CN107423668B - Electroencephalogram signal classification system and method based on wavelet transformation and sparse expression - Google Patents
Electroencephalogram signal classification system and method based on wavelet transformation and sparse expression Download PDFInfo
- Publication number
- CN107423668B CN107423668B CN201710244077.7A CN201710244077A CN107423668B CN 107423668 B CN107423668 B CN 107423668B CN 201710244077 A CN201710244077 A CN 201710244077A CN 107423668 B CN107423668 B CN 107423668B
- Authority
- CN
- China
- Prior art keywords
- learning
- dictionary
- classification
- data
- motor imagery
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention relates to an electroencephalogram signal classification system and method based on wavelet transformation and sparse expression, and the method comprises the following specific steps: dictionary learning of the motor imagery electroencephalogram signals in a dictionary learning module: and performing feature extraction, dictionary learning and class histogram calculation by taking the known motor imagery electroencephalogram signals collected by the brain-computer interface system as learning data to obtain a learning dictionary and a class histogram of the learning data, and transmitting the learning dictionary and the class histogram of the learning data to the class identification module. And (3) identifying the category of the motor imagery electroencephalogram signal in a category identification module: the unknown motor imagery electroencephalogram signals are used as classification data to carry out feature extraction, a learning dictionary in a training module is used for obtaining sparse expression of the classification data, a class histogram of the classification data is calculated, and class identification and classification of the classification data are carried out according to the comparison result of the class histogram of the learning data and the class histogram of the classification data. The feature extraction comprises signal preprocessing, wavelet transformation and feature vector establishment.
Description
Technical Field
The invention belongs to the technical field of electroencephalogram signal processing, and particularly relates to an electroencephalogram signal classification system and method based on wavelet transformation and sparse expression.
Background
At present, the brain-computer interface technology is used as a brand-new communication and control technology, and can provide a language communication and environment control means for patients with normal thinking but severe dyskinesia. A brain-computer interface is defined as a device that enables a person to communicate or control with the outside world independently of the peripheral nervous system and muscles. The brain-computer interface technology is not only applied to providing language communication and environmental control for patients, but also has potential application value in the scientific fields of automatic control, military cognition and the like. In view of its great application prospect, the brain-computer interface has attracted the high attention of the international scientific community, which is called a research hotspot in the fields of brain science, rehabilitation engineering, biomedical engineering and man-machine control.
Among all signals capable of reflecting brain activities to be monitored, Electroencephalogram (EEG) signals have the advantages of being good in time resolution, simple in monitoring instruments, non-invasive and the like, and are adopted by most brain-computer interface systems.
In all brain-computer interface systems based on EEG signals, the brain activity signals that can be used as control signals are: visual evoked potential, P300 event related potential, motor imagery, cortical slow potential, etc. Aiming at the motor imagery signal, the theoretical basis of the brain-computer interface based on the motor imagery is as follows: when a sensory organ is stimulated by one or more events, the stimulation is transmitted to the brain, causing a specific electrical activity in the brain's neuron population, which is the time-dependent Potentials (ERPs). ERPs are a special evoked potential. It has been found through research that the electrical activity of ERPs is manifested as a change in energy at a particular frequency, primarily as an increase or decrease in energy of a particular rhythm signal, which is a common result of collective electrical activity of a large number of neurons. Thus, when ERPS occurs, the rhythmic activity energy of a specific frequency of the cortical region shows a decrease in amplitude, which is called Event-related Desynchronization (ERD), whereas a phenomenon of an increase in amplitude, called Event-related Desynchronization (ERS), occurs. The brain-computer interface system based on ERD/ERS mainly discriminates EEG signals for motor imagery mental tasks, such as imagining the movements of the left hand, right hand, feet, tongue, to generate different control commands. Pfunscheler et al found through research that such event-related synchronization or event-related desynchronization of moving areas can be observed even for patients with severe limb impairment while performing motor imagery.
