CN109222966A - A kind of EEG signals sensibility classification method based on variation self-encoding encoder - Google Patents
A kind of EEG signals sensibility classification method based on variation self-encoding encoder Download PDFInfo
- Publication number
- CN109222966A CN109222966A CN201811173603.6A CN201811173603A CN109222966A CN 109222966 A CN109222966 A CN 109222966A CN 201811173603 A CN201811173603 A CN 201811173603A CN 109222966 A CN109222966 A CN 109222966A
- Authority
- CN
- China
- Prior art keywords
- eeg signals
- vae
- feature
- power density
- wave
- 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
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/369—Electroencephalography [EEG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/16—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
- A61B5/165—Evaluating the state of mind, e.g. depression, anxiety
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Abstract
The invention discloses a kind of methods of emotional semantic classification based on variation autoencoder network VAE.This method utilizes the EEG signals extracted, it is denoised first, filtering, the pretreatment such as segmentation, then the power density spectrum of EEG signals is extracted, obtained time-frequency feature is normalized, then it is put into seven layers of VAE neural network model and is trained as input, the number of input layer is 160, the number of hidden layer neuron is 100, the number of output layer neuron is 50, training label is divided by the size of V-A model, its feature is calculated, then obtained feature feeding classifier is classified, potential feature of the EEG signals by variation from after encoding effectively is utilized, improve the accuracy of emotional semantic classification.
Description
Technical field
The present invention relates to a kind of brain telecommunications for being based on variation self-encoding encoder VAE (Variational Auto-Encoder)
Number sensibility classification method, belongs to processing of biomedical signals field.
Background technique
Torsion free modules (Brain-Computer Interface, BCI) be biomedical engineering field research it
One.It provides a kind of effective method by detection EEG signals mode to control external equipment.At present had much at
The application of the BCI of function, such as spelling program and wheelchair controller.BCI can be used for controlling equipment by detection consciousness, may be used also
With the psychological condition for monitoring people, emotion recognition is one of important application.
Electroencephalogram (EEG) signal is the clue of the oscillation of the neuroelectricity as caused by the ionic current between brain neuron, can be with
The variation of affective state in the brain of people is directly reacted, however, since EEG signal-to-noise ratio (SNR) is lower, " manual " analysis brain telecommunications
Number generally also have more highly difficult.The deep learning method carried out recently in machine learning field allows to automate feature extraction
And selection, reduce manual intervention Feature Selection.In addition to other than image and voice field, deep learning method in recent years
It is widely used in processing biomedicine signals, such as electroencephalogram (EEG), electromyogram (EMG) and electrocardiogram (ECG).Currently, being directed to
The mood of EEG signal, which is classified, mainly following method: support vector machines (Supported Vector Machine, SVM),
K arest neighbors (K-Nearest Neighbor, KNN), Recurrent Fuzzy Neural Network (Recurrent Fuzzy Neural
Network, RFNN), deepness belief network (Deep Belief Network, DBN) etc..
The mood model of people can be characterized with potency and two principal dimensions of wake-up.Potency (valence) is that individual is right
Degree is detested in the attraction of specific matters, and the range that scores is from negatively to front.It is then a kind of physiological for waking up (arousal)
Psychological condition, it is from passively to awake or to stimulation the reaction of active.Mood awakening dimension (Valence-Arousal, V-
A) model is widely used in anthropopsychology's research and measurement, usually analyzes the mood EEG signals of people, general primary study delta
Wave, theta wave, alpha wave, five common frequency bands of beta wave and gamma wave estimate people using the feature of each frequency band
Cognition and emotional state.
For emotion recognition, an important problem is how to improve the accuracy rate of emotional semantic classification.Variation encodes certainly
Device (Variational Auto-Encoder, VAE) model can be fitted the distribution of available data, carry out data generation, pass through
Measurement reconstructed error and back-propagation network parameter can be good at training such network, can be more preferable using Reparameterization skill
Extraction EEG signals in potential information.In recent years, VAE is widely used in the analysis of voice, music signal,
The upper of EEG signals also has preliminary application, but carries out emotional semantic classification using VAE analysis EEG signals and do not further investigate also.
