CN109740544A - The recognition methods of sense of hearing attention state degree of awakening, device and storage medium - Google Patents
The recognition methods of sense of hearing attention state degree of awakening, device and storage medium Download PDFInfo
- Publication number
- CN109740544A CN109740544A CN201910013388.1A CN201910013388A CN109740544A CN 109740544 A CN109740544 A CN 109740544A CN 201910013388 A CN201910013388 A CN 201910013388A CN 109740544 A CN109740544 A CN 109740544A
- Authority
- CN
- China
- Prior art keywords
- awakening
- signal
- majority filter
- time series
- eeg signals
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a kind of recognition methods of sense of hearing attention state degree of awakening, device and storage medium based on EEG signals, comprising: collect required EEG signals;The first order majority filter device of EEG signals and training process building based on acquisition, carries out feature extraction of the first order based on set empirical mode decomposition and majority filter;Based on the second level majority filter device that first order feature extracted signals and training process construct, feature extraction of the second level based on set empirical mode decomposition and majority filter is carried out;Based on second level feature extracted signals, the feature vector based on variance statistic amount is carried out to extraction feature signal and is calculated;Based on the Machine learning classifiers that feature vector calculated result and training process construct, the sense of hearing attention state degree of awakening based on EEG signals in test process is extracted.The present invention realizes the degree of awakening identification of sense of hearing attention state, helps to improve sense of hearing attention state degree of awakening accuracy of identification and identification validity.
Description
Technical field
The present invention relates to EEG feature extractions and mode identification technology, more particularly to one kind to be based on EEG signals
The recognition methods of sense of hearing attention state degree of awakening, device and storage medium.
Background technique
Emotion is that the attitude whether people meets the needs of oneself to objective things and generate is experienced, it combines the sense of people
Feel, the state of thought and act.Currently, researchers not yet provide unified definition to emotion, but generally agree to: emotion is
Generated under strong nerve impulse and the subjective state inseparable with cerebral cortex, it can make one to generate it is positive or
Passive psychoreaction, so that corresponding body tissue be made to take action.Emotion is complicated, is to have Special Significance to the mankind
A kind of experience of consciousness or experience, and include the reaction of one group of coordination, wherein may include in oral, physiological, behavior
With supraneural mechanism.Currently, three kinds of widely accepted emotion models, are discrete emotion model, dimension emotion model respectively
And the emotion model based on cognitive appraisal.However, going deep into emotional problems research, scholars have found discrete type emotion not
It can reflect complexity and rich of the emotion in expression and transmittance process, for example, people often show in daily communication
The emotion of increasingly complex exquisiteness, such as thinking, disappointed, awkward, appreciation.Therefore, dimension emotion model is gradually by researcher's
Concern.Currently, most common dimension type emotion model is: awakening-potency model (Arousal-Valence model).With feelings
Feel the development of model, although only a few emotion model is still received now, potency-wake-up model is still accounted for
According to main status.Most of dimensional model all has recognised potency and wakes up the presence of the two dimensions.
The sense of hearing of people has certain rule to the perception and understanding of sound, can be divided into the sense of hearing is appreciated that, the sense of hearing pays attention to,
Sense of hearing orientation, auditory discrimination, acoustic memory, sense of hearing selection and audio feedback, eventually form sense of hearing concept, make to acoustic information
Correctly reaction, these stages are to interknit, synergistic.The sense of hearing is note that be that a kind of psychology related with the sense of hearing is living
It is dynamic, it is that people pour into sound, the activity of listening to meet certain psychological goal, it is established on the basis that the sense of hearing is appreciated that
On, and this sound still has meaning to a certain degree to hearer, can just generate the sense of hearing and pay attention to.The sense of hearing pays attention to helping the mankind
Sound interested or important (cocktail party effect) is quickly and accurately extracted from noisy acoustic environment, and is made accordingly
Further reaction.Under sense of hearing attention state, the awakening of hearer's physiological emotion and central nervous system, the feel with particular emotion
Waking up, Degree of Accord Relation is close, and for people carries out listening to specific sense of hearing affairs, the success of Auditory behavior is that hearer is needed to have
There is high degree of awakening.If arousal level is low, it will lead to that auditory response is blunt, judgement is inaccurate, it is easy to Auditory behavior occur
Failure.Therefore, the degree of awakening for how identifying sense of hearing attention state is one very with the research of practical application value.
Currently, mainly including subjective assessment, biological respinse test, bio-signal acquisition, life to the detection method of degree of awakening
Four kinds of main methods such as object chemical method.Wherein physiological signal is the signal of direct reflection human body variation, in degree of awakening detection
Using more and more extensive.Due to can relatively accurately reflect the variation of brain degree of awakening, the research of EEG signals increasingly by
To the concern of scholar.But since EEG signals are generally fainter, and it is easy to be influenced by environment, therefore for brain telecommunications
Number research still belong to the laboratory research stage at present.The existing degree of awakening recognizer based on EEG signals, mark sheet
Show and extract, be primarily present limitation following aspects: accuracy of identification is inadequate, cannot effectively extract useful feature letter
Breath;The nonlinear characteristic in EEG signals cannot be efficiently extracted;The stationarity of EEG signals there are certain requirements, brain telecommunications
Number more unstable, the effective feature mode extracted is more limited.
