CN113157094B - Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation - Google Patents

Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation Download PDF

Info

Publication number
CN113157094B
CN113157094B CN202110428950.4A CN202110428950A CN113157094B CN 113157094 B CN113157094 B CN 113157094B CN 202110428950 A CN202110428950 A CN 202110428950A CN 113157094 B CN113157094 B CN 113157094B
Authority
CN
China
Prior art keywords
matrix
label
domain data
electroencephalogram
undirected graph
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.)
Active
Application number
CN202110428950.4A
Other languages
Chinese (zh)
Other versions
CN113157094A (en
Inventor
李文政
黄文娜
王文娟
彭勇
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.)
Hangzhou Dianzi University
Original Assignee
Hangzhou Dianzi University
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 Hangzhou Dianzi University filed Critical Hangzhou Dianzi University
Priority to CN202110428950.4A priority Critical patent/CN113157094B/en
Publication of CN113157094A publication Critical patent/CN113157094A/en
Application granted granted Critical
Publication of CN113157094B publication Critical patent/CN113157094B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/011—Arrangements for interaction with the human body, e.g. for user immersion in virtual reality
    • G06F3/015—Input arrangements based on nervous system activity detection, e.g. brain waves [EEG] detection, electromyograms [EMG] detection, electrodermal response detection
    • 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
    • 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
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Dermatology (AREA)
  • Neurology (AREA)
  • Neurosurgery (AREA)
  • Human Computer Interaction (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention discloses a brain electric emotion recognition method combining feature migration and image semi-supervised label propagation. The method specifically comprises the steps of guiding a testee to watch electroencephalogram data with obvious emotional tendency, preprocessing the electroencephalogram data, extracting characteristics of the electroencephalogram data, and generating a sample matrix. And constructing a learning model for joint feature transfer learning and state estimation, wherein the learning model comprises a single mapping domain adaptation model and a semi-supervised label propagation model, and a joint optimization objective function is obtained. And then, joint optimization is realized by fixing two variables and updating the rule of the other variable according to the target function, and a subspace is shared by continuous iterative optimization features to obtain a better migration effect so as to improve the accuracy of emotion recognition. The method can be used for emotion recognition across test migrations.

