheme is that the evaluation model of the intra-class correlation coefficient method is as follows:

wherein, MS_BMean variance between classes, MS, representing characteristic parameters of each group_WThe mean variance within the class of each set of characteristic parameters is represented, and m represents the number of characteristic parameters.

In view of the urgent need of a method which has better stability and can carry out real-time emotion detection in the prior art, the scheme provides an emotion detection method based on brain wave analysis, and the method firstly utilizes original electroencephalogram data to extract features and respectively extracts fractal dimension features, energy features, statistical features and high-order cross features; then evaluating the characteristic parameters of the extracted features by utilizing an intra-class relation numerical method so as to obtain the most stable characteristic parameters; then, the obtained characteristic parameters are used for training a classification model through a support vector machine; and finally, realizing real-time emotion detection by using the trained classification model. When emotion detection is performed on a certain target, the classification model does not need to be retrained, and the classification model stored in the step 3 only needs to be downloaded and applied to emotion classification of the target. Thus, real-time detection of the mood of the target is achieved.

The invention has the following remarkable effects: the most stable characteristic parameters are obtained by utilizing an intra-class relation numerical method, and a stable and accurate classification model is successfully trained; compared with the traditional method, the method does not need to retrain the classification model, and is simpler to operate and higher in classification precision.

Drawings

FIG. 1 is a flow chart of the method of the present invention.

Detailed Description

The following provides a more detailed description of the embodiments and the operation of the present invention with reference to the accompanying drawings.

As shown in fig. 1, a method for emotion detection based on brain wave analysis includes the following steps:

step 1: extracting characteristics of the acquired brain wave sample signals, and respectively extracting fractal dimension characteristics, energy characteristics, statistical characteristics and high-order cross characteristics, wherein the specific operations are as follows:

in this example, the fractal dimension characteristics are obtained by processing the brain wave signals by the Higuchi algorithm,

firstly, the acquired brain wave sample signal is subjected to interval sampling processing according to the following formula Thereby obtaining new time series samplesFor each new time series sampleCalculating the deviation mean value L_n(k) The calculation formula is as follows:and according to the formula<L_n(k)>∝k^-FDCalculating deviationMean difference L_n(k) From which the value of the fractal dimension can be calculatedWhere n ∈ (1 to k) is the initial time and k is the interval time.

For the energy characteristics of the brain wave sample signals, the brain wave signals are processed using discrete fourier transform as follows:

the energy characteristic may be expressed as

Wherein,n is the number of input samples, N ∈ (1-N).

For the statistical characteristics of the brain waves, the mean value mu of the brain wave signals is calculated respectively_XStandard deviation σ_XMean of absolute values of first deviation_XMean value of absolute value of first deviation of standardized brain wave signalMean value gamma of absolute values of second deviation_XAnd the mean value of the absolute value of the second deviation of the normalized brain wave signalThe calculation formula is as follows:

wherein X (n), X (n +1), and X (n +2) are three different sequences of brain wave sample signals,three normalized electroencephalogram signals are provided, N is the number of input samples, N ∈ (1 to N).

For the high-order cross feature of the brain wave, the brain wave sample signal after applying the filter processing weight is extracted, and the filter function formula is expressed as follows:

wherein y (n) ═ x (n) — μ_XThen the order i cross signature is expressed as follows:

wherein, X_n(l)、X_n-1(l) Respectively two characteristic functions under the order of l, and the calculation formula isY (n) is the weighting process for the nth electroencephalogram sample signal X (n), mu_XThe mean value of the brain wave sample signals, l, is the order of the cross features, N ∈ (1 to N).

Step 2: evaluating the characteristic parameters of the extracted characteristics by adopting an intra-class relation numerical method to obtain the most stable characteristic parameters, wherein the evaluation model is as follows:

wherein, MS_BMean variance between classes, MS, representing characteristic parameters of each group_WThe mean variance within the class of each set of characteristic parameters is represented, and m represents the number of characteristic parameters.

And step 3: training a classification model through a support vector machine according to the most stable characteristic parameters obtained in the step 2;

and 4, step 4: and (4) carrying out real-time emotion detection by using the classification model trained in the step (3).

In the specific implementation process, when emotion detection is performed on a certain target, the classification model does not need to be retrained, and only the classification model stored in the step 3 needs to be downloaded and applied to emotion classification of the target. Thus, real-time detection of the mood of the target is achieved.

The invention has the following remarkable effects: the most stable characteristic parameters are obtained by utilizing an intra-class relation numerical method, and a stable and accurate classification model is successfully trained; compared with the traditional method, the method is simpler to operate and higher in classification precision.

