CN110353702A - A kind of emotion identification method and system based on shallow-layer convolutional neural networks - Google Patents

A kind of emotion identification method and system based on shallow-layer convolutional neural networks Download PDF

Info

Publication number
CN110353702A
CN110353702A CN201910591898.7A CN201910591898A CN110353702A CN 110353702 A CN110353702 A CN 110353702A CN 201910591898 A CN201910591898 A CN 201910591898A CN 110353702 A CN110353702 A CN 110353702A
Authority
CN
China
Prior art keywords
layer
shallow
convolutional neural
eeg signals
neural networks
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
Application number
CN201910591898.7A
Other languages
Chinese (zh)
Inventor
吴畏
曾柏泉
齐菲菲
俞祝良
顾正晖
李远清
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201910591898.7A priority Critical patent/CN110353702A/en
Publication of CN110353702A publication Critical patent/CN110353702A/en
Pending legal-status Critical Current

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/16—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
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • 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
    • A61B5/72—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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2134—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on separation criteria, e.g. independent component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Public Health (AREA)
  • Surgery (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Psychiatry (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Veterinary Medicine (AREA)
  • Signal Processing (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Physiology (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Biology (AREA)
  • Psychology (AREA)
  • Educational Technology (AREA)
  • Child & Adolescent Psychology (AREA)
  • Developmental Disabilities (AREA)
  • Mathematical Physics (AREA)
  • Fuzzy Systems (AREA)
  • Hospice & Palliative Care (AREA)
  • Social Psychology (AREA)
  • Image Analysis (AREA)

Abstract

The invention belongs to emotion recognition technical fields, are related to a kind of emotion identification method and system based on shallow-layer convolutional neural networks.It is pre-processed by EEG signals, designs shallow-layer convolutional neural networks according to FBCSP feature extracting method, be based on trained shallow-layer convolutional neural networks model, classify to pretreated EEG signals, obtain emotion recognition result.In conjunction with FBCSP algorithm effective to eeg signal classification and convolutional neural networks at present, and it is applied to emotion electroencephalogramrecognition recognition, the recognition accuracy of different moods can be significantly improved, and have better wide usage for different tested individuals.Using shallow-layer convolutional network, to treated, emotion EEG signals are classified, more preferable than traditional characteristic extracting method recognition effect, have good application prospect in emotion recognition research field.