Currently, the research of brain-computer interface system based on motor imagery is one of the research hotspots of the current brain-computer interface. The main reason is that the physiological basis and mathematical model of the ERS/ERD phenomenon has been studied and proven by many scientific research institutes and is currently the most common method for identifying sensory, motor and cognitive functions under normal and pathological conditions. In addition, the change of the mu rhythm and the beta wave in the phenomenon does not need to be induced by external stimulation, so that the training and the control of a testee are facilitated, and the phenomenon becomes the most popular realization way for the research of brain-computer interfaces. For an electroencephalogram multi-classification system, a Wopaw research group realizes cursor movement in a three-dimensional space in 2010 by utilizing u and beta waveband signal change characteristics when left and right hands and feet are imagined, but a tested object in the research needs long-term training, the result depends on a tested sample to a great extent, and only two motor imagery signals, namely the imagined left hand and the imagined right hand, are utilized.
In summary, the current brain-computer interface based on motor imagery has the following problems in the classification of electroencephalogram signals:
(1) the test dependence was too high: only a single digit is selected for a tested sample in the current experiment, and long-term training is needed;
(2) the classification types are few: the classification of electroencephalogram signals based on different motor imagery thought operations is very difficult, the most motor imagery thought operations which can be distinguished at present are six, and the application range of a brain-computer interface is limited by few classification types;
(3) the classification precision is not high: with the increase of the motor imagery category, the classification precision is reduced;
(4) the working and rest states cannot be distinguished at any time: if the brain-computer interface needs to continuously work if the user wears the brain-computer interface for a long time, the corresponding brain-computer interface system needs to be capable of distinguishing the motor imagery state and the rest state of the user, and the user cannot be in the motor imagery control state for a long time. The current brain-computer interface system does not have the function.
The prior art still lacks an effective solution to the above four problems.
Disclosure of Invention
In order to solve the problems, the invention solves the problems of few classification types, low classification precision and incapability of distinguishing working states and rest states at any time in the aspect of feature extraction and classification of a brain-computer interface based on motor imagery in the prior art, and provides an electroencephalogram signal classification system based on wavelet transformation and sparse expression.
In order to achieve the purpose, the invention adopts the following technical scheme:
an electroencephalogram signal classification system based on wavelet transformation and sparse expression comprises a dictionary learning module and a category identification module;
the dictionary learning module is configured to perform feature extraction, dictionary learning and class histogram calculation by taking known motor imagery electroencephalogram signals collected by the brain-computer interface system as learning data to obtain a learning dictionary and a class histogram of the learning data; the dictionary learning module transmits the obtained learning dictionary and the class histogram of the learning data to the class identification module;
the classification identification module is configured as a module which is used for extracting the characteristics of unknown motor imagery electroencephalogram signals collected by the brain-computer interface system as classification data, obtaining sparse expression of the classification data by utilizing a learning dictionary in the dictionary learning module, calculating a classification histogram of the classification data, and identifying and classifying the classification of the classification data according to the comparison result of the classification histogram of the learning data and the classification histogram of the classification data;
the feature extraction comprises signal preprocessing, wavelet transformation and feature vector establishment.
Further, the dictionary learning module comprises a first electroencephalogram signal storage module, and the first electroencephalogram signal storage module stores motor imagery electroencephalogram signals collected by the brain-computer interface system as learning data;
the category identification module comprises a second electroencephalogram signal storage module, and the second electroencephalogram signal storage module stores motor imagery electroencephalogram signals collected by the brain-computer interface system as category data.
In order to solve the problems, the invention solves the problems of few classification types, low classification precision and incapability of distinguishing working states and rest states at any time in the aspect of feature extraction and classification of a brain-computer interface based on motor imagery in the prior art, and provides an electroencephalogram signal classification method based on wavelet transformation and sparse expression.