Summary of the invention
For conventional method when classifying to emotion the lower problem of accuracy rate, the invention proposes one kind based on become
Divide the EEG signals sensibility classification method of self-encoding encoder.This method considers the temporal aspect that EEG signals have, and utilizes life
Extraction further is merged with frequency domain character progress at entropy feature of the model variation self-encoding encoder VAE to EEG signals, obtains height
Grade feature, obtains final emotion recognition result using classifier.
Technical solution proposed by the present invention is as follows:
A kind of EEG signals sensibility classification method based on variation self-encoding encoder VAE, it is characterized in that: after using pretreatment
EEG signals, the extraction of the power density spectrum signature of progress EEG signals and be normalized first, then by VAE to returning
Feature after one change is trained, and obtains the potential feature of EEG signals, is finally sent into classifier and is classified, this method includes
Following steps:
(1) several EEG signals are acquired, artefact processing and bandpass filtering screening are carried out;
(2) extraction of power density spectrum signature: extracting the time domain of EEG signals, frequency domain character, then by pair
All EEG signals carry out Short Time Fourier Transform (short-time Fourier transform, STFT) in data set, obtain
To power density spectrum signature;
(3) the obtained power density spectrum signature of step (2) is normalized, then using normalized feature as defeated
Enter to be put into and be trained in seven layers of VAE neural network model, obtains the potential feature of EEG signals;
(4) potential feature obtained in step (3) is sent into support vector machines and carries out emotion recognition.
In seven layers of VAE neural network model, the number of input layer is 160, the number of hidden layer neuron
It is 100, output layer neuron number is 50.
In step (2) and its later EEG signals frequency spectrum used in step be taken respectively from delta wave, theta wave,
Five alpha wave, beta wave and gamma wave primary bands.
The potential characteristic Z of finally obtained EEG signals in step (3) is to carry out Reparameterization by Z=μ+ε × σ
It obtains, wherein μ is the mean value of EEG signals, and σ is the variance of EEG signals, and ε is white Gaussian noise.
When carrying out emotion recognition in step (4), extracts EEG signals potency V and wake up the label of A, according to V high A high, V high
A is low, the low A high of V, the low four kinds of situations of the low A of V, is divided into four classes and carries out emotion recognition.
The present invention is based on variation self-encoding encoder VAE models to classify to affective state, in step (3), the present invention
It can be good at training such network by measurement reconstructed error and back-propagation network parameter by VAE, utilize Reparameterization skill
Ingeniously, potential information preferably can be extracted using EEG signals, potential advanced spy possessed by EEG signals is effectively utilized
Sign, improves the accuracy of emotional semantic classification.
Detailed description of the invention
Fig. 1 is that the present invention is based on the module flow diagrams that EEG signals carry out emotional semantic classification;
Fig. 2 is that the present invention is based on the variation self-encoding encoder VAE model flow figures that EEG signals carry out emotional semantic classification;
Fig. 3 is the partitioning standards schematic diagram divided in the present invention for label.
Specific embodiment
As shown in Figure 1 and Figure 2, the emotional semantic classification of the variation self-encoding encoder VAE proposed according to the present invention based on EEG signals
One embodiment of method the following steps are included:
(1) EEG signals are acquired first, and we used DEAP data sets here.Data set is by data and label two
Class data composition.The structure of data is 40 × 40 × 8064 (stimulations × channel × sample), and channel used in this experiment is
The EEG signals in preceding 32 channels, data original frequency is 512HZ, frequency 128Hz after down-sampling, then by myoelectricity, eye electricity
The data processings such as signal removal, EEG signals filtering;
(2) extraction of power density spectrum signature: the time domain of data, frequency domain character are extracted, then by data
32 all EEG signals are concentrated to carry out Short Time Fourier Transform (short-time Fouriertransform, STFT), into
One step obtains power density spectrum, and every 160 data points are one group;
(3) 32 brain electric channel numbers of all 40 kinds of stimulations of 32 people the mark of label: have been selected in data data
According to the structure of label is 40*40*4, and the first two 40 is stimulus and channel respectively, and the latter 4 is potency, is waken up, and is dominated, happiness