Summary of the invention
The present invention provides a kind of sense of hearing attention state degree of awakening recognition methods based on EEG signals, device and storage and is situated between
Matter realizes that the degree of awakening of sense of hearing attention state identifies that can effectively improve accuracy of identification and identification has using cognition EEG signals
Effect property.
To achieve the above object, the present invention provides a kind of sense of hearing attention state degree of awakening identification side based on EEG signals
Method, comprising the following steps:
Obtain EEG signals to be tested;
The first order majority filter device of EEG signals and training process building based on the test, carries out first
Feature extraction of the grade based on set empirical mode decomposition and majority filter;
Based on the second level majority filter device that first order feature extracted signals signal and training process construct, carry out
Feature extraction of the second level based on set empirical mode decomposition and majority filter;
Based on second level feature extracted signals, extraction feature signal is carried out based on the feature vector of variance statistic amount
It calculates;
Based on the Machine learning classifiers that feature vector calculated result and training process construct, base in test process is extracted
In the sense of hearing attention state degree of awakening of EEG signals.
Wherein, before described the step of obtaining EEG signals to be tested further include:
Obtain the EEG signals in training process;
Based on the EEG signals in the training process, carries out the first order and be based on set empirical mode decomposition, building first
The feature extraction of grade majority filter device and majority filter, and carry out the second level and be based on set empirical mode decomposition, building
The feature extraction of the second level majority filter device and majority filter;
The feature vector based on variance statistic amount is carried out to extraction feature signal to calculate;
Machine learning classifiers are constructed based on feature vector calculated result.
Wherein, the step of the feature extraction based on set empirical mode decomposition, building majority filter device and majority filter
Suddenly include:
Obtain the time series signal of EEG signals in training process;
Set empirical mode decomposition is carried out to time series signal, time series under high degree of awakening and low degree of awakening is obtained and believes
Number preceding quadravalence (the 1,2,3,4th rank) intrinsic mode functions ingredient;
Using time series signal under high degree of awakening and low degree of awakening preceding quadravalence (the 1,2,3,4th rank) intrinsic mode functions at
Point, construct majority filter device;
The majority filter device completed using building carries out majority filter processing to preceding quadravalence intrinsic mode functions ingredient,
Obtain four time series signals that feature extraction goes out.
Wherein, described the step of carrying out set empirical mode decomposition to time series signal, includes:
It is assumed that the length of window w initialization value of EEG signals phase space reconfiguration is 1, it is N One-dimension Time Series for length
Signal t (n), n=1,2,3, N adds white noise and zero averaging processing, obtains signal x (n);
Determine all local maximums of signal x (n) and minimum;
Using cubic spline curve, the local maximum all to signal x (n) is fitted respectively, forms coenvelope line
env_max(n);The local minizing points all to signal x (n) are fitted, and are formed lower envelope line env_min (n);
Calculate mean value m (n)=(env_max (the n)+env_min (n))/2 of envelope up and down;
Extract detail signal h (n)=t (n)-m (n);
Check whether h (n) meets the stopping criterion for iteration of intrinsic mode functions;
After meeting screening stopping criterion for iteration, w=w+1;First intrinsic mode functions IMF1 (n)=h1, k (n) are obtained,
Residual signal r (n)=x (n)-IMF1 (n);
Judge whether residual signal r (n) meets stop condition;
If finally obtaining residual signal r (n) is that a constant or variation meet preset condition, all iteration mistakes are terminated
Otherwise journey is based on r (n), repeat the second step of above-mentioned process to the 7th step, into next round iteration, until meeting iteration stopping
Condition;
After meeting iteration stopping condition, it is intrinsic to obtain each rank for the set empirical mode decomposition of deadline sequence signal
Modular function component.
Wherein, the preceding quadravalence intrinsic mode functions ingredient using time series signal under high degree of awakening and low degree of awakening,
Construct majority filter device the step of include:
Obtain the preceding quadravalence intrinsic mode functions ingredient of time series signal under high degree of awakening and low degree of awakening;
According to the preceding quadravalence intrinsic mode functions ingredient of time series signal under high degree of awakening and low degree of awakening, it is empty to find out mixing
Between covariance matrix;
Matrix- eigenvector-decomposition is carried out to the blending space covariance matrix, obtains albefaction eigenvalue matrix;
Majority filter device is constructed based on the albefaction eigenvalue matrix.
Wherein, described the step of carrying out the calculating of the feature vector based on variance statistic amount to extraction feature, includes:
There are four time series signals for the input of second level feature extraction, and each time series signal is by set Empirical Mode
After state is decomposed, quadravalence intrinsic mode functions ingredient carries out majority filter device reduced-dimensions filtering before extracting, and becomes two time series letters
Number;
Calculate separately the variance Z of this two time series signals;
According to mathematical formulae F=log10Eight characteristic values are calculated in (1+Var (Z)), and composition characteristic vector is given next
The Machine learning classifiers of grade.
Wherein, the model that the Machine learning classifiers use includes: support vector machines, linear decision device, neural network
Model.
The present invention also proposes a kind of sense of hearing attention state degree of awakening identification device based on EEG signals, comprising: memory,
Processor and the computer program being stored on the memory, the computer program are realized when being run by the processor
The step of method as described above.
