CN102512160B - Electroencephalogram emotional state feature extraction method based on adaptive tracking in different frequency bands - Google Patents

Abstract

The invention belongs to emotional state recognition technology, and provides an electroencephalogram emotional state feature extraction method based on adaptive tracking in different frequency bands, which is a more objective emotional state recognition method and is a more objective evaluation method for therapy evaluation to mental diseases. The technical scheme includes that the electroencephalogram emotional state feature extraction method based on adaptive tracking in different frequency bands comprises following steps: data acquisition and preprocessing, time frequency feature extraction, adaptive frequency band selection and emotional state recognition; the specific data acquisition and preprocessing step comprises emotionally inducing a testee by the aid of emotional images and preprocessing acquired original electroencephalogram signals, and preprocessing comprises changing reference potential, reducing sampling, realizing bandpass filtering and removing electrooculogram; time-frequency change is realized for processed electroencephalogram signals by the aid of short time Fourier transform; and adaptive frequency band selection is realized by a frequency band variable adaptive tracking method, and an SVM (support vector machine) is used for realizing classification recognition for feature frequency bands. The electroencephalogram emotional state feature extraction method is mainly applied to recognizing emotional states.

Description

But the brain electricity emotional state feature extracting method based on the frequency-division section adaptive tracing

Technical field

The invention belongs to the emotional state recognition technology, but relate to a kind of brain electricity emotional state feature extracting method based on the frequency-division section adaptive tracing.

Background technology

1872, Darwin pointed out that in " human and animal's expression " book emotion is the adaptation instrument of senior phylogenetic scale, and from then on people have begun the research of emotion experiment with theory.Through more than 100 years, flourish to 20 th century later emotion researchs, and combine with researchs such as cognition, neuroscience, brain sciences; Its research means is also varied, as brain electricity (EEG), functional mri (fMRI), function near infrared imaging (fNIRI) etc.

Along with the development of society, each age, each neighborhood people's emotional disturbance is more and more, more and more serious, and the various disease incidences relevant to emotion are more and more higher, as depression, manic disorder, anxiety neurosis, obsession, affective disorder etc.2010, according to World Health Organization's statistics, global depression rate was about 11%, and approximately there are 3.4 hundred million patients with depression and number in the whole world in continuous growth.Current depression has become the fourth-largest disease in the world, and expecting the year two thousand twenty may be only second to the second largest illness of the cardiopathic mankind with becoming, and depression will become the 21 century mankind's main killer.This has had a strong impact on people's quality of life, and the while also provides powerful power for the development of emotion research.Emotion research is broad development abroad.1985 set up in the world " international emotion EASD ".The process of its development has related to the various aspects of human society and people's mental life, and from basic theories, methodology and content, from simple emotion kind and effect thereof, emotion and cognitive relation are to complicated socialization's emotion.IAPS has been used in the research of relevant emotional problem since coming out widely, such as the relation of the cognitive activities such as physiological mechanism, mood regulation, emotion and the attention of emotion, memory etc.At present, lot of domestic and international seminar is carrying out based on the vision induced research of the Emotional Picture of IAPS, many achievements in research have also been obtained, these achievements in research are different according to signal acquisition method, can be divided into the combination of emotional experience self-report method, brain imaging technique, biofeedback therapy and above the whole bag of tricks.

In prior art, mostly there are the deficiencies such as technical sophistication, identification poor accuracy.

Summary of the invention

The present invention is intended to solution and overcomes the deficiencies in the prior art, proposes a kind of more objectively emotional state recognition methods, and an objective appraisal method more can be provided for the treatment evaluation of mental illness (as depression).For achieving the above object, the technical scheme that the present invention takes is, but the brain electricity emotional state feature extracting method based on the frequency-division section adaptive tracing comprises the steps: data acquisition and pretreatment, time-frequency characteristics extraction, self adaptation Frequency Band Selection and emotional state identification;

Data acquisition and pretreatment concrete steps comprise: select joyful degree scope to be divided into the Emotional Picture of eight grades, higher grade of the Emotional Picture emotion that induces is more positive, lower grade of the Emotional Picture emotion that induces is more passive, utilizing Emotional Picture to carry out emotion to the experimenter brings out, and record experimenter's EEG signals, the original EEG signals that gathers is carried out pretreatment, and pretreatment comprises and changes reference potential, down-sampled, bandpass filtering, four steps of removal eye electricity

It is to utilize short time discrete Fourier transform to carry out time-frequency to the complete EEG signals of pretreatment to change that time-frequency characteristics extracts;

But self adaptation Frequency Band Selection and emotional state identification refer to utilize the frequency-division section adaptive tracking method, but different tested each frequency-division section that lead are discussed respectively, obtaining each leads and has the frequency range of separability most, then use SVM to carry out Classification and Identification to the feature of extracting, thereby identification emotional state, SVM is the abbreviation of Support Vector Machine, represents support vector machine.