Description

Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation
Technical Field
The invention belongs to the technical field of electroencephalogram signal processing, and particularly relates to an electroencephalogram emotion recognition method combining feature migration and image semi-supervised label propagation.
Background
Emotion is a corresponding physiological expression produced by the brain when people are stimulated by the outside world in daily life and work, and has the functions of information transmission and behavior regulation. For a machine, the emotion recognition method can automatically and accurately recognize human emotion, and better realize emotion human-computer interaction is a research hotspot in the fields of current information science, psychology, cognitive neuroscience and the like. However, human emotional expression is affected by such factors as environmental situation, expression object and thinking cognition, thereby causing difficulty in emotion recognition, and recognition of current emotional state can be mainly classified into four categories: facial expression, text, voice and physiological signals, the former three are non-physiological signals, and because the non-physiological signals have camouflage property and cannot ensure the reliability of the recognition result, the physiological signals are usually adopted for recognition. The electroencephalogram signal is a physiological signal which is not easy to disguise, and the electroencephalogram signal is very important for improving the accuracy and reliability of emotion recognition in human-computer interaction.
Common migratory learning methods can be generally classified into four categories: model-based, feature-based, sample-based, relationship-based. Among them, the feature-based method is the most widely used migration method, and this class of methods aims to learn a shared feature representation, i.e., a shared subspace, and map the target domain and source domain data into the shared subspace by means of projection in combination with some measurement strategies, such as the maximum mean difference MMD, etc., so as to minimize the difference between the two conditional distributions and the marginal distribution. However, for target domain data without tags, calculating the condition distribution of the target domain data cannot be achieved, so if a good target domain tag can be learned, a better projection matrix can be obtained, the difference between a source domain and a target domain can be reduced, a more excellent target tag can be obtained, the model identification precision is improved, and the reliability of emotion human-computer interaction is ensured.
However, the electroencephalogram signal is an unsteady signal, and different electroencephalogram signal characteristics tested for the same emotion are different if the cross-test migration recognition is simply performed, and the requirement of human-computer interaction on emotion recognition accuracy cannot be well met.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a brain electric emotion recognition method combining feature migration and graph semi-supervised label propagation, which is implemented by aligning projection matrix W and target domain emotion label F of source domain data and target domain data in subspace t And joint iterative optimization is carried out on the undirected graph S, and a subspace is shared through continuous iterative optimization features, so that a better migration effect is obtained, and the accuracy of emotion recognition is improved.
Step 1, collecting electroencephalogram data of a tested person in K different emotional states.
And 2, preprocessing and extracting features of the electroencephalogram data acquired in the step 1, wherein each sample matrix X consists of electroencephalogram features of a testee, and the label vector y is an emotion label corresponding to the electroencephalogram features in the sample matrix X. Selecting two different sample matricesRespectively as source domain data X s And target data X t 。
And 3, constructing a learning model combining feature transfer learning and state estimation, wherein the learning model comprises a single mapping domain adaptation model and a semi-supervised label propagation model.
Step 3.1, establishing a single mapping domain adaptive model:
Figure BDA0003030687820000021
wherein the content of the first and second substances,
Figure BDA0003030687820000022
in order to map the source domain data and the target domain data to a mapping matrix in the same shared subspace, p represents a target dimension, and d represents an original dimension; />
Figure BDA0003030687820000023
C is the number of categories for the augmented source domain data tag; />
Figure BDA0003030687820000024
To augment the target domain data tag, n s 、n t The sample numbers of the source domain data and the target domain data respectively; n = n s +n t Representing the total number of samples; />