In order to achieve the purpose, the invention adopts the following technical scheme:
a classification method of EEG signals based on wavelet transformation and sparse expression is based on an EEG signal classification system based on wavelet transformation and sparse expression, and comprises the following specific steps:
(1) dictionary learning of the motor imagery electroencephalogram signals in a dictionary learning module: taking known motor imagery electroencephalogram signals collected by a brain-computer interface system as learning data to perform feature extraction, dictionary learning and class histogram calculation to obtain a learning dictionary and a class histogram of the learning data, and transmitting the learning dictionary and the class histogram of the learning data to the class identification module;
(2) and (3) identifying the category of the motor imagery electroencephalogram signal in a category identification module: the method comprises the steps of performing feature extraction by taking unknown motor imagery electroencephalogram signals collected by a brain-computer interface system as classified data, obtaining sparse expression of the classified data by utilizing a learning dictionary in a training module, calculating a class histogram of the classified data, and performing class identification and classification on the classified data according to a comparison result of the class histogram of the learning data and the class histogram of the classified data.
Further, the feature extraction of the learning data or the classification data in the step (1) and the step (2) specifically includes signal preprocessing, wavelet transformation, and feature vector establishment.
Further, the signal preprocessing comprises the following specific steps:
and (3) frequency domain filtering: performing frequency domain filtering by taking a known motor imagery electroencephalogram signal acquired by a brain-computer interface system as learning data, wherein the frequency range of the filtered motor imagery electroencephalogram signal is 0-40 Hz;
and
baseline wander removal: and removing the baseline drift of the data after the frequency domain filtering, and removing the baseline drift by adopting a cubic spline difference method.
Further, the specific steps of the wavelet transform are as follows:
performing continuous wavelet transformation on the motor imagery electroencephalogram signal f (t) after signal preprocessing:
wherein, W f (alpha, tau) after continuous wavelet transformThe motor imagery electroencephalogram signal psi (t) is a wavelet function, alpha is a scale factor, and alpha>1, tau is a translation factor;
when α increases, it means to view the entire f (t) with stretched Ψ (t); conversely, when α decreases, we look at the local part of f (t) with compressed Ψ (t).
Further, the wavelet transform uses a Morlet wavelet function:
wherein f is c Is the center frequency, f b As a bandwidth parameter, σ f As variance, bandwidth parameter f b And variance σ f The relationship therebetween is expressed by equation (3):
further, determining wavelet transformation parameters of the motor imagery electroencephalogram signals of different frequency bands before wavelet transformation: center frequency f c And a bandwidth parameter f b ；
The motor imagery brain electrical signals are divided into delta frequency band, theta frequency band, alpha frequency band, beta frequency band and gamma frequency band;
the frequency of the delta frequency band is set to be less than or equal to 4 Hz;
the frequency of the theta band is set to be greater than 4Hz and less than 7 Hz;
the frequency of the alpha band is set to be more than 8Hz and less than 13 Hz;
the frequency of the beta band is set to be greater than 14Hz and less than 25 Hz;
the frequency of the gamma band is set to be greater than 26 Hz.
Further, the specific steps of establishing the feature vector are as follows:
performing wavelet analysis on the motor imagery electroencephalogram signals after wavelet transformation: decomposing the motor imagery electroencephalogram signal into wavelets of different levels according to a wavelet decomposition algorithm, wherein the number of layers of the wavelet decomposition depends on the useful components and sampling frequency of each node motor imagery electroencephalogram signal for extracting characteristics; the wavelet coefficient after the decomposition of the motor imagery electroencephalogram signal expresses the energy distribution of the wavelet coefficient in the time domain and the frequency domain;
selecting the frequency range of the ERD/ERS of the motor imagery electroencephalogram signal, selecting the decomposed motor imagery electroencephalogram signal for further wavelet analysis, and extracting three statistics of the mean value, the energy mean value and the mean square error of the wavelet coefficient as feature vectors.
Further, the dictionary learning in the step (1) obtains sparse expression of the eigenvector matrix Y through compressive sensing according to the eigenvector matrix Y after wavelet transformation, and the specific steps are as follows:
obtaining a learning dictionary phi epsilon IR satisfying the following formula by utilizing a K-SVD algorithm n×m (m > n) and sparse expressions
Wherein x is i Comprising k (k)<<n) non-zero elements or less:
wherein | · | purple sweet F Is Frobenius norm, | · | | purple 0 Is a 1 0 And (5) half norm, and calculating non-zero elements contained in the vector.