The present invention also proposes a kind of computer readable storage medium, and calculating is stored on the computer readable storage medium
The step of machine program, the computer program realizes method as described above when being run by processor.
Compared with prior art, a kind of sense of hearing attention state degree of awakening identification side based on EEG signals proposed by the present invention
Method, device and storage medium, character representation and extract be it is a kind of based on data driving character representation and extraction, very
It is suitble to non-linear, non stationary state EEG feature extraction, the degree of awakening for how identifying sense of hearing attention state is realized, and mention
High accuracy of identification and identification validity.
Detailed description of the invention
Fig. 1 is the system block diagram of the sense of hearing attention state degree of awakening recognizer the present invention is based on EEG signals;
Fig. 2 is the present invention figure feature extraction flow chart based on set empirical mode decomposition and majority filter;
Fig. 3 is the set empirical mode decomposition flow chart of time series signal according to the present invention;
Fig. 4 is the process of the majority filter device building of time series signal intrinsic mode functions ingredient according to the present invention
Figure;
Fig. 5 is the feature vector calculation method flow chart according to the present invention based on variance statistic amount.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
Specifically, Fig. 1 is please referred to, Fig. 1 is that the sense of hearing attention state degree of awakening proposed by the present invention based on EEG signals is known
The flow diagram of other embodiment of the method.
The embodiment of the present invention proposes a kind of sense of hearing attention state degree of awakening recognition methods based on EEG signals, including following
Step:
S1 obtains EEG signals to be tested;
S2 is carried out the EEG signals first order to be measured and is based on set empirical mode decomposition based on the EEG signals of the test,
The intrinsic mode functions ingredient of EEG signals is obtained, the first order majority filter device constructed using training process realizes brain to be measured
The feature extraction of electric signal intrinsic mode functions ingredient progress majority filter;
S3 is based on the first order feature extraction signal, carries out the second level and is based on set empirical mode decomposition, recycles instruction
The second level majority filter device for practicing process building, the intrinsic mode functions ingredient to the second level based on set empirical mode decomposition,
Carry out the feature extraction of majority filter;
S4 carries out the feature vector based on variance statistic amount and calculates to extraction feature signal;
S5 is realized based on the Machine learning classifiers for calculating resulting feature vector and training process building and is based on brain
The sense of hearing attention state degree of awakening of electric signal identifies.
Further, the S1, before the step of obtaining EEG signals to be tested further include:
S01 obtains trained EEG signals;
S02 is carried out the EEG signals first order and is based on set empirical modal point based on the EEG signals in the training process
Solution, obtains the intrinsic mode functions ingredient of the EEG signals of the training;Utilize the intrinsic mode functions of the EEG signals of the training
Ingredient constructs first order majority filter device, and the feature extraction of majority filter is carried out to EEG signals intrinsic mode functions ingredient,
And it carries out the second level and is taken out based on the feature of set empirical mode decomposition, building second level majority filter device and majority filter
It takes;
S03 carries out the feature vector based on variance statistic amount and calculates to extraction feature signal;
S04, based on resulting feature vector is calculated, training machine Study strategies and methods realize the structure of Machine learning classifiers
It builds.
Wherein, the step of the feature extraction based on set empirical mode decomposition, building majority filter device and majority filter
Suddenly include:
Obtain the EEG signals in training process;
Set empirical mode decomposition is carried out to EEG signals, obtains preceding four of EEG signals under high degree of awakening and low degree of awakening
Rank (the 1,2,3,4th rank) intrinsic mode functions ingredient;
Using preceding quadravalence (the 1st, 2,3,4 rank) intrinsic mode functions ingredient of EEG signals under high degree of awakening and low degree of awakening,
Construct majority filter device;
Using constructed majority filter device, majority filter processing is carried out to preceding quadravalence intrinsic mode functions ingredient, is obtained
Four time series signals gone out to feature extraction.
Wherein, described the step of carrying out set empirical mode decomposition to EEG signals, includes:
It is assumed that the length of window w initialization value of EEG signals phase space reconfiguration is 1, it is N One-dimension Time Series for length
Signal t (n), n=1,2,3, N adds white noise and zero averaging processing, obtains signal x (n);
Determine all local maximums of signal x (n) and minimum;
Using cubic spline curve, the local maximum all to signal x (n) is fitted respectively, forms coenvelope line
env_max(n);The local minizing points all to signal x (n) are fitted, and are formed lower envelope line env_min (n);
Calculate mean value m (n)=(env_max (the n)+env_min (n))/2 of envelope up and down;
Extract detail signal h (n)=t (n)-m (n);
Check whether h (n) meets the stopping criterion for iteration of intrinsic mode functions;
After meeting screening stopping criterion for iteration, w=w+1;First intrinsic mode functions IMF1 (n)=h1, k (n) are obtained,
Residual signal r (n)=x (n)-IMF1 (n);
Judge whether residual signal r (n) meets stop condition;
If finally obtaining residual signal r (n) is that a constant or variation meet preset condition, all iteration mistakes are terminated
Otherwise journey is based on r (n), repeat the second step of above-mentioned process to the 7th step, into next round iteration, until meeting iteration stopping
Condition;
After meeting iteration stopping condition, the set empirical mode decomposition of EEG signals is completed, obtains each rank eigen mode letter
Number component.