It is good.In label data, the label of all 40 kinds of stimulations corresponding potency and wake-up of 32 people is extracted, and according to V-A mould
The combination of two of type bidimensional height, wherein V-A model is a kind of more commonly used index that emotion is measured, its abscissa
The potency to things emotion is represented, ordinate represents the awakening degree of emotion.It is generally used for the emotion of perception in experimentation
Conversion is for convenience of the data measured.With 4.5 it is threshold value in the present invention, is divided into four classes, division rule is as shown in figure 3, four
Four quadrants in a class corresponding diagram 3, that is to say, that extract EEG signals potency V and wake up the label of A, according to V high A high, V
High A is low, the low A high of V, the low four kinds of situations of the low A of V, is divided into four classes and carries out emotion recognition;
(4) (2) obtained time-frequency feature is normalized, is then put into normalized feature as input
It is trained in seven layers of VAE neural network model, the number of input layer is 160, and the number of hidden layer neuron is
100, the number of output layer neuron is 50, utilizes " distance " size and training process of control VAE mode input outlet chamber
In to approach the divergence distance (KLD) being originally inputted between introduced new distribution and true input, to carry out entire model
Optimization.After the Reparameterization in network mean value and variance remain, calculate obtained mean μ and variance with front
σ is calculated as follows: Z=μ+ε × σ, wherein ε is the noise variance of Gaussian distributed, has used Reparameterization here
(reparemerization) data obtained after Reparameterization are put into SVM and classify by skill, get the bid with (3)
Sign the classification accuracy for comparing to the end.
Effect of the invention can be further illustrated by experiment.
Experiment tests method proposed by the invention on standard DEAP eeg data collection, this experiment EEG signals are adopted
Sample rate 128Hz has carried out denoising to original EEG signals, and has been filtered to the signal after denoising, segmentation.For
The label of experimental data passes through the combination of two of V-A bidimensional height, is threshold value with 4.5, is divided into four classes and is calculated.
Table 1 compares algorithm proposed by the invention and is compared with the estimated result of existing algorithm, by final prediction
Probability is as it can be seen that after applying the present invention, effectively improve for the emotion predictablity rate of EEG signals.
Classification results under 1 distinct methods of table
Method name | Svm classifier result (four classes) |
SVM | 0.42 |
LDA | 0.22 |
SWLDA | 0.21 |
AE(SVM) | 0.361 |
SAE(SVM) | 0.43 |
SSAE(SVM) | 0.42 |
VAE(SVM) | 0.46 |
Claims (5)
1. a kind of EEG signals sensibility classification method based on variation self-encoding encoder VAE, it is characterized in that: after using pretreatment
EEG signals, the first extraction of the power density spectrum signature of progress EEG signals are simultaneously normalized, then by VAE to normalizing
Feature after change is trained, and obtains the potential feature of EEG signals, finally be sent into classifier classify, this method include with
Lower step:
(1) several EEG signals are acquired, artefact processing and bandpass filtering screening are carried out;
(2) extraction of power density spectrum signature: the time domain of EEG signals, frequency domain character are extracted, then by data
It concentrates all EEG signals to carry out Short Time Fourier Transform, obtains power density spectrum signature;
(3) the obtained power density spectrum signature of step (2) is normalized, is then put normalized feature as input
Enter into seven layers of VAE neural network model and be trained, obtains the potential feature of EEG signals;
(4) potential feature obtained in step (3) is sent into support vector machines and carries out emotion recognition.
2. the method as described in claim 1, it is characterised in that: in seven layers of VAE neural network model, input layer
Number be 160, the number of hidden layer neuron is 100, and output layer neuron number is 50.
3. the method as described in claim 1, it is characterised in that: in step (2) and its later EEG signals used in step
Frequency spectrum is taken respectively from five delta wave, theta wave, alpha wave, beta wave and gamma wave primary bands.
4. the method as described in claim 1, it is characterised in that: the potential feature of finally obtained EEG signals in step (3)
Z is to carry out Reparameterization by Z=μ+ε × σ to obtain, and wherein μ is the mean value of EEG signals, and σ is the variance of EEG signals, ε
It is white Gaussian noise.