Description

A kind of emotion identification method and system based on shallow-layer convolutional neural networks
Technical field
The invention belongs to emotion recognition technical fields, are related to a kind of emotion identification method based on shallow-layer convolutional neural networks And system.
Background technique
The brain of people is the highest part of nervous system, controls the behavior, thought words and deeds and happiness anger sorrow of the mankind It is happy, it is the part that structure is most complicated in human body.The secret explored and disclose cerebral function is always what scientists were chased Dream and target.In in the past few decades, brain science research constantly makes a breakthrough.In first world brain in 1999 In machine interface conference, brain-computer interface (BCI) this concept is formally proposed for the first time, is set by the brain in human or animal with the external world Communication road is established between standby, carries out information exchange, is directly the control instruction of computer by EEG Processing.
The research direction in the field BCI has much at present, achieves good research achievement, for example P300 event is mutually powered-down Position, imagination movement brain electricity, epileptic electroencephalogram (eeg) identification etc., mood electroencephalogramrecognition recognition is also the one of popular research direction BCI.Due to The smeared out boundary of mood, carrying out detection identification to mood using conventional method is still a huge challenge, and there is many Problem, and it is used for most of feature, expression, behavior, speech or even the heartbeat etc. of Emotion identification, it can be pseudo- easily It loads, this kind of external feature not necessarily represents the true emotional in the minds of people.External feature more than comparing, brain electricity The physiological signals such as signal can more truly reveal out the time of day of people, if we can be by EEG signals come to people's Mood is identified that result has more reliability than the result obtained by external feature.
Summary of the invention
In view of the deficiencies of the prior art, the present invention provides a kind of emotion identification method based on shallow-layer convolutional neural networks, It is pre-processed by EEG signals, designs shallow-layer convolutional neural networks according to FBCSP feature extracting method, be based on trained shallow-layer Convolutional neural networks model classifies to pretreated EEG signals, obtains emotion recognition result.
The present invention also provides a kind of emotion recognition systems based on shallow-layer convolutional neural networks.
Emotion identification method of the invention adopts the following technical scheme that realization:
A kind of emotion identification method based on shallow-layer convolutional neural networks, comprising:
EEG signals under S1, acquisition different emotions state, classify by different labels；
S2, all EEG signals are pre-processed, obtains the emotion of the high s/n ratio of several set time length EEG signals；
S3, pretreated emotion EEG signals are inputted into the shallow-layer convolution mind according to the building of FBCSP feature extracting method In network, a series of training are carried out, obtain shallow-layer convolutional neural networks optimal parameter, obtain trained shallow-layer convolutional Neural Network；
S4, emotion recognition knot is obtained to emotion eeg signal classification based on trained shallow-layer convolutional neural networks model Fruit.
Preferably, EEG signals, which pre-process, includes:
1., EEG signals sample separation, the EEG signals segment after being separated；
2., removal separation after EEG signals segment in bad sample and bad channel；
3., EEG signals segment bandpass filtering and down-sampled processing；
4., independent component analysis and principal component analysis；
5., using machine learning method independent element and the signal source that principal component analysis is isolated are judged whether it is Noise source, and noise source is removed.
Preferably, it carries out judging whether it is bad sample by following calculation formula:
Wherein: an,cFor the average value in c-th of channel of n-th of sample, mcAnd σcIt is then all period segments in the channel c respectively Mean value and standard deviation；
One threshold value t=3 is set, z is worked asn,cWhen > 3, illustrate amplitude and other periods of entire channel in the channel c of the sample Segment differs greatly, and can be judged as of poor quality；
For the same sample, if it exceeds 20% channel is considered to of poor quality, then entire sample will be moved directly It removes, sample number subtracts 1；If being no more than 20%, which can be retained, but bad channel can be removed and according to peripheral channel number According to being repaired.
Preferably, the detection in bad channel is judged by calculating the maximum correlation coefficient in each channel and other channels, phase It is as follows to close coefficient formulas:
X, Y is two signals, and r (X, Y) is the related coefficient of two signals, and Cov (X, Y) is covariance, Var [X] and Var [Y] is respectively the variance of X, Y；
If there is the related coefficient maximum value with other channels less than 0.4 in the epochs that certain channel is more than 2% Phenomenon, then the channel is judged as bad channel.
Preferably, bandpass filter frequency band is 0.1-49.5Hz.
Preferably, it further comprises the steps of:
Building shallow-layer convolutional neural networks are carried out according to FBCSP feature extracting method.
Further, shallow-layer convolutional neural networks are constructed according to FBCSP feature extracting method, and first two layers is convolution Layer carries out feature extraction by different convolution kernels；First convolutional layer is used to extract the temporal signatures of eeg data, is equivalent to Time domain filtering corresponds to the Filter banks step in FBCSP algorithm, for extracting the feature of different frequency；Second Convolutional layer is for carrying out space characteristics extraction, the spatial filtering step being equivalent in common space pattern algorithm, by data projection To another space；What and then two convolutional layers exported is a quadratic nonlinearity activation primitive and pond layer, and pond layer swashs The log-variance characteristic extraction step for being log activation primitive, corresponding in FBCSP of function selection living；The last layer is Full articulamentum adds softmax activation primitive, plays a part of classifier.
Preferably, shallow-layer convolutional neural networks include l0~l4Several layers of structures, in which:
l0:
The layer is input layer, and as input X by taking 1-sec EEG signals segment as an example, size is 62 × 100 matrixes；
l1:
For l1The output of t-th of convolution kernel of layer, is one 62 × 70 matrix,WithRespectively t-th The weight and offset of convolution kernel, i, j are respectively the transverse and longitudinal coordinate of output matrix；Due to l1Layer is set as 10 convolution kernels, therefore Export a(1)Size is 10 × 62 × 70；All convolution process all only carry out on a channel, and the time domain for only extracting data is special Sign；
l2:
a(2)=(BN (z(2)))2
For l2S-th of spatial filter output of layer, j represent j-th of neuron of filter output, and matrix is big Small is 1 × 70, therefore 1≤j≤70；WithThe weight and offset of respectively s-th convolution kernel,Indicating should Some specific value of convolution kernel two-dimensional matrix；l2Layer one shares 30 spatial filters, therefore exports z(2)Size is 30 × 70；l2 Convolution is carried out in Spatial Dimension, only extracts spatial feature；Subsequent convolutional layer input enters BN layers and square non-linear layer, obtains most Output a eventually(2)；
l3:
a(3)=g (m(3))
l3It is mean value pond layer, whereinFor l2The output of s-th of spatial filter of layer is after BN and quadratic nonlinearity As a result,ForMean value Chi Huahou output, size be 1 × 5；Subsequent m(3)By log activation primitive g (x)= Logmax (x, 10-6) obtain final output a(3)；l3The output a of layer(3)Size be 30 × 5；
l4:
d(4)=dropout (a(3), p)
f(4)=flatten (d(4))
l4It is full articulamentum, d(4)It is a(3)Output after carrying out dropout according to Probability p, j represent l4J-th nerve Member；Then 30 × 5 matrixes are shakeout to obtain f(4), size becomes 1 × 150；Be represent c classification output layer neuron it is defeated Out, it is calculated by softmax layers, according to the maximum classification of output layer value as final classification classification.
Emotion recognition system of the invention adopts the following technical scheme that realization:
A kind of emotion recognition system based on shallow-layer convolutional neural networks, comprising: eeg signal acquisition device, EEG signals Pretreatment unit, shallow-layer convolutional neural networks and thymencephalon electricity tranining database, eeg signal acquisition device and EEG signals are pre- Processing unit is connected, and shallow-layer convolutional neural networks are connected with EEG signals pretreatment unit and thymencephalon electricity tranining database It connects, in which:
Eeg signal acquisition device is usually wrapped for acquiring the EEG signals under the different emotions state as test set Include a series of facilities such as electrode cap, signal amplifier, PC machine；
EEG signals pretreatment unit, for handling the brain electricity original signal of eeg signal acquisition device acquisition, removal is not The noise signal needed improves EEG signals signal-to-noise ratio, obtains pretreated emotion EEG signals；
Pretreated emotion EEG signals are classified as classifier, obtain emotion by shallow-layer convolutional neural networks Recognition result；
Thymencephalon electricity tranining database, for providing training sample for shallow-layer convolutional neural networks, comprising sufficient amount of Training sample has been subjected to pretreatment.
Further, shallow-layer convolutional neural networks are constructed according to FBCSP feature extracting method, and first two layers is convolution Layer carries out feature extraction by different convolution kernels；First convolutional layer is used to extract the temporal signatures of eeg data, is equivalent to Time domain filtering corresponds to the Filter banks step in FBCSP algorithm, for extracting the feature of different frequency；Second Convolutional layer is for carrying out space characteristics extraction, the spatial filtering step being equivalent in common space pattern algorithm, by data projection To another space；What and then two convolutional layers exported is a quadratic nonlinearity activation primitive and pond layer, and pond layer swashs The log-variance characteristic extraction step for being log activation primitive, corresponding in FBCSP of function selection living；The last layer is Full articulamentum adds softmax activation primitive, plays a part of classifier.
Compared with prior art, the invention has the following beneficial effects:
(1) FBCSP algorithm effective to eeg signal classification and convolutional neural networks at present are combined, and are applied In emotion electroencephalogramrecognition recognition, the recognition accuracy of different moods can be significantly improved, and is had more for different tested individuals Good wide usage.
(2) activation primitive of log and square nonlinear function as convolutional network is used, not with general network structure Together, feature extraction effect similar with FBCSP can be reached.
(3) using shallow-layer convolutional network, to treated, emotion EEG signals are classified, than traditional characteristic extracting method Recognition effect is more preferable, has good application prospect in emotion recognition research field.
(4) emotion identification method of the present invention is huge in training sample, individual difference it is biggish across individual experiment in can play The advantage of deep learning, and effectively improve across emotion recognition accuracy rate in individual experiment, there is preferable application prospect.
Detailed description of the invention
Fig. 1 is emotion identification method flow chart in one embodiment of the invention；
Fig. 2 is EEG signals pretreatment process figure in one embodiment of the invention；
Fig. 3 is FBCSP feature extracting method flow chart in one embodiment of the invention；
Fig. 4 is that the improved shallow-layer convolutional neural networks frame of FBCSP feature extracting method is based in one embodiment of the invention Frame figure.
Specific embodiment
The present invention is described in further detail below by specific embodiment, but embodiments of the present invention are not It is limited to this.
A kind of emotion identification method based on shallow-layer convolutional neural networks, as shown in Figure 1, comprising:
EEG signals under S1, acquisition different emotions state, classify by different labels；
S2, all EEG signals are pre-processed, obtains the emotion of the high s/n ratio of several set time length EEG signals；
In the present embodiment, EEG signals pretreatment is as shown in Figure 2, comprising:
1., EEG signals sample separation, the EEG signals segment after being separated；
Emotion EEG signals to be processed are read, and are separated into the segment of several set time length.General point From the EEG signals segment for becoming 1 second or 4 seconds, each EEG signals segment is a sample.
2., removal separation after EEG signals segment in bad segment (also known as " bad sample ") and bad channel；
The segment of poor quality in EEG signals segment after separation and acquisition channel are removed, and pass through relevant method It is repaired.
Step each sample that 1. middle segmentation obtains can carry out judging whether it is bad sample by following calculation formula:
Wherein: an,cFor the average value in c-th of channel of n-th of sample, mcAnd σcIt is then all period segments in the channel c respectively Mean value and standard deviation.One threshold value t=3 is set, z is worked asn,cWhen > 3, illustrates the amplitude in the channel c of the sample and entirely lead to Other period segments of road differ greatly, and can be judged as of poor quality.For the same sample, if it exceeds 20% channel all by Think of poor quality, then entire sample will be removed directly, and sample number subtracts 1；If being no more than 20%, which can be retained, but It is that bad channel can be removed and be repaired according to peripheral channel data.
The detection in bad channel judges by calculating the maximum correlation coefficient in each channel and other channels, related coefficient meter It is as follows to calculate formula:
X, Y is two signals, and r (X, Y) is the related coefficient of two signals, and Cov (X, Y) is covariance, Var [X] and Var [Y] is respectively the variance of X, Y.
If all there is the related coefficient maximum value with other channels less than 0.4 in the period segment that certain channel is more than 2% The phenomenon that, then the channel is judged as bad channel.The signal data in entire channel can be then removed, and according to circumferential passages signal It carries out repairing filling data.All bad channels re-start an average reference after repairing, then again to bad channel into Row detection repeats above step until can't detect any one bad channel.
3., EEG signals segment bandpass filtering and down-sampled processing；
Emotion recognition does not need too high-frequency signal, it is therefore desirable to filter out extra make an uproar by a bandpass filtering Sound, bandpass filter filter out high-frequency noise and low frequency floats for screening to the EEG signals segment of different frequency range Move phenomenon；
In the present embodiment, bandpass filter frequency band can remove the alternating current power frequency of 50Hz using 0.1-49.5Hz Interference and high-frequency noise, while can also inhibit the drift phenomenon of low frequency.
It then carries out down-sampled, EEG signals sample rate is reduced to 100Hz, subsequent process is reduced and calculates pressure, improve Efficiency of algorithm.
4., independent component analysis and principal component analysis；
Independent component analysis (ICA) is for removing still remaining various noises, such as electrocardio interference, blink and eye movement Interfere, do not remove clean electromyography signal etc..EEG signals can be separated into multiple signal sources by ICA, and some of them is identification institute The signal source needed, some are then noise source.In addition, also using principal component analysis reduces ingredient number, to reduce calculation amount.
5., using machine learning method, above-mentioned independent element and the signal source that principal component analysis is isolated are judged Whether it is noise source, and noise source is removed.
Classified using the method for machine learning to signal source, can effectively identify noise source and removed it.
S3, pretreated emotion EEG signals are inputted into the shallow-layer convolution mind according to the building of FBCSP feature extracting method In network, a series of training are carried out, obtain shallow-layer convolutional neural networks optimal parameter, obtain trained shallow-layer convolutional Neural Network；
FBCSP (filtering group common space mode) feature extracting method as shown in figure 3, comprising steps of
A, brain electricity original signal is separated to multiple and different frequency ranges using multiple bandpass filters.
B, CSP common space mode is calculated to the signal of each frequency range, it is maximum that EEG signals is transformed into inter-class variance In space, can preferably it classify.
C, the signal after spatial alternation is subjected to feature extraction, is extracted according to following calculation formula:
F=log (1+VAR (Z))
Wherein: f is the feature finally extracted, and Z is transformed signal, and VAR is variance.
D, classify to the step C feature extracted, generally use SVM or neural network as classifier.
In the present embodiment, shallow-layer convolutional neural networks model is constructed according to FBCSP feature extracting method, such as Fig. 4 institute Show, including l0~l4Several layers of structures, in which:
l0:
The layer is input layer, and as input X by taking 1-sec EEG signals segment as an example, size is 62 × 100 matrixes.
l1:
For l1The output of t-th of convolution kernel of layer, is one 62 × 70 matrix,WithRespectively t-th The weight and offset of convolution kernel, i, j are respectively the transverse and longitudinal coordinate of output matrix.Due to l1Layer is set as 10 convolution kernels, therefore Export a(1)Size is 10 × 62 × 70.All convolution process all only carry out on a channel, and the time domain for only extracting data is special Sign.
l2:
a(2)=(BN (z(2)))2
For l2S-th of spatial filter output of layer, j represent j-th of neuron of filter output, and matrix is big Small is 1 × 70, therefore 1≤j≤70.WithThe weight and offset of respectively s-th convolution kernel,Indicating should Some specific value of convolution kernel two-dimensional matrix.l2Layer one shares 30 spatial filters, therefore exports z(2)Size is 30 × 70.l2 Convolution is carried out in Spatial Dimension, only extracts spatial feature.Subsequent convolutional layer input enters BN layers and square non-linear layer, obtains most Output a eventually(2)。
l3:
a(3)=g (m(3))
l3It is mean value pond layer, whereinFor l2The output of s-th of spatial filter of layer is after BN and quadratic nonlinearity As a result,ForMean value Chi Huahou output, size be 1 × 5.Subsequent m(3)By log activation primitive g (x)= Logmax (x, 10-6) obtain final output a(3), choose x and 10-6Maximum value as the input of log function be in order to avoid pole Small value exports a very big negative by log function.l3The output a of layer(3)Size be 30 × 5.
l4:
d(4)=dropout (a(3), p)
f(4)=flatten (d(4))
l4It is full articulamentum, d(4)It is a(3)Output after carrying out dropout according to Probability p, j represent l4J-th nerve Member；Then 30 × 5 matrixes are shakeout to obtain f(4), size becomes 1 × 150；Be represent c classification output layer neuron it is defeated Out, it is calculated by softmax layers, finally according to the maximum classification of output layer value as final classification classification.
The propagated forward process of shallow-layer convolutional neural networks SCN above, network parameter training by back transfer into Row, more sorter networks for softmax as the last layer generally use negative log-likelihood function as loss function, right It is X in input, label is the sample of y, and loss function is shown below:
Wherein: n is output layer neuron number, i.e. classification sum, lcFor the corresponding label value of c class, y is network output,It is the output for representing the output layer neuron of c classification, δ (x) is impulse function.
The set time length sample obtained after pretreatment is generally 1 second or 4 seconds segments, for different time spans, Used shallow-layer convolutional neural networks can be also adjusted therewith.