Wherein, the preceding quadravalence intrinsic mode functions ingredient using EEG signals under high degree of awakening and low degree of awakening, building
The step of majority filter device includes:
Obtain the preceding quadravalence intrinsic mode functions ingredient of EEG signals under high degree of awakening and low degree of awakening;
According to the preceding quadravalence intrinsic mode functions ingredient of EEG signals under high degree of awakening and low degree of awakening, blending space association is found out
Variance matrix;
Matrix- eigenvector-decomposition is carried out to the blending space covariance matrix, obtains albefaction eigenvalue matrix;
Majority filter device is constructed based on the albefaction eigenvalue matrix.
Wherein, described to extraction feature signal, carrying out the step of feature vector based on variance statistic amount calculates includes:
There are four time series signals for the input of second level feature extraction, and each time series signal is by set Empirical Mode
State is decomposed, and quadravalence (the 1st, 2,3,4 rank) intrinsic mode functions ingredient carries out majority filter device reduced-dimensions filtering before extracting, and obtains two
Time series signal;
Calculate separately the variance Z of this two time series signals;
According to mathematical formulae F=log10Eight characteristic values are calculated in (1+Var (Z)), and composition characteristic vector is given next
The Machine learning classifiers of grade.
Wherein, the model that the Machine learning classifiers use includes: support vector machines, linear decision device, neural network
Model.
The sense of hearing attention state degree of awakening recognizer based on EEG signals that the invention proposes a kind of should be based on brain telecommunications
Number degree of awakening recognizer, character representation and extract be it is a kind of based on data driving character representation and extraction,
Main advantage is as follows:
The identification of the high degree of awakening, low degree of awakening of sense of hearing attention state is directly realized using EEG signals;
The pattern feature of sense of hearing attention state degree of awakening based on EEG signals is not only based on set empirical mode decomposition
The character representation realized with majority filter and extraction;Additionally using a kind of cascade mode and constructing a kind of depth characteristic indicates
It is character representation and extraction of the one kind efficiently based on data driving with extraction frame;
Sense of hearing attention state degree of awakening recognizer proposed by the present invention based on EEG signals, due to being a kind of based on certainly
The schema extraction algorithm of body data-driven is very suitable to non-linear, non stationary state EEG feature extraction.
Below in conjunction with attached drawing, technical solution of the present invention and embodiment are described in detail.
Referring to Fig.1, Fig. 1 is the system frame of the sense of hearing attention state degree of awakening recognizer the present invention is based on EEG signals
Figure.The system block diagram of the sense of hearing attention state degree of awakening recognizer, is mainly formed by training and testing two processes.It trained
Journey relates generally to 4 modules, is respectively: feature extraction of the first order based on set empirical mode decomposition and majority filter;The
Feature extraction of the second level based on set empirical mode decomposition and majority filter;Feature vector based on variance statistic amount calculates;
Machine learning classifiers.The major function of training process is building and the engineering for realizing the first, second majority filter device
Practise the training of classifier.Test process, the block process of design and the entirety of training process are identical, and mainly difference is: instruction
Practice the first, second majority filter device and Machine learning classifiers of process building, directly tested process uses;Test process
The building of the first, second majority filter device and the building of Machine learning classifiers are not will do it.Therefore, it is based on training process
Model parameter, it can be achieved that sense of hearing attention state high degree of awakening and low degree of awakening based on EEG signals identification.
The system block diagram of the sense of hearing attention state degree of awakening recognizer based on EEG signals described in Fig. 1, the first order
It is a kind of combination set empirical mode decomposition and master based on the feature extraction module of set empirical mode decomposition and majority filter
The feature extraction algorithm based on data driving that ingredient filtering is realized, flow chart are as shown in Figure 2.The first order is based on set
The feature extraction module of empirical mode decomposition and majority filter, three subprocess related generally to: the first subprocess, time sequence
The set empirical mode decomposition of column signal;The present invention, the second subprocess utilize preceding quadravalence (the 1st, 2,3,4 rank) intrinsic mode functions
Ingredient constructs majority filter device;Third subprocess, the majority filter device completed using building, to preceding quadravalence eigen mode letter
Number ingredient implementation processing, to realize feature extraction of the first order based on set empirical mode decomposition and majority filter.
The system block diagram of the sense of hearing attention state degree of awakening recognizer based on EEG signals described in Fig. 1, brain telecommunications
Feature extraction number by the first order based on set empirical mode decomposition and majority filter, the dimension of time series signal by
One-dimension Time Series signal originally becomes 4 dimension time series signals.Then, by first order feature extraction treated 4
Time series signal, then carry out feature extraction of the second level based on set empirical mode decomposition and majority filter.Second level base
In the feature extraction module of set empirical mode decomposition and majority filter, the method for realization and being passed through based on set for the first order
It is identical with the Feature Extraction Algorithm process of majority filter to test mode decomposition.The first order and second level feature extracting method pass through grade
Connection mode is effectively combined into a kind of depth characteristic extraction model.This depth characteristic extracts model, is a kind of combination set experience
Automatic Feature Extraction method that mode decomposition and majority filter are realized, based on data driving.
Next, three of the feature extraction module to first (two) grade based on set empirical mode decomposition and majority filter
The technic relization scheme of a subprocess is described in detail.