5. the method as described in claim 1, it is characterised in that: when carrying out emotion recognition in step (4), extract EEG signals effect
Valence V and the label for waking up A, the low four kinds of situations of, V low A high low according to V high A high, V high A, the low A of V are divided into four classes and carry out emotion knowledge
Not.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811173603.6A CN109222966A (en) | 2018-10-09 | 2018-10-09 | A kind of EEG signals sensibility classification method based on variation self-encoding encoder |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811173603.6A CN109222966A (en) | 2018-10-09 | 2018-10-09 | A kind of EEG signals sensibility classification method based on variation self-encoding encoder |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109222966A true CN109222966A (en) | 2019-01-18 |
Family
ID=65054370
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811173603.6A Pending CN109222966A (en) | 2018-10-09 | 2018-10-09 | A kind of EEG signals sensibility classification method based on variation self-encoding encoder |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109222966A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109924977A (en) * | 2019-03-21 | 2019-06-25 | 西安交通大学 | A kind of surface electromyogram signal classification method based on CNN and LSTM |
CN110141258A (en) * | 2019-05-16 | 2019-08-20 | 深兰科技(上海)有限公司 | A kind of emotional state detection method, equipment and terminal |
CN111338482A (en) * | 2020-03-04 | 2020-06-26 | 太原理工大学 | Brain-controlled character spelling recognition method and system based on supervised self-encoding |
CN111368686A (en) * | 2020-02-27 | 2020-07-03 | 西安交通大学 | Electroencephalogram emotion classification method based on deep learning |
CN113032558A (en) * | 2021-03-11 | 2021-06-25 | 昆明理工大学 | Variational semi-supervised hundred-degree encyclopedia classification method fusing wiki knowledge |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106264460A (en) * | 2016-07-29 | 2017-01-04 | 北京好运到信息科技有限公司 | The coding/decoding method of cerebration multidimensional time-series signal based on self study and device |
CN107157477A (en) * | 2017-05-24 | 2017-09-15 | 上海交通大学 | EEG signals Feature Recognition System and method |
CN107361767A (en) * | 2017-08-04 | 2017-11-21 | 西南大学 | A kind of human emotion's potency classifying identification method using EEG signals |
US20180157923A1 (en) * | 2010-06-07 | 2018-06-07 | Affectiva, Inc. | Vehicular cognitive data collection using multiple devices |
CN108171176A (en) * | 2017-12-29 | 2018-06-15 | 中车工业研究院有限公司 | A kind of subway driver's emotion identification method and device based on deep learning |
CN108256629A (en) * | 2018-01-17 | 2018-07-06 | 厦门大学 | The unsupervised feature learning method of EEG signal based on convolutional network and own coding |
WO2018141061A1 (en) * | 2017-02-01 | 2018-08-09 | Cerebian Inc. | System and method for measuring perceptual experiences |
US20180268262A1 (en) * | 2017-03-15 | 2018-09-20 | Fuji Xerox Co., Ltd. | Information processing device and non-transitory computer readable medium |
-
2018
- 2018-10-09 CN CN201811173603.6A patent/CN109222966A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180157923A1 (en) * | 2010-06-07 | 2018-06-07 | Affectiva, Inc. | Vehicular cognitive data collection using multiple devices |
CN106264460A (en) * | 2016-07-29 | 2017-01-04 | 北京好运到信息科技有限公司 | The coding/decoding method of cerebration multidimensional time-series signal based on self study and device |
WO2018141061A1 (en) * | 2017-02-01 | 2018-08-09 | Cerebian Inc. | System and method for measuring perceptual experiences |
US20180268262A1 (en) * | 2017-03-15 | 2018-09-20 | Fuji Xerox Co., Ltd. | Information processing device and non-transitory computer readable medium |
CN107157477A (en) * | 2017-05-24 | 2017-09-15 | 上海交通大学 | EEG signals Feature Recognition System and method |
CN107361767A (en) * | 2017-08-04 | 2017-11-21 | 西南大学 | A kind of human emotion's potency classifying identification method using EEG signals |
CN108171176A (en) * | 2017-12-29 | 2018-06-15 | 中车工业研究院有限公司 | A kind of subway driver's emotion identification method and device based on deep learning |
CN108256629A (en) * | 2018-01-17 | 2018-07-06 | 厦门大学 | The unsupervised feature learning method of EEG signal based on convolutional network and own coding |