Further, the specific steps of obtaining sparse expression calculation of the classification data by using the learning dictionary in the training module in the step (2) are as follows:
combining the characteristic vector matrix Q obtained by characteristic extraction of the classified data in the step (2) with the learning dictionary phi obtained in the step (1),
according to the following steps:
obtaining sparse representation X of classified data Q 。
Further, the class histogram of the learning data in the step (1) and the class histogram of the classification data in the step (2) are respectively based on:
obtaining a class histogram of the learning data in the step (1) as h i The class histogram of the learning data in the step (2) is h Q ；
The specific steps of classifying the data in the step (2) are as follows:
according to the following steps:
determining the category of the classification data.
The invention has the beneficial effects that:
1. the classification precision is improved: according to the electroencephalogram signal classification system and method based on wavelet transformation and sparse expression, due to the fact that the difference of ERS/ERD appears in different frequency band ranges, characteristics of the motor imagery electroencephalogram signals are extracted through wavelet analysis, the difference is reflected to the maximum, and classification accuracy of the electroencephalogram signals is improved.
2. The classification category is improved: in the electroencephalogram signal classification system and method based on wavelet transformation and sparse expression, in the dictionary learning of the motor imagery electroencephalogram signals in the step (1), different thinking operations can establish a common learning dictionary phi, and in the class identification of the motor imagery electroencephalogram signals in the step (2), the class can be identified in the class identification process as long as the thinking operation appears in the dictionary learning part, so that the classification type of the algorithm is improved;
3. the motor imagery and the rest state can be detected: according to the electroencephalogram signal classification system and method based on wavelet transformation and sparse expression, a rest state is taken as a thinking operation, the state is taken into consideration when a dictionary phi is established, and in subsequent tests, when the rest state occurs, the rest state of a user can be detected;
4. the classification speed is fast: according to the electroencephalogram signal classification system and method based on wavelet transformation and sparse expression, a common dictionary is established by different thinking operations, and classification results can be obtained at one time by the different thinking operations in a test.
Drawings
FIG. 1 is a schematic diagram of the system of the present invention;
fig. 2 is a flow chart of the method of the present invention.
The specific implementation mode is as follows:
it should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
The embodiments and features of the embodiments in the present application may be combined with each other without conflict. The invention is further described with reference to the following figures and examples.
Example 1:
as introduced by the background technology, the brain-computer interface based on the motor imagery in the prior art has the problems of few classification types, low classification precision and incapability of distinguishing working states and rest states at any time in the aspects of feature extraction and classification, and provides an electroencephalogram signal classification system and method based on wavelet transformation and sparse expression.
In the implementation of the electroencephalogram signal classification system based on wavelet transformation and sparse representation,
in order to achieve the purpose, the invention adopts the following technical scheme:
a classification system of EEG signals based on wavelet transformation and sparse expression is shown in figure 1 and comprises a dictionary learning module and a category identification module;
the dictionary learning module is configured to take known motor imagery electroencephalogram signals collected by the brain-computer interface system as learning data to perform feature extraction, dictionary learning and class histogram calculation to obtain a learning dictionary and a class histogram of the learning data; the dictionary learning module transmits the obtained learning dictionary and the class histogram of the learning data to the class identification module;
the classification identification module is configured as a module which is used for extracting the characteristics of unknown motor imagery electroencephalogram signals collected by the brain-computer interface system as classification data, obtaining sparse expression of the classification data by utilizing a learning dictionary in the dictionary learning module, calculating a classification histogram of the classification data, and identifying and classifying the classification of the classification data according to the comparison result of the classification histogram of the learning data and the classification histogram of the classification data;
the feature extraction comprises signal preprocessing, wavelet transformation and feature vector establishment.
The dictionary learning module comprises a first electroencephalogram signal storage module, and the first electroencephalogram signal storage module stores motor imagery electroencephalogram signals collected by the brain-computer interface system as learning data;
the category identification module comprises a second electroencephalogram signal storage module, and the second electroencephalogram signal storage module stores the motor imagery electroencephalogram signals collected by the brain-computer interface system as classified data.