It further, is the detailed protocol gathering the subprocess technology of empirical mode decomposition and realizing in Fig. 2, as shown in Figure 3.It is main
It will be by following below scheme:
It is assumed that the length of window w initialization value of EEG signals phase space reconfiguration is 1, it is N One-dimension Time Series for length
Signal t (n), n=1,2,3, N;
White noise and zero averaging processing are added, signal x (n) is obtained;
Determine all local maximums of signal x (n) and minimum;
Using cubic spline curve, the local maximum all to signal x (n) is fitted respectively, forms coenvelope line
env_max(n);The local minizing points all to signal x (n) are fitted, and are formed lower envelope line env_min (n);
Calculate mean value m (n)=(env_max (the n)+env_min (n))/2 of envelope up and down;
Extract detail signal h (n)=t (n)-m (n);
Check whether h (n) meets the stopping criterion for iteration of intrinsic mode functions:
h1,k(i) value of the kth time iteration of the 1st detail signal is indicated, SD is that iteration screening threshold value (generally takes 0.2-
0.3), m1,1For the mean value of upper lower envelope, detail signal h1,k(i) initial value is that envelope mean value obtains above and below x (n) is subtracted.
Value of the embodiment of the present invention 0.2, when SD is less than 0.2, the screening iteration ends of epicycle intrinsic mode functions component.
After meeting screening stopping criterion for iteration, w=w+1;First intrinsic mode functions IMF1 (n)=h1, k (n) are obtained, is remained
Remaining signal r (n)=x (n)-IMF1 (n).
Judge whether residual signal r (n) meets stop condition.
If finally obtaining residual signal r (n) is that a constant or variation are sufficiently small, all iterative process are terminated, otherwise,
Based on r (n), item of the process second step of above-mentioned process to the 7th step, into next round iteration, until meeting iteration stopping is repeated
Part.
After meeting iteration stopping condition, the set empirical mode decomposition of deadline sequence signal obtains each rank eigen mode
Function component.
Further, the detailed protocol that the subprocess technology of the building of majority filter device is realized is constructed in Fig. 2, such as Fig. 4 institute
Show.Relate generally to following steps:
The first step is asked according to the preceding quadravalence intrinsic mode functions ingredient of time series signal under high degree of awakening and low degree of awakening
Blending space covariance matrix out.Assuming that under high degree of awakening and low degree of awakening time series signal preceding quadravalence intrinsic mode functions at
Matrix composed by point, is respectively as follows: IMF41And IMF42, length is the physiological time sequence of N, then for IMF41And IMF42Table
The matrix dimensionality shown is 4 × N.
Second step solves IMF41And IMF42Normalized covariance matrix, respectively R1And R2, specific mathematical table
It is as follows up to formula,
Wherein, on trace () representing matrix diagonal line element sum.
Third step, acquiring blending space covariance matrix R is,
WithThe unusual spectral component IMF4 of physiological time sequence under respectively two kinds of physiological status1And IMF42It is average
Covariance matrix.
4th step carries out feature decomposition, finds out albefaction eigenvalue matrix to blending space covariance matrix R.First to mixing
Space covariance matrix R carries out Eigenvalues Decomposition, and U and λ are respectively as follows: eigenvectors matrix and its corresponding eigenvalue matrix is (special
The characteristic value of value indicative matrix, is arranged in decreasing order).
R=U × λ × UT (5)
Then, albefaction value matrix P can be expressed as follows:
5th step constructs majority filter device.Based on albefaction value matrix, to matrix R1And R2Carry out such as down conversion:
S1=P × R1×PT (7)
S2=P × R2×PT (8)
Then, to matrix S1And S2Feature decomposition is done, is had,
It can prove matrix S1Feature vector and matrix S2Eigenvectors matrix be equal, that is,
B1=B2=B (11)
At the same time, the sum of diagonal matrix λ 1 and λ 2 of two characteristic values are unit matrix, it may be assumed that
λ 1+ λ 2=I (12)
It is always 1 since the characteristic value of two matroids is added, then S1Maximum eigenvalue corresponding to feature vector make S2Have
The smallest characteristic value, vice versa.Albefaction physiological time sequence is to feature vector corresponding with the maximum eigenvalue in λ 1 and λ 2
Transformation, for separation two signal matrix in variance be optimal.
Therefore, optimal majority filter device W can be constructed at this time, mathematical form is,
W=BT×P (13)
Further, preceding quadravalence intrinsic mode functions ingredient implementation is handled using the majority filter device that building is completed in Fig. 2
Subprocess, main implementer's case, such as mathematic(al) representation below:
Training process filters the time series signal of high degree of awakening and low degree of awakening by the principal component of construction
After wave device W processing, the characteristic Z 1 and Z2 extracted are as follows:
Z1=W × IMF41 (14)
Z2=W × IMF42 (15)
For test process, the time series signal of test is obtained after the majority filter device W of construction processing
To characteristic Z _ test of extraction are as follows:
Z_test=W × IMF4_test (16)
In the system block diagram of the sense of hearing attention state degree of awakening recognizer based on EEG signals described in Fig. 1, it is based on
The Implementation Technology that the feature vector of variance statistic amount calculates, as shown in Figure 5.When the input of second level feature extraction has 4
Between sequence signal, each time series after gathering empirical mode decomposition, extract before quadravalence intrinsic mode functions ingredient led
After filter-divider reduced-dimensions filtering, two time series signals are obtained, the side of this two time series signals is then calculated separately
Poor Z.Further according to mathematical formulae F=log108 characteristic values can be obtained in (1+Var (Z)), and composition characteristic vector gives next stage
Machine learning classifiers.