Non-Patent Citations (3)
Title |
---|
QINGYANG XU 等: "《The Difference Learning of Hidden Layer between Autoencoder and Variational Autoencoder》", 《2017 29TH CHINESE CONTROL AND DECISION CONFERENCE》 * |
SUWICHAJIRAYUCHAROENSAK 等: "《EEG-Based Emotion Recognition Using Deep Learning Network》", 《THE SCIENTIFIC WORLD JOURNAL》 * |
ZHONG YIN 等: "《Physiological Feature Based Emotion Recognition via an Ensemble Deep Autoencoder with Parsimonious Structure》", 《 IFAC PAPERS ONLINE》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109924977A (en) * | 2019-03-21 | 2019-06-25 | 西安交通大学 | A kind of surface electromyogram signal classification method based on CNN and LSTM |
CN110141258A (en) * | 2019-05-16 | 2019-08-20 | 深兰科技(上海)有限公司 | A kind of emotional state detection method, equipment and terminal |
CN111368686A (en) * | 2020-02-27 | 2020-07-03 | 西安交通大学 | Electroencephalogram emotion classification method based on deep learning |
CN111338482A (en) * | 2020-03-04 | 2020-06-26 | 太原理工大学 | Brain-controlled character spelling recognition method and system based on supervised self-encoding |
CN111338482B (en) * | 2020-03-04 | 2023-04-25 | 太原理工大学 | Brain-controlled character spelling recognition method and system based on supervision self-coding |
CN113032558A (en) * | 2021-03-11 | 2021-06-25 | 昆明理工大学 | Variational semi-supervised hundred-degree encyclopedia classification method fusing wiki knowledge |
CN113032558B (en) * | 2021-03-11 | 2023-08-29 | 昆明理工大学 | Variable semi-supervised hundred degree encyclopedia classification method integrating wiki knowledge |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Budak et al. | An effective hybrid model for EEG-based drowsiness detection | |
CN109222966A (en) | A kind of EEG signals sensibility classification method based on variation self-encoding encoder | |
CN106886792B (en) | Electroencephalogram emotion recognition method for constructing multi-classifier fusion model based on layering mechanism | |
Diykh et al. | EEG sleep stages classification based on time domain features and structural graph similarity | |
AU2020100027A4 (en) | Electroencephalogram-based negative emotion recognition method and system for aggressive behavior prediction | |
Kalaivani et al. | Analysis of EEG signal for the detection of brain abnormalities | |
CN110969108B (en) | Limb action recognition method based on autonomic motor imagery electroencephalogram | |
CN114052735B (en) | Deep field self-adaption-based electroencephalogram emotion recognition method and system | |
Kumari et al. | Seizure detection in EEG using time frequency analysis and SVM | |
Shahnaz et al. | Emotion recognition based on wavelet analysis of Empirical Mode Decomposed EEG signals responsive to music videos | |
CN110604565A (en) | Brain health screening method based on portable EEG equipment | |
CN110897648A (en) | Emotion recognition classification method based on electroencephalogram signal and LSTM neural network model | |
CN111184509A (en) | Emotion-induced electroencephalogram signal classification method based on transfer entropy | |
CN110960191A (en) | Epilepsia electroencephalogram signal classification method based on frequency spectrum energy diagram | |
Agarwal et al. | Classification of alcoholic and non-alcoholic EEG signals based on sliding-SSA and independent component analysis | |
Ramos-Aguilar et al. | Analysis of EEG signal processing techniques based on spectrograms | |
Anh-Dao et al. | A multistage system for automatic detection of epileptic spikes | |
Xu et al. | EEG emotion classification based on baseline strategy | |
Katyal et al. | EEG signal and video analysis based depression indication | |
Sulthan et al. | Emotion recognition using brain signals | |
Samal et al. | Ensemble median empirical mode decomposition for emotion recognition using EEG signal | |
Li et al. | Multi-modal emotion recognition based on deep learning of EEG and audio signals | |
Gnana Rajesh | Analysis of MFCC features for EEG signal classification | |
Chu et al. | An enhanced EEG microstate recognition framework based on deep neural networks: an application to Parkinson's disease | |
Pan et al. | Epileptic seizure detection with hybrid time-frequency EEG input: a deep learning approach |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190118 |