Utilizing short time discrete Fourier transform to carry out time-frequency to the complete EEG signals of pretreatment changes and to be:

1. utilizing short time discrete Fourier transform to carry out the time-frequency variation to the complete EEG signals of pretreatment is at first to use the observation window W (t) of a finite width that signal x (t) is observed, then the signal after windowing is carried out that Fourier transform obtains, is specially:

$\mathrm{STFT}(t,\mathrm{\ω})=\underset{-\∞}{\overset{+\∞}{\∫}}x\left(\mathrm{\τ}\right)W^{*}(\mathrm{\τ}-t)e^{-\mathrm{j\ω\τ}}\mathrm{d\τ}---\left(1\right)$

Here ω is angular frequency, W ^*(τ-t) is the W (complex conjugate function of τ-t); The observation window of limited value length along the time shaft translation, is obtained the time dependent information of spectrum distribution of signal on the time-frequency plane of two dimension, obtain the two-dimentional time-frequency matrix I of EEG signals _n(f, t);

2. calculate the fisher ratio, weighing in class with it is to be capacity volume variance between different emotion grades between same emotion grade and class:

$S_W(f,t)={\mathrm{\Σ}}_{k=1}^C{\mathrm{\Σ}}_{n=1}^{n_k}{(I_n(f,t)-m_k(f,t))(I_n(f,t)-m_k(f,t))}^T---\left(2\right)$

$S_B(f,t)={\mathrm{\Σ}}_{k=1}^C{n}_k(m(f,t)-m_k(f,t)){(m(f,t)-m_k(f,t))}^T---\left(3\right)$

$F_R(f,t)=\frac{S_B(f,t)}{S_W(f,t)}---\left(4\right)$

Wherein, S _W, S _B, m _k, m and F _RTwo-dimensional matrix, S _W(f, t), S _B(f, t) represent respectively in class and class between difference, m _k(f, t) is the frequent degree of the mean time of k class, and m (f, t) is the frequent degree of the mean time of all classes, and C represents classification number, n _kIt is the sample number of k apoplexy due to endogenous wind; T represents transposition;

3. DW (f) can be tried to achieve by the Fisher ratio:

$\mathrm{DW}\left(f\right)={\mathrm{\Σ}}_{t=1}^t{F}_R(f,t)$

Wherein, t represents the time period when STFT calculates;

4. after obtaining DW (f), calculate DFC by wave band iteration selection method, iterations need to equal the band number of acquisition; DFC is the abbreviation of Discriminative Frequency Components, can calculate with Step1 cited below the frequency range of tool separability to the Step5 five steps, then the weight DW (f) under will tool separability frequency range is set to zero, calculate again separability and be positioned at deputy frequency range, constantly repeat this process until the band number that need to obtain:

Step1, at first determine to need selecteed frequency range be 1-45Hz, and the frequency window of slip changes with step-length 1Hz between 3-7Hz, obtains 5 different bandwidth parameters and is designated as BW _h, h=1,2,3,4,5;

Step2, when frequency window moves along the frequency axis of DW (f), according to formula (6) calculating energy distribution α:

$\mathrm{\α}(F_i,{\mathrm{BW}}_h)={\mathrm{\Σ}}_{f=F_i-{\mathrm{BW}}_h/2}^{f=F_i+{\mathrm{BW}}_h/2}\mathrm{DW}\left(f\right)---\left(6\right)$

F wherein _iRepresent the mid frequency of i frequency range when frequency window moves along frequency axis

Step3, according to ceiling capacity distribution α, at all F _iMiddle selection is best

$F_h^{\mathrm{opt}}=\arg \underset{F_i}{\max }\mathrm{\α}(F_i,{\mathrm{BW}}_h)---\left(7\right)$

To each BW _hAll to obtain one

Therefore, each h, all corresponding best mid frequency

And optimum capacity distributes

Step4, for each BW relatively _hResolution capability, calculate

Relative variation, utilize following formula to calculate δ _h:

${\mathrm{\δ}}_h=\frac{{\mathrm{\α}}_h^{\mathrm{opt}}-{\mathrm{\α}}_{h-1}^{\mathrm{opt}}}{{\mathrm{\α}}_h^{\mathrm{opt}}}\×100\%---\left(8\right)$

Step5, calculated δ _hAfterwards, establish a threshold value δ _min, each subjects uses same threshold value, relatively δ ₂With δ _minSize, if δ ₂Greater than δ _min, then compare δ ₃With δ _minSize, until find δ _hLess than δ _min, the position of h-1 is exactly the frequency range of tool separability.