Figure BDA0003030687820000025
Figure BDA0003030687820000026
The matrix is a diagonal matrix, wherein the kth diagonal element in the matrix is the reciprocal of the number of data samples of the kth source domain or target domain, and k =1, 2., c +1; wherein +>
Figure BDA0003030687820000027
Is a central matrix, I is a unit matrix, R is a unit matrix>
Figure BDA0003030687820000028
Represents FroCalculating benius norm; superscript T denotes transpose.
Step 3.2, establishing a semi-supervised label propagation model based on an undirected graph:
Figure BDA0003030687820000029
wherein S is ij Representing the elements of the ith row and the jth column of the undirected graph correlation matrix S, i.e., the degree of correlation between the ith sample and the jth sample (i =1, 2.. Multidot.n; j =1, 2.. Multidot.n); alpha and gamma are parameters;
Figure BDA00030306878200000210
is a Laplace transform matrix, is obtained by a correlation matrix S, L = S-D, D is an n-dimensional diagonal matrix, and the ith diagonal element->
Figure BDA00030306878200000211
Figure BDA00030306878200000212
For label matrix for label propagation, F s 、F t Labels for the source domain and the target domain, respectively; 1 is a full 1 matrix; tr (-) represents the trace of the matrix; />
Figure BDA00030306878200000213
Represents the calculation of a 2 norm; the first term in the formula (2) is semi-supervised label propagation, and the second term and the third term are correlation matrixes for solving an undirected graph so as to ensure that a good undirected graph is constructed.
Step 3.3, combining the two models constructed in the steps 3.1 and 3.2, integrating domain adaptation and undirected graph-based semi-supervised label propagation into a unified framework for joint optimization, wherein the optimization model is as follows:
Figure BDA0003030687820000031
λ, α, γ are parameters.
Step 4, according to the optimization model established in the step 3.3, mapping matrix W and target domain data label F t And performing joint iterative optimization on the undirected graph incidence matrix S.
Step 4.1, initialize the data label F of the target domain t Has a value of
Figure BDA0003030687820000032
And constructing an initial undirected graph using a neural network. />
Preferably, the initial undirected graph is constructed from a KNN network.
Step 4.2, fixing the target domain data label F t And an undirected graph incidence matrix S, updating a mapping matrix W, wherein the objective function is as follows:
Figure BDA0003030687820000033
and solving the formula (4) to obtain an updated mapping matrix.
Step 4.3, fixing the mapping matrix W and the undirected graph correlation matrix S, and updating a target domain data label Ft, wherein the target function is as follows:
Figure BDA0003030687820000034
and solving the formula (5) to obtain the updated target domain data label.
Step 4.4, fix target domain data tag F t And a mapping matrix W, updating an undirected graph correlation matrix S, wherein the objective function is as follows:
Figure BDA0003030687820000035
and (5) solving the formula (6) to obtain an updated undirected graph correlation matrix S.
Step 4.5, repeating steps 4.2, 4.3 and 4.4 for multiple times to complete mapping matrix W and target domain data label F t And joint iterative optimization of the undirected graph correlation matrix S.
And 5, inputting the sample matrix X obtained in the step 2 into the objective function subjected to iterative optimization in the step 4 to obtain a corresponding predicted value label, wherein the predicted value label is the emotional state of the testee corresponding to the sample at the acquisition moment.
The invention has the following beneficial effects:
1. the combined electroencephalogram feature migration and emotional state estimation method model provided by the invention provides an effective tool with higher accuracy for emotional man-machine interaction, and the target label is continuously iteratively optimized through the mathematical model, so that the emotional state of the testee can be accurately identified according to electroencephalogram data.
2. Aiming at the situation that the electroencephalogram research field is difficult to cross the tested situation, iteration optimization is carried out by combining the undirected graph construction, the semi-supervised label propagation and the domain adaptation, the undirected graph constructed based on sample data is continuously optimized, a better target label is obtained by the semi-supervised label propagation optimization, a more excellent mapping matrix is obtained, the migration effect is improved, the source domain data and the target domain data which are closer are obtained for composition, and the iteration and the optimization are continuously carried out, so that a more excellent emotion state recognition result is obtained, and the emotion recognition accuracy of the cross-tested migration is improved.
Drawings