In order to achieve the purpose, the invention adopts the following technical scheme:
a electroencephalogram signal classification method based on wavelet transformation and sparse expression is based on an electroencephalogram signal classification system based on wavelet transformation and sparse expression, as shown in figure 2, and the method comprises the following specific steps:
(1) dictionary learning of the motor imagery electroencephalogram signals in a dictionary learning module: taking known motor imagery electroencephalogram signals collected by a brain-computer interface system as learning data to perform feature extraction, dictionary learning and class histogram calculation to obtain a learning dictionary and a class histogram of the learning data, and transmitting the learning dictionary and the class histogram of the learning data to the class identification module;
the specific steps of taking the motor imagery electroencephalogram signals collected by the brain-computer interface system in the step (1) as learning data are as follows:
the user wears the EEG electrode cap to perform motor imagery of different psychological operations, such as imagining left-hand movement and right-hand movement, and corresponding electroencephalogram signals are acquired to be X ═ X L :X R ]And the first electroencephalogram signal storage module stores the electroencephalogram signals. When the brain of a user is in a rest state, the brain-electrical signal acquisition device can be used as a psychological operation mode to acquire brain-electrical signals.
When the brain performs different motor imagery, the brain electrical signals at different electrode locations change over time. The changes caused by the same motor imagery task are similar, and the task of feature extraction is to find out what can describe the similarity aiming at a certain motor imagery task.
The step (1) of extracting the characteristics of the learning data specifically comprises signal preprocessing, wavelet transformation and characteristic vector establishment.
In the step (1), the signal preprocessing specifically comprises the following steps:
and (3) frequency domain filtering: and performing frequency domain filtering by taking the known motor imagery electroencephalogram signals collected by the brain-computer interface system as learning data, wherein the frequency range of the filtered motor imagery electroencephalogram signals is 0-40 Hz.
The beneficial effects of frequency filtering are: because the electroencephalogram signals related to the motor imagery are mainly concentrated below 40Hz, the frequency domain filtering is carried out, and the interference of other frequency band signals on the electroencephalogram signals related to the motor imagery is effectively prevented.
Baseline wander removal: and removing the baseline drift of the data after the frequency domain filtering, and removing the baseline drift by adopting a cubic spline difference method.
The beneficial effects of baseline wander removal are as follows: because EEG adopts lead electrodes, a non-invasive signal acquisition mode, muscle tissue and scalp surface impedance, some interferences such as involuntary moving head of a tested person and the like can often cause the EEG signal to drift along with time change, and baseline-removing drift effectively removes the phenomenon that the EEG signal drifts along with the time change, such as involuntary moving head and the like.
In the step (1), the wavelet transform specifically includes the steps of:
performing continuous wavelet transformation on the motor imagery electroencephalogram signal f (t) after signal preprocessing:
wherein, W f (alpha, tau) is a motor imagery electroencephalogram signal after continuous wavelet transformation, psi (t) is a wavelet function, alpha is a scale factor, and alpha is>1, tau is a translation factor;
when α increases, it means to view the entire f (t) with stretched Ψ (t); conversely, when α decreases, the part of f (t) is observed as compressed Ψ (t).
In this embodiment, wavelet transform is performed by using a Morlet wavelet function, which is defined as formula (2):
wherein, f c Is the center frequency, f b As a bandwidth parameter, σ f As variance, bandwidth parameter f b And variance σ f The relationship between them is expressed by equation (3):
the Morlet wavelet is very special in characteristic, and the waveform of the Morlet wavelet is Gaussian in time domain and frequency domain, so that the user can accurately extract the wanted signal frequency by using the Morlet wavelet by adjusting the parameters of the Morlet wavelet. Meanwhile, the wavelet has higher resolution in both time domain and frequency domain, which is also the reason for the wavelet transform using the Morlet wavelet in the invention.