In the system block diagram of the sense of hearing attention state degree of awakening recognizer based on EEG signals described in Fig. 1, machine
The technic relization scheme of Study strategies and methods can use classical taxonomy device model, such as support vector machines, linear decision device, nerve
Network etc..
In addition, the present invention also proposes a kind of sense of hearing attention state degree of awakening identification device based on EEG signals, comprising: deposit
Reservoir, processor and the computer program being stored on the memory, the computer program are run by the processor
The step of Shi Shixian method as described above.
In addition, the present invention also proposes a kind of computer readable storage medium, stored on the computer readable storage medium
There is the step of computer program, the computer program realizes method as described above when being run by processor.
Compared with prior art, a kind of sense of hearing attention state degree of awakening identification side based on EEG signals proposed by the present invention
Method, device and storage medium, character representation and extract be it is a kind of based on data driving character representation and extraction, realize
How to identify the degree of awakening of sense of hearing attention state, and improves accuracy of identification and identification validity.
The above description is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all utilizations
Equivalent structure made by description of the invention and accompanying drawing content or process transformation, are applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (9)
1. a kind of sense of hearing attention state degree of awakening recognition methods based on EEG signals, which comprises the following steps:
Obtain EEG signals to be tested;
Based on the EEG signals to be tested, carries out the first order and be based on set empirical mode decomposition, recycle training process structure
The first order majority filter device built carries out majority filter to first order set empirical mode decomposition intrinsic mode functions component
Feature extraction;
Based on first order feature extracted signals, carries out the second level and be based on set empirical mode decomposition, recycle training process building
Second level majority filter device, to the second level set empirical mode decomposition intrinsic mode functions component carry out majority filter spy
Sign extracts;
Based on second level feature extracted signals, the feature vector based on variance statistic amount is carried out to extraction feature signal and is calculated;
Based on the Machine learning classifiers that feature vector calculated result and training process construct, extracts and be based on brain in test process
The sense of hearing attention state degree of awakening of electric signal.
2. the method according to claim 1, wherein before described the step of obtaining EEG signals to be tested also
Include:
Obtain the EEG signals in training process;
Based on the EEG signals in the training process, carries out the first order and be based on set empirical mode decomposition, building first order master
At the feature extraction of filter-divider and majority filter, and carries out the second level and be based on set empirical mode decomposition, building second
The feature extraction of grade majority filter device and majority filter;
The feature vector based on variance statistic amount is carried out to extraction feature signal to calculate;
Machine learning classifiers are constructed based on feature vector calculated result.
3. according to the method described in claim 2, it is characterized in that, based on set empirical mode decomposition, building majority filter
The step of feature extraction of device and majority filter includes:
Obtain the time series signal of EEG signals in training process;
Set empirical mode decomposition is carried out to time series signal, obtains time series signal under high degree of awakening and low degree of awakening
Preceding quadravalence (the 1,2,3,4th rank) intrinsic mode functions ingredient;
Using preceding quadravalence (the 1st, 2,3,4 rank) intrinsic mode functions ingredient of time series signal under high degree of awakening and low degree of awakening,
Construct majority filter device;
The majority filter device completed using building is carried out majority filter processing to preceding quadravalence intrinsic mode functions ingredient, obtained
Four time series signals that feature extraction goes out.
4. according to the method described in claim 3, it is characterized in that, described carry out set empirical modal point to time series signal
The step of solution includes:
It is assumed that the length of window w initialization value of EEG signals phase space reconfiguration is 1, it is N One-dimension Time Series signal t for length
(n), n=1,2,3, N adds white noise and zero averaging processing, obtains signal x (n);
Determine all local maximums of signal x (n) and minimum;
Using cubic spline curve, the local maximum all to signal x (n) is fitted respectively, forms coenvelope line env_
max(n);The local minizing points all to signal x (n) are fitted, and are formed lower envelope line env_min (n);
Calculate mean value m (n)=(env_max (the n)+env_min (n))/2 of envelope up and down;
Extract detail signal h (n)=t (n)-m (n);
Check whether h (n) meets the stopping criterion for iteration of intrinsic mode functions;
After meeting screening stopping criterion for iteration, w=w+1;First intrinsic mode functions IMF1 (n)=h1, k (n) are obtained, it is remaining
Signal r (n)=x (n)-IMF1 (n);
Judge whether residual signal r (n) meets stop condition;
If finally obtaining residual signal r (n) is that a constant or variation meet preset condition, all iterative process are terminated,
Otherwise, it is based on r (n), repeats the second step of above-mentioned process to the 7th step, into next round iteration, until meeting iteration stopping
Condition;
After meeting iteration stopping condition, the set empirical mode decomposition of deadline sequence signal obtains each rank eigen mode letter
Number component.