Network, using small lot gradient decline (Mini-batch gradient descent, MBGD), is criticized big in training It is small to be selected as 128.
The training of shallow-layer convolutional neural networks specifically includes:
S31, the database for establishing different emotions EEG signals collect the specific emotional brain telecommunications of different tested individuals Number, need to guarantee that each EEG signals clip size is consistent, i.e., port number is identical with time span；
S32, EEG signals are improved into signal-to-noise ratio by the preprocessing process of step S2, and divides training according to a certain percentage Collection and test set and verifying collection；
S33, shallow-layer convolutional neural networks model is built by pytorch deep learning library, first initialization model hyper parameter, Including each layer of neuron number, convolution nuclear volume and size, activation primitive, batch size etc.；
S34, the data of training set are inputted into training in the shallow-layer convolutional neural networks of initialization, is calculated by reverse conduction Method adjusts each parameter of shallow-layer convolutional neural networks, stops instruction when verifying collection accuracy rate higher value does not occur within 30 periods Practice, taking the network model of current highest accuracy rate is best model.In addition to this, network model also takes a series of processing and mentions High model generalization ability allows to apply in the big subject of multiple individual differences.
Shallow-layer convolutional neural networks are the convolution nets improved according to filtering group common space mode (FBCSP) algorithm Network, first two layers of network are convolutional layer, carry out feature extraction by different convolution kernels.First convolutional layer is for extracting brain electricity The temporal signatures of data, are equivalent to time domain filtering, correspond to the Filter banks step in FBCSP algorithm, can extract The feature of different frequency.Second convolutional layer is equivalent in common space mode (CSP) algorithm for carrying out space characteristics extraction Spatial filtering step, by data projection to another space.What and then two convolutional layers exported is that a quadratic nonlinearity swashs Function and pond layer living, the activation primitive selection of pond layer is log activation primitive, corresponds to the log- in FBCSP Variance characteristic extraction step.The last layer is full articulamentum plus softmax activation primitive, plays a part of classifier.
In the present embodiment, shallow-layer convolutional neural networks hyper parameter specifically:
The feature of input is the EEG signals sample of 62*100 (1 second length sample, 4 seconds length are 62*400), network Batch size is 128, and convolution kernel size used in first convolutional layer is 1 × 31, step-length 1, convolution nuclear volume be 10 (if It is 4-sec sample, convolution nuclear volume increases to 20).Second convolutional layer convolution kernel size and step-length are respectively 62 × 1 and 1, Convolution nuclear volume increases to 30 (4-sec sample, convolution nuclear volume are constant).The Chi Huahe size of pond layer is 1 × 30, step-length It is 10, local average is carried out to the feature that convolutional layer extracts.
Above-mentioned shallow-layer convolutional neural networks are in order to improve the measure that generalization ability is taken are as follows: batch normalization (BN) and Dropout layers, BN layers can be normalized a certain layer input data, accelerate model convergence rate, improve convolutional network and stablize Property, it is applied between the output of spatial convoluted layer and quadratic nonlinearity layer；Dropout layers are that can effectively inhibit deep learning network mistake One of method of fitting phenomenon, applied to last full articulamentum to reduce network complexity.
S4, it is based on trained shallow-layer convolutional neural networks model, classified to emotion EEG signals, obtain emotion knowledge Other result.
A kind of emotion recognition system based on shallow-layer convolutional neural networks, comprising: eeg signal acquisition device, EEG signals Pretreatment unit, shallow-layer convolutional neural networks and thymencephalon electricity tranining database, eeg signal acquisition device and EEG signals are pre- Processing unit is connected, and shallow-layer convolutional neural networks are connected with EEG signals pretreatment unit and thymencephalon electricity tranining database It connects, in which:
Eeg signal acquisition device is usually wrapped for acquiring the EEG signals under the different emotions state as test set Include a series of facilities such as electrode cap, signal amplifier, PC machine；
EEG signals pretreatment unit, for handling the brain electricity original signal of eeg signal acquisition device acquisition, removal is not The noise signal needed improves EEG signals signal-to-noise ratio, obtains pretreated emotion EEG signals；
Pretreated emotion EEG signals are classified as classifier, obtain emotion by shallow-layer convolutional neural networks Recognition result；
Thymencephalon electricity tranining database, for providing training sample for shallow-layer convolutional neural networks, comprising sufficient amount of Training sample has been subjected to pretreatment.
Shallow-layer convolutional neural networks in the present embodiment emotion recognition system are set according to FBCSP feature extracting method Meter, first two layers is convolutional layer, carries out feature extraction by different convolution kernels；First convolutional layer is for extracting eeg data Temporal signatures are equivalent to time domain filtering, correspond to the Filter banks step in FBCSP algorithm, can extract different frequencies The feature of rate；Second convolutional layer is for carrying out space characteristics extraction, the space filtering being equivalent in common space pattern algorithm Step, by data projection to another space；What and then two convolutional layers exported is a quadratic nonlinearity activation primitive and pond Change layer, the log-variance feature extraction for being log activation primitive, corresponding in FBCSP of the activation primitive selection of pond layer Step；The last layer is full articulamentum plus softmax activation primitive, plays a part of classifier.
The present invention is directed to emotion recognition problem, and the method for brain electric treatment and deep learning is incorporated wherein.By being based on The improved shallow-layer convolutional neural networks of FBCSP feature extracting method carry out feature extraction and model learning to emotion EEG signals, The identification mission to the EEG signals under different emotions state is realized, emotion signal identification rate is effectively improved.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention, It should be equivalent substitute mode, be included within the scope of the present invention.