Determining wavelet transformation parameters of the motor imagery electroencephalogram signals of different frequency bands before wavelet transformation: center frequency f c And a bandwidth parameter f b ；
The motor imagery electroencephalogram signal is divided into a delta frequency band, a theta frequency band, an alpha frequency band, a beta frequency band and a gamma frequency band;
the frequency of the delta frequency band is set to be less than or equal to 4 Hz;
the frequency of the theta frequency band is set to be greater than 4Hz and less than 7 Hz;
the frequency of the alpha band is set to be more than 8Hz and less than 13 Hz;
the frequency of the beta band is set to be greater than 14Hz and less than 25 Hz;
the frequency of the gamma band is set to be greater than 26 Hz.
The wavelet transformation parameter settings described above are specifically as follows in this embodiment:
because the difference of the electroencephalogram signals of the motor imagery is obvious in a frequency domain, before wavelet transformation is carried out, the electroencephalogram signals of different motor imagery are divided into five frequency bands: delta band (. ltoreq.4 Hz), theta band (4-7 Hz), alpha band (8-13 Hz), beta band (14-25 Hz), gamma band (>26 Hz).
Aiming at the five frequency bands, different wavelet transformation parameters are adopted for carrying out wavelet transformation, and a motor imagery electroencephalogram signal wavelet transformation parameter table is shown in table 1
TABLE 1
In the step (1), the specific steps of establishing the feature vector are as follows:
performing wavelet analysis on the motor imagery electroencephalogram signals after wavelet transformation: decomposing the motor imagery electroencephalogram signal into wavelets of different levels according to a wavelet decomposition algorithm, wherein the number of layers of the wavelet decomposition depends on the useful components and sampling frequency of each node motor imagery electroencephalogram signal for extracting characteristics; the wavelet coefficient after the decomposition of the motor imagery electroencephalogram signal expresses the energy distribution of the wavelet coefficient in the time domain and the frequency domain;
selecting the frequency range of the ERD/ERS of the motor imagery electroencephalogram signal, selecting the decomposed motor imagery electroencephalogram signal for further wavelet analysis, and extracting three statistics of the mean value, the energy mean value and the mean square error of the wavelet coefficient as the feature vector.
In the embodiment, the decomposition layer number of the wavelet determines the frequency range covered by each node motor imagery electroencephalogram signal for extracting the characteristics, and the frequency range covered by each characteristic value is reduced along with the increase of the decomposition layer number; when wavelet analysis is carried out on the motor imagery electroencephalogram signal, the number of decomposed layers depends on the useful components and sampling frequency of the specific motor imagery electroencephalogram signal.
If the sampling frequency of the motor imagery electroencephalogram signal is 128Hz, the motor imagery electroencephalogram signal is decomposed to the third layer, and the corresponding lowest frequency band is 0-8 Hz. According to the wavelet decomposition algorithm, the detail part of the decomposed motion image electroencephalogram signal is D1-D3, the approximate part is A3, and the frequency band range of each layer after wavelet decomposition is shown in Table 2.
TABLE 2
The wavelet coefficient after the decomposition of the electroencephalogram signal by motor imagery expresses the energy distribution of the electroencephalogram signal in the time domain and the frequency domain, and as can be seen from the table 2, D3 (8-16 Hz) is near the alpha (8-13 Hz) wave band range of the electroencephalogram signal, D2 (16-32 Hz) is near the beta (14-25 Hz) wave band of the electroencephalogram signal, and ERD/ERS of the electroencephalogram signal mainly appears in the two wave bands. Therefore, D2 and D3 are selected for further analysis, and in order to further reduce the dimension of the feature vector, the invention extracts three statistics of the mean, the energy mean and the mean square error of the wavelet coefficients as the feature vector.
The dictionary learning in the step (1) is to obtain sparse expression of the eigenvector matrix Y through compressive sensing according to the eigenvector matrix Y after wavelet transformation, and the specific steps are as follows:
obtaining a learning dictionary phi epsilon IR meeting the following formula by utilizing a K-SVD algorithm n×m (m > n) and sparse expressions
Wherein x is i Comprising k (k)<<n) non-zero elements or less:
wherein | · | purple sweet F Is Frobenius norm, | | · |. non-calculation 0 Is a 0 And half norm, calculating non-zero elements contained in the vector.