FIG. 1 is a flow chart of the method.
Detailed Description
The invention is further explained below with reference to the drawings;
as shown in fig. 1, the electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation specifically comprises the following steps:
step 1, electroencephalogram data acquisition.
The human emotion does not appear very strong under daily conditions, therefore, in order to acquire strong emotion information, certain induction needs to be carried out on a human subject, 4 film segments with obvious emotion tendencies are selected to be respectively played to the human subject at different times for watching, and the 4 film segments are connected to corresponding brain areas through brain cap leads while watching a film to acquire brain electrical data of the human subject as an original emotion brain electrical data set.
And 2, preprocessing data.
Sampling the electroencephalogram data acquired in the step 1, wherein the sampling rate is 200Hz, filtering noise and artifacts by a band-pass filter of 1 Hz-75 Hz, and calculating Differential Entropy (DE) of the electroencephalogram data as a sample matrix S in 5 frequency bands (Delta (1-4 Hz), theta (4-8 Hz), alpha (8-14 Hz), beta (14-31 Hz) and Gamma (31-50 Hz)) respectively:
Figure BDA0003030687820000041
wherein σ is a standard deviation of the probability density function; μ is the expectation of the probability density function.
It can be seen that the differential entropy signature is essentially a logarithmic form of the power spectral density signature, i.e.
Figure BDA0003030687820000042
The pretreatment of the brain electrical signals aims to improve the signal to noise ratio, thereby improving the pretreatment effect of data and reducing interference.
And the label vector y is an emotion label corresponding to the sample matrix X.
Step 3, establishing a combined electroencephalogram feature migration and emotional state estimation method model, integrating a single mapping domain adaptation model and a semi-supervised label propagation model based on an undirected graph into a unified framework, and obtaining a combined optimization objective function, wherein the specific steps are as follows:
step 3.1, establishing a single mapping domain adaptive model:
Figure BDA0003030687820000051
wherein the content of the first and second substances,
Figure BDA0003030687820000052
in order to map the source domain data and the target domain data to a mapping matrix in the same shared subspace, p represents a target dimension, d represents an original dimensionDegree; x s 、X t Respectively source domain data and target domain data;
Figure BDA0003030687820000053
c is the number of categories for the augmented source domain data tag; />
Figure BDA0003030687820000054
Unknown in a single domain adaptation to augment target domain data tags, n s 、n t The number of samples of the source domain data and the target domain data respectively; />
Figure BDA0003030687820000055
The matrix is a diagonal matrix, and the kth diagonal element in the matrix is the reciprocal of the data sample number of the kth source domain or the target domain; wherein +>
Figure BDA0003030687820000056
Is a central matrix, and I is a unit matrix; />
Figure BDA0003030687820000057
Representing the calculation of the Frobenius norm; superscript T denotes transpose.
Step 3.2, establishing a semi-supervised label propagation model based on an undirected graph:
Figure BDA0003030687820000058
wherein S is ij Representing the elements of the ith row and the jth column of the undirected graph correlation matrix S, i.e., the degree of correlation between the ith sample and the jth sample (i =1, 2.. Multidot.n; j =1, 2.. Multidot.n); alpha and gamma are parameters;
Figure BDA0003030687820000059
is a Laplace transform matrix, is found by a correlation matrix S, L = S-D, D is an n-dimensional diagonal matrix, the ith diagonal element->
Figure BDA00030306878200000510
Figure BDA00030306878200000511
For label matrices for label propagation, F s 、F t Labels for the source domain and the target domain, respectively; 1 is a full 1 matrix; tr (-) denotes the trace of the matrix; sigma (-) is a summation formula; />
Figure BDA00030306878200000512
Represents the calculation of a 2 norm; the first term in the formula (9) is semi-supervised label propagation, and the second term and the third term are correlation matrixes for solving an undirected graph so as to ensure that a good undirected graph is constructed.
Step 3.3, combining the two models constructed in the steps 3.1 and 3.2, and integrating the models into a unified framework to obtain a joint optimization objective function:
Figure BDA0003030687820000061
λ, α, γ are parameters.
Step 4, according to the optimization model established in the step 3.3, mapping matrix W and target domain data label F t And performing joint iterative optimization on the undirected graph incidence matrix S.
Step 4.1, initialize the data label F of the target domain t Is 0.25 and an initial undirected graph is constructed using a KNN network.