Claims (10)

1. a kind of emotion identification method based on shallow-layer convolutional neural networks characterized by comprising
EEG signals under S1, acquisition different emotions state, classify by different labels；
S2, all EEG signals are pre-processed, obtains the thymencephalon electricity of the high s/n ratio of several set time length Signal；
S3, pretreated emotion EEG signals are inputted into the shallow-layer convolutional Neural net according to the building of FBCSP feature extracting method In network, a series of training are carried out, obtain shallow-layer convolutional neural networks optimal parameter, obtain trained shallow-layer convolutional Neural net Network；
S4, emotion recognition result is obtained to emotion eeg signal classification based on trained shallow-layer convolutional neural networks model.
2. emotion identification method according to claim 1, which is characterized in that EEG signals, which pre-process, includes:
1., EEG signals sample separation, the EEG signals segment after being separated；
2., removal separation after EEG signals segment in bad sample and bad channel；
3., EEG signals segment bandpass filtering and down-sampled processing；
4., independent component analysis and principal component analysis；
5., using machine learning method the signal source that independent element and principal component analysis are isolated is carried out judging whether it is noise Source, and noise source is removed.
3. emotion identification method according to claim 2, which is characterized in that judged whether by following calculation formula For bad sample:
Wherein: aN, cFor the average value in c-th of channel of n-th of sample, mcAnd σcIt is then the equal of all period segments in the channel c respectively Value and standard deviation；
One threshold value t=3 is set, z is worked asN, cWhen > 3, illustrate the amplitude and other period segments of entire channel in the channel c of the sample It differs greatly, can be judged as of poor quality；
For the same sample, if it exceeds 20% channel is considered to of poor quality, then entire sample will be removed directly, sample This number subtracts 1；If be no more than 20%, which can be retained, but bad channel can be removed and according to peripheral channel data into Row is repaired.
4. emotion identification method according to claim 2, which is characterized in that the detection in bad channel is by calculating each channel Judge with the maximum correlation coefficient in other channels, related coefficient calculation formula is as follows:
X, Y is two signals, and r (X, Y) is the related coefficient of two signals, and Cov (X, Y) is covariance, and Var [X] and Var [Y] divide Not Wei X, Y variance；
If there is with the related coefficient maximum value in other channels the phenomenon that less than 0.4 in the epochs that certain channel is more than 2%, Then the channel is judged as bad channel.
5. the emotion identification method according to any one of claim 2-4, which is characterized in that the bandpass filter frequency Band is 0.1-49.5Hz.
6. emotion identification method according to claim 1, which is characterized in that further comprise the steps of:
Building shallow-layer convolutional neural networks are carried out according to FBCSP feature extracting method.
7. emotion identification method according to claim 6, which is characterized in that shallow-layer convolutional neural networks are special according to FBCSP Sign extracting method is constructed, and first two layers is convolutional layer, carries out feature extraction by different convolution kernels；First convolutional layer is used In the temporal signatures for extracting eeg data, it is equivalent to time domain filtering, corresponds to the Filter banks step in FBCSP algorithm Suddenly, for extracting the feature of different frequency；Second convolutional layer is equivalent to common space mode for carrying out space characteristics extraction Spatial filtering step in algorithm, by data projection to another space；What and then two convolutional layers exported is one square non- Linear activation primitive and pond layer, the activation primitive selection of pond layer is log activation primitive, corresponds to the log- in FBCSP Variance characteristic extraction step；The last layer is full articulamentum plus softmax activation primitive, plays a part of classifier.
8. emotion identification method according to claim 7, which is characterized in that shallow-layer convolutional neural networks include l0~l4It is several Layer structure, in which:
l0:
The layer is input layer, and as input X by taking 1-sec EEG signals segment as an example, size is 62 × 100 matrixes；
l1:
For l1The output of t-th of convolution kernel of layer, is one 62 × 70 matrix,WithRespectively t-th of convolution The weight and offset of core, i, j are respectively the transverse and longitudinal coordinate of output matrix；Due to l1Layer is set as 10 convolution kernels, therefore exports a(1)Size is 10 × 62 × 70；All convolution process all only carry out on a channel, only extract the temporal signatures of data；
l2:
a(2)=(BN (z(2)))2
For l2S-th of spatial filter output of layer, j represent j-th of neuron of filter output, matrix size 1 × 70, therefore 1≤j≤70；WithThe weight and offset of respectively s-th convolution kernel,Indicate the convolution kernel Some specific value of two-dimensional matrix；l2Layer one shares 30 spatial filters, therefore exports z(2)Size is 30 × 70；l2In space Dimension carries out convolution, only extracts spatial feature；Subsequent convolutional layer input enters BN layers and square non-linear layer, obtains final output a(2)；
l3:
a(3)=g (m(3))
l3It is mean value pond layer, whereinFor l2Knot of the output of s-th of spatial filter of layer after BN and quadratic nonlinearity Fruit,ForMean value Chi Huahou output, size be 1 × 5；Subsequent m(3)By log activation primitive g (x)=logmax (x, 10-6) obtain final output a(3)；l3The output a of layer(3)Size be 30 × 5；
l4:
d(4)=dropout (a(3), p)
f(4)=flatten (d(4))
l4It is full articulamentum, d(4)It is a(3)Output after carrying out dropout according to Probability p, j represent l4J-th of neuron；So Afterwards 30 × 5 matrixes are shakeout to obtain f(4), size becomes 1 × 150；It is the output for representing the output layer neuron of c classification, by Softmax layers are calculated, according to the maximum classification of output layer value as final classification classification.
9. a kind of emotion recognition system based on shallow-layer convolutional neural networks characterized by comprising eeg signal acquisition dress Set, EEG signals pretreatment unit, shallow-layer convolutional neural networks and thymencephalon electricity tranining database, eeg signal acquisition device with EEG signals pretreatment unit is connected, and shallow-layer convolutional neural networks and EEG signals pretreatment unit and thymencephalon electricity train number It is connected according to library, in which:
Eeg signal acquisition device generally includes electricity for acquiring the EEG signals under the different emotions state as test set A series of facilities of polar cap, signal amplifier, PC machine；
EEG signals pretreatment unit, for handling the brain electricity original signal of eeg signal acquisition device acquisition, removal is not needed Noise signal, improve EEG signals signal-to-noise ratio, obtain pretreated emotion EEG signals；
Pretreated emotion EEG signals are classified as classifier, obtain emotion recognition by shallow-layer convolutional neural networks As a result；
Thymencephalon electricity tranining database includes sufficient amount of training for providing training sample for shallow-layer convolutional neural networks Sample has been subjected to pretreatment.
10. emotion recognition system according to claim 9, which is characterized in that shallow-layer convolutional neural networks are special according to FBCSP Sign extracting method is constructed, and first two layers is convolutional layer, carries out feature extraction by different convolution kernels；First convolutional layer is used In the temporal signatures for extracting eeg data, it is equivalent to time domain filtering, corresponds to the Filter banks step in FBCSP algorithm Suddenly, for extracting the feature of different frequency；Second convolutional layer is equivalent to common space mode for carrying out space characteristics extraction Spatial filtering step in algorithm, by data projection to another space；What and then two convolutional layers exported is one square non- Linear activation primitive and pond layer, the activation primitive selection of pond layer is log activation primitive, corresponds to the log- in FBCSP Variance characteristic extraction step；The last layer is full articulamentum plus softmax activation primitive, plays a part of classifier.
CN201910591898.7A 2019-07-02 2019-07-02 A kind of emotion identification method and system based on shallow-layer convolutional neural networks Pending CN110353702A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910591898.7A CN110353702A (en) 2019-07-02 2019-07-02 A kind of emotion identification method and system based on shallow-layer convolutional neural networks