In this embodiment, four emotions are identified, and the experimental results are shown in table 1, where F1 represents a traditional emotion classification method, and F2 represents a combined feature (fractal dimension feature, 5 statistical features, 1-order high-order cross feature, and 4 charged energy features) emotion classification method proposed by the present solution, which can be seen easily that the present solution improves the accuracy of emotion identification.

TABLE 1

Claims (7)

1. A mood detection method based on brain wave analysis is characterized by comprising the following steps:

step 1: performing feature extraction on the acquired brain wave sample signal;

step 2: evaluating the characteristic parameters of the extracted characteristics by adopting an intra-class relation numerical method to obtain the most stable characteristic parameters;

and step 3: training a classification model through a support vector machine according to the most stable characteristic parameters obtained in the step 2;

and 4, step 4: and (4) performing real-time emotion detection by adopting the classification model trained in the step (3).

2. The emotion detection method based on electroencephalogram analysis as claimed in claim 1, wherein: the features extracted in step 1 are fractal dimension features, energy features, statistical features and high-order cross features respectively.

3. The brain wave analysis-based emotion detection method according to claim 2, wherein: the fractal dimension characteristics are extracted after being processed by a Higuchi algorithm, and the calculation formula is as follows:

$FD=-\underset{k\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{log<L_n\left(k\right)>}{\log k},$

wherein FD is a value of the fractal dimension,<L_n(k)>is composed ofDeviation mean value L of_n(k) The average number of (a) is,the new time-series samples of the brain wave sample signal x (n) are n ∈ (1-k) as the initial time, and k as the interval time.

4. The brain wave analysis-based emotion detection method according to claim 2, wherein: the energy characteristics are obtained by processing brain wave sample signals by using discrete Fourier transform, and the processing formula is as follows:

${\hat{S}}_{NX\left(\mathrm{\&omega;}\right)}=\frac{1}{N}|X\left(e^{j\mathrm{\&omega;}}\right){|}^2,$

wherein,is the energy characteristic of the brain wave sample signal, X (e)^jω) Is a frequency spectrum of the brain wave sample signal x (n),n is the number of input samples, N ∈ (1-N).

5. The brain wave analysis-based emotion detection method according to claim 2, wherein: the statistical characteristics are mean values mu of brain wave signals_XStandard deviation σ_XMean of absolute values of first deviation_XMean value of absolute value of first deviation of standardized brain wave signalMean value gamma of absolute values of second deviation_XAnd the mean value of the absolute value of the second deviation of the normalized brain wave signalThe calculation formula is as follows:

${\mathrm{\&mu;}}_X=\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}X\left(n\right),$

${\mathrm{\&sigma;}}_X=\sqrt{\frac{1}{N}{\mathrm{\&Sigma;}}_{n=1}^N{\left(X\right(n)-{\mathrm{\&mu;}}_X)}^2},$

${\mathrm{\&delta;}}_X=\frac{1}{N-1}\underset{n=1}{\overset{N-1}{\&Sigma;}}|X(n+1)-X\left(n\right)|,$

$\stackrel{\&OverBar;}{{\mathrm{\&delta;}}_X}=\frac{{\mathrm{\&delta;}}_X}{{\mathrm{\&sigma;}}_X},$

${\mathrm{\&gamma;}}_X=\frac{1}{N-2}\underset{n=1}{\overset{N-2}{\&Sigma;}}|X(n+2)-X\left(n\right)|,$

$\stackrel{\&OverBar;}{{\mathrm{\&gamma;}}_X}=\frac{{\mathrm{\&gamma;}}_X}{{\mathrm{\&sigma;}}_X},$

wherein, X (N), X (N +1) and X (N +2) are three different sequences of a brain wave sample signal, N is the number of input samples, and N belongs to (1-N).

6. The brain wave analysis-based emotion detection method according to claim 2, wherein: the high-order cross feature is obtained by processing the weighted brain wave sample signal through a filter, and the calculation formula is as follows:

$D_l=\underset{n=2}{\overset{N}{\&Sigma;}}{\&lsqb;X_n\left(l\right)-X_{n-1}\left(l\right)\&rsqb;}^2,$

wherein D is_lIs a cross feature of order l, X_n(l)、X_n-1(l) Two characteristic functions at order l, i being the order of the cross characteristic, N ∈ (1-N), are respectively provided.

7. The emotion detection method based on electroencephalogram analysis as claimed in claim 1, wherein: the evaluation model of the intra-class relation method is as follows:

$ICC=\frac{{\mathrm{MS}}_B-{\mathrm{MS}}_W}{{\mathrm{MS}}_B+(m-1){\mathrm{MS}}_W},$

wherein, MS_BMean variance between classes, MS, representing characteristic parameters of each group_WThe mean variance within the class of each set of characteristic parameters is represented, and m represents the number of characteristic parameters.