Step 4.2, fixing the target domain data label F t And an undirected graph incidence matrix S, updating a mapping matrix W, wherein the objective function is as follows:
Figure BDA0003030687820000062
the objective function is an equality constraint optimization problem, and is calculated by adopting a Lagrangian function to construct the Lagrangian function:
Figure BDA0003030687820000063
where phi is the lagrangian multiplier corresponding to the equality constraint,
Figure BDA0003030687820000064
equation (12) can be obtained by partial derivation of W:
Figure BDA0003030687820000065
/>
thus, the device
(CC T +2αXLX T )W＝ΦXHX T W (14)
Equation (14) is a generalized eigenvalue decomposition problem, let M = (XHX) T ) -1 (CC T +2αXLX T ) And simultaneously taking W as a group of standard orthogonal basis vectors, performing eigenvalue decomposition on the matrix W, and taking the eigenvectors corresponding to the first p smallest eigenvalues to obtain an updated mapping matrix W. p is the shared subspace dimension.
Step 4.3, fixing the mapping matrix W and the undirected graph correlation matrix S, and updating the target domain data label F t The objective function is:
Figure BDA0003030687820000066
order to
Figure BDA0003030687820000067
Equation (15) is reduced to:
Figure BDA0003030687820000071
converting equation (16) to the trace form:
Figure BDA0003030687820000072
in the formula (17), the target variable exists in two forms, one is augmented, and the other is normal, and the augmented form of the target variable is disassembled, that is, the target variable is subjected to
Figure BDA0003030687820000073
In formula (17):
the first item:
Figure BDA0003030687820000074
the second term is:
Figure BDA0003030687820000075
wherein
Figure BDA0003030687820000076
Since the trace of the matrix is the sum of the major diagonal elements and the target variable is F t Thus, equation (17) is transformed to:
Figure BDA0003030687820000077
wherein the content of the first and second substances,
Figure BDA0003030687820000078
the formula (18) is simplified in a line-by-line solving mode, and is taken
Figure BDA0003030687820000079
Is->
Figure BDA00030306878200000710
Ith t A column vector, i.e.>
Figure BDA00030306878200000711
Is F t Ith of (2) t Line row vector, i t ＝1，2，...，n t 。
The first item:
Figure BDA00030306878200000712
the second term is:
Figure BDA00030306878200000713
the third item:
Figure BDA00030306878200000714
wherein
Figure BDA00030306878200000715
I th of M t A column vector; j is a function of t ＝1，2，...，n t . At this time, the ith t The objective function of the row is: />
Figure BDA00030306878200000716
The first term and the third term in the formula (19) are both in the form of products, only the second term is a calculation in the form of 2 norms, and the second term is decomposed:
Figure BDA00030306878200000717
due to the fact that
Figure BDA00030306878200000718
Is a constant whose trace is itself, so equation (19) can be written as:
Figure BDA0003030687820000081
wherein the content of the first and second substances,
Figure BDA0003030687820000082
at the same time, get
Figure BDA0003030687820000083
The target formula can be simplified as follows:
Figure BDA0003030687820000084
solving the formula (22) to obtain the updated target domain data label F t 。
Step 4.4, fix target domain data tag F t And a mapping matrix W, updating an undirected graph correlation matrix S, wherein the objective function is as follows:
Figure BDA0003030687820000085
and similarly, adopting a line-by-line solving mode, firstly, disassembling a first item trace in the objective function:
Figure BDA0003030687820000086
since the calculation of the two-norm in equation (24) is independent of the target variable, let:
Figure BDA0003030687820000087
then equation (24) is solved in rows, and equation (26) is row i:
Figure BDA0003030687820000088
get
Figure BDA0003030687820000089
Then equation (26) can be written as: />
Figure BDA00030306878200000810
Order to
Figure BDA00030306878200000811
At this time, equation (27) can be converted into:
Figure BDA00030306878200000812
and solving the formula (28) to obtain the updated undirected graph correlation matrix S.
Step 4.5, repeating steps 4.2, 4.3 and 4.4 for multiple times to complete mapping matrix W and target domain data label F t And joint iterative optimization of the undirected graph correlation matrix S.
And 5, inputting the sample matrix X obtained in the step 2 into the objective function subjected to iterative optimization in the step 4 to obtain a corresponding predicted value label, wherein the predicted value label is the emotional state of the testee corresponding to the sample at the acquisition moment.
Figure BDA0003030687820000091
TABLE 1
Figure BDA0003030687820000092
Figure BDA0003030687820000101
TABLE 2
As can be seen from the data in the above two tables, the recognition accuracy of the result of this embodiment is higher than that of other migration methods.