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910591898.7A CN110353702A (en) 2019-07-02 2019-07-02 A kind of emotion identification method and system based on shallow-layer convolutional neural networks

Publications (1)

Publication Number Publication Date
CN110353702A true CN110353702A (en) 2019-10-22

Family

ID=68217768

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910591898.7A Pending CN110353702A (en) 2019-07-02 2019-07-02 A kind of emotion identification method and system based on shallow-layer convolutional neural networks

Country Status (1)

Country Link
CN (1) CN110353702A (en)

Cited By (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111461204A (en) * 2020-03-30 2020-07-28 华南理工大学 Emotion identification method based on electroencephalogram signals and used for game evaluation
CN111671445A (en) * 2020-04-20 2020-09-18 广东食品药品职业学院 Consciousness disturbance degree analysis method
CN111709267A (en) * 2020-03-27 2020-09-25 吉林大学 Electroencephalogram signal emotion recognition method of deep convolutional neural network
CN111783942A (en) * 2020-06-08 2020-10-16 北京航天自动控制研究所 Brain cognition process simulation method based on convolution cyclic neural network
CN111860463A (en) * 2020-08-07 2020-10-30 北京师范大学 Emotion identification method based on joint norm
CN111882036A (en) * 2020-07-22 2020-11-03 广州大学 Convolutional neural network training method, electroencephalogram signal identification method, device and medium
CN112036229A (en) * 2020-06-24 2020-12-04 宿州小马电子商务有限公司 Intelligent bassinet electroencephalogram signal channel configuration method with demand sensing function
CN112084935A (en) * 2020-09-08 2020-12-15 南京邮电大学 Emotion recognition method based on expansion of high-quality electroencephalogram sample
CN112101152A (en) * 2020-09-01 2020-12-18 西安电子科技大学 Electroencephalogram emotion recognition method and system, computer equipment and wearable equipment
CN112381008A (en) * 2020-11-17 2021-02-19 天津大学 Electroencephalogram emotion recognition method based on parallel sequence channel mapping network
CN112465069A (en) * 2020-12-15 2021-03-09 杭州电子科技大学 Electroencephalogram emotion classification method based on multi-scale convolution kernel CNN
CN112488002A (en) * 2020-12-03 2021-03-12 重庆邮电大学 Emotion recognition method and system based on N170
CN112869711A (en) * 2021-01-19 2021-06-01 华南理工大学 Automatic sleep staging and migration method based on deep neural network
CN113052113A (en) * 2021-04-02 2021-06-29 中山大学 Depression identification method and system based on compact convolutional neural network
CN113069117A (en) * 2021-04-02 2021-07-06 中山大学 Electroencephalogram emotion recognition method and system based on time convolution neural network
CN113128353A (en) * 2021-03-26 2021-07-16 安徽大学 Emotion sensing method and system for natural human-computer interaction
CN113180659A (en) * 2021-01-11 2021-07-30 华东理工大学 Electroencephalogram emotion recognition system based on three-dimensional features and cavity full convolution network
CN113191225A (en) * 2021-04-19 2021-07-30 华南师范大学 Emotional electroencephalogram recognition method and system based on graph attention network
CN113576493A (en) * 2021-08-23 2021-11-02 安徽七度生命科学集团有限公司 User state identification method for health physiotherapy cabin
CN113627518A (en) * 2021-08-07 2021-11-09 福州大学 Method for realizing multichannel convolution-recurrent neural network electroencephalogram emotion recognition model by utilizing transfer learning
CN113642528A (en) * 2021-09-14 2021-11-12 西安交通大学 Hand movement intention classification method based on convolutional neural network
CN113723557A (en) * 2021-09-08 2021-11-30 山东大学 Depression electroencephalogram classification system based on multiband time-space convolution network
CN113791691A (en) * 2021-09-18 2021-12-14 中国科学院自动化研究所 Electroencephalogram signal band positioning method and device
CN114578967A (en) * 2022-03-08 2022-06-03 天津理工大学 Emotion recognition method and system based on electroencephalogram signals
CN116671881A (en) * 2023-08-03 2023-09-01 北京九叁有方物联网科技有限公司 Head-wearing brain body operation capability assessment device and method based on graph neural network

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5601090A (en) * 1994-07-12 1997-02-11 Brain Functions Laboratory, Inc. Method and apparatus for automatically determining somatic state
CN107714057A (en) * 2017-10-01 2018-02-23 南京邮电大学盐城大数据研究院有限公司 A kind of three classification Emotion identification model methods based on convolutional neural networks
CN109508651A (en) * 2018-10-24 2019-03-22 辽宁师范大学 Brain electricity sensibility classification method based on convolutional neural networks

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5601090A (en) * 1994-07-12 1997-02-11 Brain Functions Laboratory, Inc. Method and apparatus for automatically determining somatic state
CN107714057A (en) * 2017-10-01 2018-02-23 南京邮电大学盐城大数据研究院有限公司 A kind of three classification Emotion identification model methods based on convolutional neural networks
CN109508651A (en) * 2018-10-24 2019-03-22 辽宁师范大学 Brain electricity sensibility classification method based on convolutional neural networks

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ROBIN TIBOR SCHIRRMEISTER等: "Deep Learning With Convolutional Neural Networks for EEG Decoding and Visualization", 《HUMAN BRAIN MAPPING》 *
张本禹: "基于卷积神经网络的EEG情绪分类研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
邱嘉裕: "基于静息态脑电的脑卒中患者大脑异常活动研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑辑》 *