The present invention has following technique effect:

The present invention utilizes short time discrete Fourier transform to extract time-frequency characteristics, can divide the frequency band adaptive tracking method to carry out the self adaptation Frequency Band Selection, the accuracy of emotional state identification can be effectively improved, then an objective appraisal method more can be provided for the treatment evaluation of mental illness.

Description of drawings

Fig. 1 the technology of the present invention flow chart.

Fig. 2 DFC calculation process.

The specific embodiment

Fig. 1 is the technology of the present invention flow chart.Comprise data acquisition and pretreatment, time-frequency characteristics extraction, self adaptation Frequency Band Selection and four major parts of emotional state identification.

The IAPS that uses in the present invention (International Affective Picture System) by many have the emotion arousal effects, international addressable, form and these colour pictures carried out the scoring of three dimensions by the colour picture of extensive semantic domain.It is by the state-run mental health of U.S. research center (National Institute of Mental Health, NIMH) emotion and attention institute (Emotion ﹠amp; Attention) go through the cover standardization emotional distress picture system that the several years works out.The present invention has selected joyful degree scope to be divided into the picture of eight grades from IAPS.Higher grade of the Emotional Picture emotion that induces is more positive, and lower grade of the Emotional Picture emotion that induces is more passive.Utilize these pictures to carry out emotion to the experimenter and bring out, and record its EEG signals.The original EEG signals that gathers is carried out pretreatment, comprise changing reference potential, down-sampled, bandpass filtering, four steps of removal eye electricity.

Utilizing short time discrete Fourier transform to carry out time-frequency to the complete EEG signals of pretreatment changes.The method of Short Time Fourier Transform is at first to use the observation window W (t) of a finite width that signal x (t) is observed, and then the signal after windowing is carried out that Fourier transform obtains,

$\mathrm{STFT}(t,\mathrm{\ω})=\underset{-\∞}{\overset{+\∞}{\∫}}x\left(\mathrm{\τ}\right)W^{*}(\mathrm{\τ}-t)e^{-\mathrm{j\ω\τ}}\mathrm{d\τ}---\left(1\right)$

Here ω is angular frequency, W ^*(τ-t) is the W (complex conjugate function of τ-t).

When the observation window of limited value length along the time shaft translation, can obtain the time dependent information of spectrum distribution of signal on the time-frequency plane of two dimension, can obtain so the two-dimentional time-frequency matrix In of EEG signals.Calculate afterwards fisher ratio F _RWeigh in class and the capacity volume variance between class.

$S_W(f,t)={\mathrm{\Σ}}_{k=1}^C{\mathrm{\Σ}}_{n=1}^{n_k}{(I_n(f,t)-m_k(f,t))(I_n(f,t)-m_k(f,t))}^T---\left(2\right)$

$S_B(f,t)={\mathrm{\Σ}}_{k=1}^C{n}_k(m(f,t)-m_k(f,t)){(m(f,t)-m_k(f,t))}^T---\left(3\right)$

$F_R(f,t)=\frac{S_B(f,t)}{S_W(f,t)}---\left(4\right)$

Wherein, S _W, S _B, m _k, m and F _RTwo-dimensional matrix, S _W(f, t), S _B(f, t) represent respectively in class and class between difference, m _k(f, t) is the frequent degree of the mean time of k class, and m (f, t) is the frequent degree of the mean time of all classes, and C represents classification number, n _kIt is the sample number of k apoplexy due to endogenous wind; T represents transposition;

Can be obtained by the Fisher ratio

$\mathrm{DW}\left(f\right)={\mathrm{\Σ}}_{t=1}^t{F}_R(f,t)---\left(5\right)$

Wherein, t represents the time period when STFT calculates.