5. according to the method described in claim 4, it is characterized in that, described utilize time series under high degree of awakening and low degree of awakening
The preceding quadravalence intrinsic mode functions ingredient of signal, construct majority filter device the step of include:
Obtain the preceding quadravalence intrinsic mode functions ingredient of time series signal under high degree of awakening and low degree of awakening;
According to the preceding quadravalence intrinsic mode functions ingredient of time series signal under high degree of awakening and low degree of awakening, blending space association is found out
Variance matrix;
Matrix- eigenvector-decomposition is carried out to the blending space covariance matrix, obtains albefaction eigenvalue matrix;
Majority filter device is constructed based on the albefaction eigenvalue matrix.
6. according to the method described in claim 4, it is characterized in that, described carry out based on variance statistic amount extraction feature signal
Feature vector calculate the step of include:
There are four time series signals for the input of second level feature extraction, and each time series signal is by set empirical modal point
Xie Hou, quadravalence intrinsic mode functions ingredient carries out majority filter device reduced-dimensions filtering before extracting, and obtains two time series signals;
Calculate separately the variance Z of this two time series signals;
According to mathematical formulae F=log10Eight characteristic values are calculated in (1+Var (Z)), and composition characteristic vector gives next stage
Machine learning classifiers.
7. according to the method described in claim 6, it is characterized in that, the model that the Machine learning classifiers use includes: branch
Hold vector machine, linear decision device, neural network model.
8. a kind of sense of hearing attention state degree of awakening identification device based on EEG signals characterized by comprising memory, place
Reason device and the computer program being stored on the memory, are realized such as when the computer program is run by the processor
The step of method of any of claims 1-7.
9. a kind of computer readable storage medium, which is characterized in that be stored with computer on the computer readable storage medium
Program is realized when the computer program is run by processor such as the step of method of any of claims 1-7.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910013388.1A CN109740544B (en) | 2019-01-07 | 2019-01-07 | Auditory attention state arousal degree identification method and device and storage medium |
PCT/CN2019/117074 WO2020143300A1 (en) | 2019-01-07 | 2019-11-11 | Auditory attention state arousal level recognition method and apparatus, and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910013388.1A CN109740544B (en) | 2019-01-07 | 2019-01-07 | Auditory attention state arousal degree identification method and device and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109740544A true CN109740544A (en) | 2019-05-10 |
CN109740544B CN109740544B (en) | 2021-09-07 |
Family
ID=66363710
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910013388.1A Expired - Fee Related CN109740544B (en) | 2019-01-07 | 2019-01-07 | Auditory attention state arousal degree identification method and device and storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109740544B (en) |
WO (1) | WO2020143300A1 (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020143300A1 (en) * | 2019-01-07 | 2020-07-16 | 哈尔滨工业大学(深圳) | Auditory attention state arousal level recognition method and apparatus, and storage medium |
CN112347087A (en) * | 2020-09-30 | 2021-02-09 | 中国铁路上海局集团有限公司 | Quality control method for wind speed observation data along high-speed rail |
CN112364710A (en) * | 2020-10-20 | 2021-02-12 | 西安理工大学 | Plant electric signal classification and identification method based on deep learning algorithm |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112989918A (en) * | 2020-12-25 | 2021-06-18 | 北京工业大学 | Electroencephalogram signal online prediction method based on kernel recursive least square adaptive tracking algorithm |
CN114218986B (en) * | 2021-12-10 | 2024-05-07 | 中国航空综合技术研究所 | State classification method based on EEG (electroencephalogram) data |
CN115034272A (en) * | 2022-06-29 | 2022-09-09 | 上海术理智能科技有限公司 | Motor imagery electroencephalogram signal classification method based on iterative learning |
CN116172580B (en) * | 2023-04-20 | 2023-08-22 | 华南理工大学 | Auditory attention object decoding method suitable for multi-sound source scene |
CN117349603B (en) * | 2023-12-06 | 2024-03-12 | 小舟科技有限公司 | Adaptive noise reduction method and device for electroencephalogram signals, equipment and storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6631325B1 (en) * | 1996-08-12 | 2003-10-07 | The United States As Represented By The Administrator Of The National Aeronautics And Space Administration | Computer implemented empirical mode decomposition method apparatus, and article of manufacture utilizing curvature extrema |
CN102568205A (en) * | 2012-01-10 | 2012-07-11 | 吉林大学 | Traffic parameter short-time prediction method based on empirical mode decomposition and classification combination prediction in abnormal state |
CN104091172A (en) * | 2014-07-04 | 2014-10-08 | 北京工业大学 | Characteristic extraction method of motor imagery electroencephalogram signals |
US20150065813A1 (en) * | 2013-09-03 | 2015-03-05 | SILVERMEDIA Spolka z ograniczona odpowiedzialnoscia | System for recording and processing signal for diagnosing auditory system and method for recording and processing signal for diagnosing auditory system |
CN104833535A (en) * | 2015-05-15 | 2015-08-12 | 西南交通大学 | Railway vehicle tire tread scratch detection method |
CN107239142A (en) * | 2017-06-01 | 2017-10-10 | 南京邮电大学 | A kind of EEG feature extraction method of combination public space pattern algorithm and EMD |