Cited By (41)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111709267A (en) * 2020-03-27 2020-09-25 吉林大学 Electroencephalogram signal emotion recognition method of deep convolutional neural network
CN111709267B (en) * 2020-03-27 2022-03-29 吉林大学 Electroencephalogram signal emotion recognition method of deep convolutional neural network
CN111461204A (en) * 2020-03-30 2020-07-28 华南理工大学 Emotion identification method based on electroencephalogram signals and used for game evaluation
CN111461204B (en) * 2020-03-30 2023-05-26 华南理工大学 Emotion recognition method based on electroencephalogram signals for game evaluation
CN111671445A (en) * 2020-04-20 2020-09-18 广东食品药品职业学院 Consciousness disturbance degree analysis method
CN111783942A (en) * 2020-06-08 2020-10-16 北京航天自动控制研究所 Brain cognition process simulation method based on convolution cyclic neural network
CN111783942B (en) * 2020-06-08 2023-08-01 北京航天自动控制研究所 Brain cognitive process simulation method based on convolutional recurrent neural network
CN112036229A (en) * 2020-06-24 2020-12-04 宿州小马电子商务有限公司 Intelligent bassinet electroencephalogram signal channel configuration method with demand sensing function
CN112036229B (en) * 2020-06-24 2024-04-19 宿州小马电子商务有限公司 Intelligent bassinet electroencephalogram signal channel configuration method with demand sensing function
CN111882036A (en) * 2020-07-22 2020-11-03 广州大学 Convolutional neural network training method, electroencephalogram signal identification method, device and medium
CN111882036B (en) * 2020-07-22 2023-10-31 广州大学 Convolutional neural network training method, electroencephalogram signal identification method, device and medium
CN111860463B (en) * 2020-08-07 2024-02-02 北京师范大学 Emotion recognition method based on joint norm
CN111860463A (en) * 2020-08-07 2020-10-30 北京师范大学 Emotion identification method based on joint norm
CN112101152A (en) * 2020-09-01 2020-12-18 西安电子科技大学 Electroencephalogram emotion recognition method and system, computer equipment and wearable equipment
CN112101152B (en) * 2020-09-01 2024-02-02 西安电子科技大学 Electroencephalogram emotion recognition method, electroencephalogram emotion recognition system, computer equipment and wearable equipment
CN112084935A (en) * 2020-09-08 2020-12-15 南京邮电大学 Emotion recognition method based on expansion of high-quality electroencephalogram sample
CN112084935B (en) * 2020-09-08 2022-07-26 南京邮电大学 Emotion recognition method based on expansion of high-quality electroencephalogram sample
CN112381008A (en) * 2020-11-17 2021-02-19 天津大学 Electroencephalogram emotion recognition method based on parallel sequence channel mapping network
CN112381008B (en) * 2020-11-17 2022-04-29 天津大学 Electroencephalogram emotion recognition method based on parallel sequence channel mapping network
CN112488002A (en) * 2020-12-03 2021-03-12 重庆邮电大学 Emotion recognition method and system based on N170
CN112465069A (en) * 2020-12-15 2021-03-09 杭州电子科技大学 Electroencephalogram emotion classification method based on multi-scale convolution kernel CNN
CN112465069B (en) * 2020-12-15 2024-02-06 杭州电子科技大学 Electroencephalogram emotion classification method based on multi-scale convolution kernel CNN
CN113180659B (en) * 2021-01-11 2024-03-08 华东理工大学 Electroencephalogram emotion recognition method based on three-dimensional feature and cavity full convolution network
CN113180659A (en) * 2021-01-11 2021-07-30 华东理工大学 Electroencephalogram emotion recognition system based on three-dimensional features and cavity full convolution network
CN112869711A (en) * 2021-01-19 2021-06-01 华南理工大学 Automatic sleep staging and migration method based on deep neural network
CN113128353B (en) * 2021-03-26 2023-10-24 安徽大学 Emotion perception method and system oriented to natural man-machine interaction
CN113128353A (en) * 2021-03-26 2021-07-16 安徽大学 Emotion sensing method and system for natural human-computer interaction
CN113052113A (en) * 2021-04-02 2021-06-29 中山大学 Depression identification method and system based on compact convolutional neural network
CN113069117A (en) * 2021-04-02 2021-07-06 中山大学 Electroencephalogram emotion recognition method and system based on time convolution neural network
CN113191225A (en) * 2021-04-19 2021-07-30 华南师范大学 Emotional electroencephalogram recognition method and system based on graph attention network
CN113191225B (en) * 2021-04-19 2023-07-04 华南师范大学 Emotion electroencephalogram recognition method and system based on graph attention network
CN113627518B (en) * 2021-08-07 2023-08-08 福州大学 Method for realizing neural network brain electricity emotion recognition model by utilizing transfer learning
CN113627518A (en) * 2021-08-07 2021-11-09 福州大学 Method for realizing multichannel convolution-recurrent neural network electroencephalogram emotion recognition model by utilizing transfer learning
CN113576493A (en) * 2021-08-23 2021-11-02 安徽七度生命科学集团有限公司 User state identification method for health physiotherapy cabin
CN113723557B (en) * 2021-09-08 2023-08-08 山东大学 Depression brain electricity classifying system based on multiband space-time convolution network
CN113723557A (en) * 2021-09-08 2021-11-30 山东大学 Depression electroencephalogram classification system based on multiband time-space convolution network
CN113642528A (en) * 2021-09-14 2021-11-12 西安交通大学 Hand movement intention classification method based on convolutional neural network
CN113791691B (en) * 2021-09-18 2022-05-20 中国科学院自动化研究所 Electroencephalogram signal band positioning method and device
CN113791691A (en) * 2021-09-18 2021-12-14 中国科学院自动化研究所 Electroencephalogram signal band positioning method and device
CN114578967A (en) * 2022-03-08 2022-06-03 天津理工大学 Emotion recognition method and system based on electroencephalogram signals
CN116671881A (en) * 2023-08-03 2023-09-01 北京九叁有方物联网科技有限公司 Head-wearing brain body operation capability assessment device and method based on graph neural network

Similar Documents

Publication Publication Date Title
CN110353702A (en) A kind of emotion identification method and system based on shallow-layer convolutional neural networks
Kachenoura et al. ICA: a potential tool for BCI systems
CN105956624B (en) Mental imagery brain electricity classification method based on empty time-frequency optimization feature rarefaction representation
CN110353673B (en) Electroencephalogram channel selection method based on standard mutual information
Bashar et al. Human identification from brain EEG signals using advanced machine learning method EEG-based biometrics
CN112656427A (en) Electroencephalogram emotion recognition method based on dimension model
Pan et al. Emotion recognition based on EEG using generative adversarial nets and convolutional neural network
CN112488002B (en) Emotion recognition method and system based on N170
CN114533086B (en) Motor imagery brain electrolysis code method based on airspace characteristic time-frequency transformation
CN111091074A (en) Motor imagery electroencephalogram signal classification method based on optimal region common space mode
CN106725452A (en) Based on the EEG signal identification method that emotion induces
CN108256579A (en) A kind of multi-modal sense of national identity quantization measuring method based on priori
CN113180659B (en) Electroencephalogram emotion recognition method based on three-dimensional feature and cavity full convolution network
Cheng et al. Emotion recognition algorithm based on convolution neural network
Lu et al. Combined CNN and LSTM for motor imagery classification
Wang et al. Classification of EEG signal using convolutional neural networks
CN111000556A (en) Emotion recognition method based on deep fuzzy forest
AU2013100576A4 (en) Human Identification with Electroencephalogram (EEG) for the Future Network Security
CN113576498B (en) Visual and auditory aesthetic evaluation method and system based on electroencephalogram signals
CN113128353B (en) Emotion perception method and system oriented to natural man-machine interaction
CN114081505A (en) Electroencephalogram signal identification method based on Pearson correlation coefficient and convolutional neural network
CN110192864A (en) A kind of cross-domain electrocardiogram biometric identity recognition methods
CN117883082A (en) Abnormal emotion recognition method, system, equipment and medium
Rammy et al. Sequence-to-sequence deep neural network with spatio-spectro and temporal features for motor imagery classification
CN116421200A (en) Brain electricity emotion analysis method of multi-task mixed model based on parallel training

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20191022

RJ01 Rejection of invention patent application after publication