Due to the frequency band of tool separability not necessarily of the most obvious feature band on frequency, and differently tested different best separability frequency bands may be arranged, utilization of the present invention can divide the frequency band adaptive tracking method, and different tested divided frequency bands are discussed respectively.Can divide frequency band adaptive tracing (Discriminative Frequency Components, DFC) is very important to extracting accurately feature, improving classification accuracy rate.The main process of DFC as shown in Figure 2.

Step1, at first determine to need selecteed frequency range be 1-45Hz, and the frequency window of slip changes with step-length 1Hz between 3-7Hz.Therefore, we can obtain 5 different bandwidth parameters and be designated as BW _j(j=1,2,3,4,5).

Step2, when frequency window moves along the frequency axis of DW (f), according to formula (6) calculating energy distribution α:

$\mathrm{\α}(F_i,{\mathrm{BW}}_h)={\mathrm{\Σ}}_{f=F_i-{\mathrm{BW}}_h/2}^{f=F_i+{\mathrm{BW}}_h/2}\mathrm{DW}\left(f\right)---\left(6\right)$

F wherein _iRepresent the mid frequency of i frequency range when frequency window moves along frequency axis

Step3, according to ceiling capacity distribution α, at all F _iMiddle selection is best

$F_h^{\mathrm{opt}}=\arg \underset{F_i}{\max }\mathrm{\α}(F_i,{\mathrm{BW}}_h)---\left(7\right)$

To each BW _hAll to obtain one

Therefore, each h, all corresponding best mid frequency

And optimum capacity distributes

Step4, for each BW relatively _hResolution capability, calculate

Relative variation, utilize following formula to calculate δ _h:

${\mathrm{\δ}}_h=\frac{{\mathrm{\α}}_h^{\mathrm{opt}}-{\mathrm{\α}}_{h-1}^{\mathrm{opt}}}{{\mathrm{\α}}_h^{\mathrm{opt}}}\×100\%---\left(8\right)$

Step5, calculated δ _hAfterwards, establish a threshold value δ _min, each subjects uses same threshold value, experiment showed, for different threshold values, as, 10%, 20%, 30%, 40% ..., threshold value is less, and this algorithm more levels off to and selects frequency window is the frequency band of 3HZ.Compare δ ₂With δ _minSize, if δ ₂Greater than δ _min, then compare δ ₃With δ _minSize, until find δ _hLess than δ _min, the position of h-1 is exactly the frequency range of tool separability.

Use SVM (Support Vector Machine) to carry out Classification and Identification to characteristic spectra, the emotional state discrimination of acquisition is up to 82.8%, than the accuracy of not passing through the adaptive tracing frequency range (79.3%) height.

The present invention is intended to propose a kind of new emotional state feature extracting method, emotional state by time frequency analysis means identification people, then an objective appraisal method more be can provide for the treatment evaluation of mental illness, perhaps information or feedback provided for mood regulation.This invention can improve the accuracy of emotional state identification effectively, and obtains considerable Social benefit and economic benefit.Optimum implementation intends adopting patent transfer, patent grant, technological cooperation or product development.

Claims (1)

1. but the brain electricity emotional state feature extracting method based on the frequency-division section adaptive tracing, is characterized in that, comprises the steps: data acquisition and pretreatment, time-frequency characteristics extraction, self adaptation Frequency Band Selection and emotional state identification;

Data acquisition and pretreatment concrete steps comprise: select joyful degree scope to be divided into the Emotional Picture of eight grades, higher grade of the Emotional Picture emotion that induces is more positive, lower grade of the Emotional Picture emotion that induces is more passive, utilizing Emotional Picture to carry out emotion to the experimenter brings out, and record experimenter's EEG signals, the original EEG signals that gathers is carried out pretreatment, and pretreatment comprises and changes reference potential, down-sampled, bandpass filtering, four steps of removal eye electricity

It is to utilize short time discrete Fourier transform to carry out time-frequency to the complete EEG signals of pretreatment to change that time-frequency characteristics extracts;

But self adaptation Frequency Band Selection and emotional state identification refer to utilize the frequency-division section adaptive tracking method, but different tested each frequency-division section that lead are discussed respectively, obtaining each leads and has the frequency range of separability most, then use SVM to carry out Classification and Identification to the feature of extracting, thereby identification emotional state, SVM is the abbreviation of Support Vector Machine, represents support vector machine;

Utilizing short time discrete Fourier transform to carry out time-frequency to the complete EEG signals of pretreatment changes and to be:

1. utilizing short time discrete Fourier transform to carry out the time-frequency variation to the complete EEG signals of pretreatment is at first to use the observation window W (t) of a finite width that signal x (t) is observed, then the signal after windowing is carried out that Fourier transform obtains, is specially:

$\mathrm{STFT}(t,\mathrm{\ω})=\underset{-\∞}{\overset{+\∞}{\∫}}x\left(\mathrm{\τ}\right)W^{*}(\mathrm{\τ}-t)e^{-\mathrm{j\ωt}}\mathrm{d\τ}---\left(1\right)$

Here ω is angular frequency, and wherein, t represents the time period when STFT calculates, W ^*(τ-t) is the W (complex conjugate function of τ-t); The observation window of limited value length along the time shaft translation, is obtained the time dependent information of spectrum distribution of signal on the time-frequency plane of two dimension, obtain the two-dimentional time-frequency matrix I of EEG signals _n(f, t);

2. calculate the fisher ratio, weighing in class with it is to be capacity volume variance between different emotion grades between same emotion grade and class:

$S_W(f,t)={\mathrm{\Σ}}_{k=1}^C{\mathrm{\Σ}}_{n=1}^{n_k}{(I_n(f,t)-m_k(f,t))(I_n(f,t)-m_k(f,t))}^T---\left(2\right)$

$S_B(f,t)={\mathrm{\Σ}}_{k=1}^C{n}_k{(m(f,t)-m_k(f,t))(m(f,t)-m_k(f,t))}^T---\left(3\right)$

$F_R(f,t)=\frac{S_B(f,t)}{S_W(f,t)}---\left(4\right)$

Wherein, S _W, S _B, m _k, m and F _RTwo-dimensional matrix, S _W(f, t), S _B(f,, t) represent respectively in class and class between difference, m _k(f, t) is the frequent degree of the mean time of k class, and m (f, t) is the frequent degree of the mean time of all classes, and C represents classification number, n _kIt is the sample number of k apoplexy due to endogenous wind; T represents transposition;

3. DW (f) can be tried to achieve by the Fisher ratio:

$\mathrm{DW}\left(f\right)={\mathrm{\Σ}}_{t=1}^t{F}_R(f,t);$

4. after obtaining DW (f), calculate DFC by wave band iteration selection method, iterations need to equal the band number of acquisition;

DFC is the abbreviation of Discriminative Frequency Components, can calculate with Step1 cited below the frequency range of tool separability to the Step5 five steps, then the weight DW (f) under will tool separability frequency range is set to zero, calculate again separability and be positioned at deputy frequency range, constantly repeat this process until the band number that need to obtain:

Step1, at first determine to need selecteed frequency range be 1-45Hz, and the frequency window of slip changes with step-length 1Hz between 3-7Hz, obtains 5 different bandwidth parameters and is designated as BW _h, h=1,2,3,4,5;

Step2, when frequency window moves along the frequency axis of DW (f), according to formula (6) calculating energy distribution α:

$\mathrm{\α}(F_i,{\mathrm{BW}}_h)={\mathrm{\Σ}}_{f=F_i{-}^{{\mathrm{BW}}_h/2}}^{f=F_i{+}^{{\mathrm{BW}}_h/2}}\mathrm{DW}\left(f\right)---\left(6\right)$

F wherein _iRepresent the mid frequency of i frequency range when frequency window moves along frequency axis;

Step3, according to ceiling capacity distribution α, at all F _iMiddle selection is best

$F_h^{\mathrm{opt}}=\arg \underset{F_i}{\max }\mathrm{\α}(F_i,{\mathrm{BW}}_h)---\left(7\right)$

To each BW _hAll to obtain one

Therefore, each h, all corresponding best mid frequency

And optimum capacity distributes

Step4, for each BW relatively _hResolution capability, calculate

Relative variation, utilize following formula to calculate δ _h:

${\mathrm{\δ}}_h=\frac{{\mathrm{\α}}_h^{\mathrm{opt}}-{\mathrm{\α}}_{h-1}^{\mathrm{opt}}}{{\mathrm{\α}}_h^{\mathrm{opt}}}\×100\%---\left(8\right)$

Step5, calculated δ _hAfterwards, establish a threshold value δ _min, each subjects uses same threshold value, relatively δ ₂With δ _minSize, if δ ₂Greater than δ _min, then compare δ ₃With δ _minSize, until find δ _hLess than δ _min, the position of h-1 is exactly the frequency range of tool separability.

Images