CN108888264A (en) * | 2018-05-03 | 2018-11-27 | 南京邮电大学 | EMD and CSP merges power spectral density brain electrical feature extracting method |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108038429A (en) * | 2017-11-30 | 2018-05-15 | 西安交通大学 | A kind of single brain electrical feature extraction sorting technique of Motor execution |
CN109039500A (en) * | 2018-08-28 | 2018-12-18 | 广东工业大学 | A kind of frequency spectrum sensing method, device, equipment, system and storage medium |
CN109740544B (en) * | 2019-01-07 | 2021-09-07 | 哈尔滨工业大学(深圳) | Auditory attention state arousal degree identification method and device and storage medium |
-
2019
- 2019-01-07 CN CN201910013388.1A patent/CN109740544B/en not_active Expired - Fee Related
- 2019-11-11 WO PCT/CN2019/117074 patent/WO2020143300A1/en active Application Filing
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6631325B1 (en) * | 1996-08-12 | 2003-10-07 | The United States As Represented By The Administrator Of The National Aeronautics And Space Administration | Computer implemented empirical mode decomposition method apparatus, and article of manufacture utilizing curvature extrema |
CN102568205A (en) * | 2012-01-10 | 2012-07-11 | 吉林大学 | Traffic parameter short-time prediction method based on empirical mode decomposition and classification combination prediction in abnormal state |
US20150065813A1 (en) * | 2013-09-03 | 2015-03-05 | SILVERMEDIA Spolka z ograniczona odpowiedzialnoscia | System for recording and processing signal for diagnosing auditory system and method for recording and processing signal for diagnosing auditory system |
CN104091172A (en) * | 2014-07-04 | 2014-10-08 | 北京工业大学 | Characteristic extraction method of motor imagery electroencephalogram signals |
CN104833535A (en) * | 2015-05-15 | 2015-08-12 | 西南交通大学 | Railway vehicle tire tread scratch detection method |
CN107239142A (en) * | 2017-06-01 | 2017-10-10 | 南京邮电大学 | A kind of EEG feature extraction method of combination public space pattern algorithm and EMD |
CN108888264A (en) * | 2018-05-03 | 2018-11-27 | 南京邮电大学 | EMD and CSP merges power spectral density brain electrical feature extracting method |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020143300A1 (en) * | 2019-01-07 | 2020-07-16 | 哈尔滨工业大学(深圳) | Auditory attention state arousal level recognition method and apparatus, and storage medium |
CN112347087A (en) * | 2020-09-30 | 2021-02-09 | 中国铁路上海局集团有限公司 | Quality control method for wind speed observation data along high-speed rail |
CN112347087B (en) * | 2020-09-30 | 2024-05-03 | 中国铁路上海局集团有限公司 | Quality control method for high-speed rail wind speed observation data |
CN112364710A (en) * | 2020-10-20 | 2021-02-12 | 西安理工大学 | Plant electric signal classification and identification method based on deep learning algorithm |
CN112364710B (en) * | 2020-10-20 | 2024-04-05 | 西安理工大学 | Plant electric signal classification and identification method based on deep learning algorithm |
Also Published As
Publication number | Publication date |
---|---|
CN109740544B (en) | 2021-09-07 |
WO2020143300A1 (en) | 2020-07-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109740544A (en) | The recognition methods of sense of hearing attention state degree of awakening, device and storage medium | |
CN111461176B (en) | Multi-mode fusion method, device, medium and equipment based on normalized mutual information | |
CN103294199B (en) | A kind of unvoiced information identifying system based on face's muscle signals | |
CN105877766B (en) | A kind of state of mind detection system and method based on the fusion of more physiological signals | |
Kello et al. | The pervasiveness of 1/f scaling in speech reflects the metastable basis of cognition | |
CN104871160B (en) | System and method for feeling and recognizing anatomy | |
CN110604565A (en) | Brain health screening method based on portable EEG equipment | |
CN106569604B (en) | Audiovisual bimodal semantic matches and semantic mismatch collaboration stimulation brain-machine interface method | |
CN110070105B (en) | Electroencephalogram emotion recognition method and system based on meta-learning example rapid screening | |
Hendrix et al. | Distinct ERP signatures of word frequency, phrase frequency, and prototypicality in speech production. | |
CN107066514A (en) | The Emotion identification method and system of the elderly | |
CN108577865A (en) | A kind of psychological condition determines method and device | |
CN109243491A (en) | Method, system and the storage medium of Emotion identification are carried out to voice on frequency spectrum | |
CN111329474A (en) | Electroencephalogram identity recognition method and system based on deep learning and information updating method | |
CN114533086B (en) | Motor imagery brain electrolysis code method based on airspace characteristic time-frequency transformation | |
Noroozi et al. | Supervised vocal-based emotion recognition using multiclass support vector machine, random forests, and adaboost | |
CN113191225B (en) | Emotion electroencephalogram recognition method and system based on graph attention network | |
CN103258545A (en) | Pathological voice subdivision method | |
Holt et al. | Work cognitions in multicultural interaction | |
CN112185493A (en) | Personality preference diagnosis device and project recommendation system based on same | |
CN109657646A (en) | The character representation and extracting method, device and storage medium of physiological time sequence | |
Trammel et al. | Decoding semantic relatedness and prediction from EEG: A classification method comparison | |
CN111466930A (en) | Audio-visual evoked emotion recognition method and system based on electroencephalogram signals | |
CN101840506B (en) | Remote education student characteristic signal extraction and recognition method | |
Eqlimi | Exploring neural markers modulated by learning from speech in environmental noise using single-trial EEG |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210907 Termination date: 20220107 |
|
CF01 | Termination of patent right due to non-payment of annual fee |