Claims (7)

1. The electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation is characterized by comprising the following steps: the method specifically comprises the following steps:
step 1, collecting electroencephalogram data of a testee in K different emotional states;
step 2, preprocessing and extracting characteristics of the electroencephalogram data acquired in the step 1, wherein each sample matrix X consists of electroencephalogram characteristics of a testee, and a label vector y is an emotion label corresponding to the electroencephalogram characteristics in the sample matrix X; selecting two different sample matrixes as source domain data X respectively s And target domain data X t ；
Step 3, constructing a learning model for joint feature transfer learning and state estimation, and integrating a single mapping domain adaptation model and a semi-supervised label propagation model based on an undirected graph into a unified frame to obtain a joint optimization objective function; the method specifically comprises the following steps:
step 3.1, establishing a single mapping domain adaptive model:
Figure FDA0004079916660000011
s.t.W T XHX T W＝I (1)
wherein the content of the first and second substances,
Figure FDA0004079916660000012
in order to map the source domain data and the target domain data to a mapping matrix in the same shared subspace, p represents a target dimension, and d represents an original dimension; />
Figure FDA0004079916660000013
To augment the source domain data tag, Y s Is a source domain label, and c is the number of categories; />
Figure FDA0004079916660000014
To augment the target domain data tag, n s 、n t The number of samples of the source domain data and the target domain data respectively; n = n s +n t Representing the total number of samples; />
Figure FDA0004079916660000015
Figure FDA0004079916660000016
Is a diagonal matrix, N s 、N t Are diagonal matrices, wherein the kth diagonal element is the reciprocal of the number of kth source domain or target domain data samples, k =1, 2., c +1; wherein +>
Figure FDA0004079916660000017
Is a central matrix, I is a unit matrix, and>
Figure FDA0004079916660000018
representing the calculation of the Frobenius norm; superscript T denotes transpose;
step 3.2, establishing a semi-supervised label propagation model based on an undirected graph:
Figure FDA0004079916660000019
Figure FDA00040799166600000110
wherein S is ij Representing the element of the ith row and the jth column of the undirected graph correlation matrix S, namely the correlation degree between the ith sample and the jth sample; alpha and gamma are parameters;
Figure FDA0004079916660000021
is a Laplace transform matrix, L = S-D, D is an n-dimensional diagonal matrix, the ith diagonal element->
Figure FDA0004079916660000022
Figure FDA0004079916660000023
For label matrix for label propagation, F s 、F t Labels for the source domain and the target domain, respectively; 1 is a full 1 matrix; tr (-) represents the trace of the matrix; />
Figure FDA0004079916660000024
Represents the calculation of a 2 norm;
step 3.3, combining the two models constructed in the steps 3.1 and 3.2, integrating domain adaptation and undirected graph-based semi-supervised label propagation into a unified framework for joint optimization, wherein the optimization model is as follows:
Figure FDA0004079916660000025
Figure FDA0004079916660000026
lambda, alpha and gamma are parameters;
step 4, initializing a target domain data label F t Constructing and obtaining an initial undirected graph by using a neural network; then according to the target function of the joint optimization obtained in the step 3, a mapping matrix W and a target domain data label F are sequentially subjected to a method of fixing two variables and updating the other variable t Optimizing the undirected graph incidence matrix S, and repeating the optimization process for multiple times to realize joint iterative optimization;
and 5, inputting the sample matrix X obtained in the step 2 into the objective function subjected to iterative optimization in the step 4 to obtain a corresponding predicted value label, wherein the predicted value label is the emotional state of the testee corresponding to the sample at the acquisition moment.
2. The electroencephalogram emotion recognition method combining feature migration and map semi-supervised label propagation as recited in claim 1, wherein: sampling the acquired electroencephalogram data at the frequency of 200Hz, and then filtering noise and artifacts by passing the sampled data through a 1-75 Hz band-pass filter; and the method is divided into five frequency bands of 1-4Hz, 4-8Hz, 8-14Hz, 14-31Hz and 31-50Hz, and differential entropy under each frequency band is respectively calculated to be used as the electroencephalogram characteristic in the sample matrix X.
3. The electroencephalogram emotion recognition method combining feature migration and map semi-supervised label propagation as recited in claim 1, wherein: step 4, labeling the target domain data F t Is initialized to 1/K.
4. The electroencephalogram emotion recognition method combining feature migration and map semi-supervised label propagation as recited in claim 1, wherein: the initial undirected graph is constructed from the KNN network.
5. The electroencephalogram emotion recognition method combining feature migration and map semi-supervised label propagation as recited in claim 1, wherein: the specific method of the joint iterative optimization in the step 4 comprises the following steps:
step 4.1, fix target domain data tag F t And updating a mapping matrix W by using an undirected graph correlation matrix S, wherein the target function is as follows:
Figure FDA0004079916660000031
Figure FDA0004079916660000032
solving the formula (4) to obtain an updated mapping matrix;
step 4.2, fixing the mapping matrix W and the undirected graph incidence matrix S, and updating the target domain data label F t The objective function is:
Figure FDA0004079916660000033
s.t.F≥0，F1＝1 (5)
solving the formula (5) to obtain an updated target domain data label;
step 4.3, fix target domain data label F t And a mapping matrix W, updating an undirected graph correlation matrix S, wherein the target function is as follows:
Figure FDA0004079916660000034
Figure FDA0004079916660000035
/>
solving a formula (6) to obtain an updated undirected graph correlation matrix S;
step 4.4, repeating steps 4.1, 4.2 and 4.3 for multiple times to complete mapping matrix W and target domain data label F t And joint iterative optimization of the undirected graph correlation matrix S.
6. The electroencephalogram emotion recognition method combining feature migration and map semi-supervised label propagation as recited in claim 5, wherein: in step 4.1, the objective function of the optimized mapping matrix W is solved through a Lagrangian function.
7. The electroencephalogram emotion recognition method combining feature migration and map semi-supervised label propagation as recited in claim 5, wherein: simplifying and optimizing target domain data label F through line solving method in steps 4.2 and 4.3 t And solving after the objective function of the undirected graph incidence matrix S.
CN202110428950.4A 2021-04-21 2021-04-21 Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation Active CN113157094B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110428950.4A CN113157094B (en) 2021-04-21 2021-04-21 Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110428950.4A CN113157094B (en) 2021-04-21 2021-04-21 Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation

Publications (2)

Publication Number Publication Date
CN113157094A CN113157094A (en) 2021-07-23
CN113157094B true CN113157094B (en) 2023-04-07

Family

ID=76867491

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110428950.4A Active CN113157094B (en) 2021-04-21 2021-04-21 Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation

Country Status (1)

Country Link
CN (1) CN113157094B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113516231B (en) * 2021-08-10 2024-03-29 大连海事大学 Daily behavior migration identification method based on DSN deep countermeasure migration network
CN114330422B (en) * 2021-12-08 2024-04-12 杭州电子科技大学 Cross-test migration learning method for estimating brain-electricity emotion characteristics in real time
CN114343674B (en) * 2021-12-22 2024-05-03 杭州电子科技大学 Combined discrimination subspace mining and semi-supervised electroencephalogram emotion recognition method

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112488081A (en) * 2020-12-23 2021-03-12 杭州电子科技大学 Electroencephalogram mental state detection method based on DDADSM (distributed denial of service) cross-test transfer learning

Also Published As

Publication number Publication date
CN113157094A (en) 2021-07-23

Similar Documents

Publication Publication Date Title
CN113157094B (en) Electroencephalogram emotion recognition method combining feature migration and graph semi-supervised label propagation
Shi et al. Feature extraction for incomplete data via low-rank tensor decomposition with feature regularization
CN113052113B (en) Depression identification method and system based on compact convolutional neural network
Shoaib et al. Signal processing with direct computations on compressively sensed data
Saboksayr et al. EEG-based emotion classification using graph signal processing
CN108470182B (en) Brain-computer interface method for enhancing and identifying asymmetric electroencephalogram characteristics
CN115919330A (en) EEG Emotional State Classification Method Based on Multi-level SE Attention and Graph Convolution
Devi et al. Deep learning-based cognitive state prediction analysis using brain wave signal
Gao et al. EEG classification based on sparse representation and deep learning
Anil Kumar et al. Performance analysis of blind source separation using canonical correlation
CN114330422A (en) Cross-test migration learning method for estimating electroencephalogram emotional characteristics in real time
Santana et al. Extraction of signals with specific temporal structure using kernel methods
CN114818822A (en) Electroencephalogram migration emotion recognition method combining semi-supervised regression and icon label propagation
CN115969392A (en) Cross-period brainprint recognition method based on tensor frequency space attention domain adaptive network
Gersch Nearest neighbor rule classification of stationary and nonstationary time series
CN114343674A (en) Combined judgment subspace mining and semi-supervised electroencephalogram emotion recognition method
CN111265214B (en) Electroencephalogram signal analysis method based on data structured decomposition
Bich et al. Aggressively prunable MAM²-based Deep Neural Oracle for ECG acquisition by Compressed Sensing
Sattar et al. Time-scale-based segmentation for degraded PCG signals using NMF
CN111990992A (en) Electroencephalogram-based autonomous movement intention identification method and system
Pruttiakaravanich et al. A review on dependence measures in exploring brain networks from fMRI data
CN110765978B (en) Channel selection method based on fractal dimension
CN117708682B (en) Intelligent brain wave acquisition and analysis system and method
Al-Anbary et al. A PROPOSED SIGN LANGUAGE MODEL FOR SPEECHLESS PERSONS USING EEG SIGNALS
Wang Adaptive Fourier Decomposition: Fast Implementation and Extension to Multi-channel Signal Analysis

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
GR01 Patent grant
GR01 Patent grant