The K-SVD is a classical dictionary training algorithm, SVD decomposition is carried out on error terms according to the principle of minimum error, the decomposition terms which enable the minimum error are selected as updated dictionary atoms and corresponding atom coefficients, and an optimized solution is obtained through continuous iteration.
In this embodiment, a compressed sensing technology is applied, a learning dictionary with sparse expression is obtained after a feature vector matrix Y is compressed, and a subsequent electroencephalogram signal establishes its own sparse expression by using the learning dictionary, and uses: if the subsequent brain electrical signals belong to a certain thinking operation existing in the learning dictionary, the sparse expression coefficient of the thinking operation part should contain the minimum non-zero value, and a sparse expression histogram is established to classify the thinking operation. And (3) establishing a common learning dictionary by using different types of electroencephalogram signals, and once the learning dictionary is established, saving a large amount of time in the classification process of the subsequent step (2).
If the resting state of the user is thought to be a thinking operation in the dictionary learning process, the resting state of the user can be distinguished at any time in the subsequent classification process, so that the problem that most of the existing brain-computer interface systems cannot solve is solved, the brain-computer interface is used by the user for a long time, and the brain-computer interface system can identify the working state and the resting state of the user.
The learning data class histogram in the step (1) is [ h ═ h 1 ,h 2 ,...,h K ]Calculated according to the following formula:
in this embodiment, the training part of step (1) ends, and two results are obtained: 1) a learning dictionary phi; 2) class histogram h ═ h 1 ,h 2 ,...,h K ]。
(2) And (3) identifying the category of the motor imagery electroencephalogram signal in a category identification module: the method comprises the steps of performing feature extraction by taking unknown motor imagery electroencephalogram signals collected by a brain-computer interface system as classified data, obtaining sparse expression of the classified data by utilizing a learning dictionary in a training module, calculating a class histogram of the classified data, and performing class identification and classification on the classified data according to a comparison result of the class histogram of the learning data and the class histogram of the classified data.
The specific steps of using the motor imagery electroencephalogram signals collected by the brain-computer interface system in the step (2) as classification data are as follows:
the user wears the EEG electrode cap to perform motor imagery of different psychological operations, such as imagining left-hand movement and right-hand movement, and corresponding electroencephalogram signals are acquired to be X ═ X L :X R ]And storing the electroencephalogram signal in a second electroencephalogram signal storage module.
The step (2) of extracting the characteristics of the classified data specifically comprises signal preprocessing, wavelet transformation and characteristic vector establishment.
In this embodiment, the signal preprocessing, wavelet transform and feature vector establishment of the classified data in step (2) are completely consistent with the method formula adopted in step (1), and are not described herein again.
And obtaining the sparse expression of the classified data by utilizing a training dictionary in the training module to obtain a category histogram of the classified data.
The specific steps of sparse expression calculation of the classified data in the step (2) are as follows:
combining a characteristic vector matrix Q obtained by characteristic extraction of the classified data in the step (2) with the learning dictionary phi obtained in the step (1),
according to the following steps:
obtaining sparse representation X of classified data Q 。
The class histogram of the classified data in the step (2) is respectively according to:
obtaining the classification histogram of the classification data in the step (2) as h Q ；
And classifying the classified data according to the comparison result of the class histogram of the learning data and the class histogram of the classified data.
The step (2) of classifying the data specifically comprises the following steps:
according to the following steps:
determining the category of the classified data.
The invention has the beneficial effects that:
1. the classification precision is improved: according to the electroencephalogram signal classification system and method based on wavelet transformation and sparse expression, due to the fact that the difference of ERS/ERD appears in different frequency band ranges, characteristics of the motor imagery electroencephalogram signals are extracted through wavelet analysis, the difference is reflected to the maximum, and classification accuracy of the electroencephalogram signals is improved.
2. The classification category is improved: in the electroencephalogram signal classification system and method based on wavelet transformation and sparse expression, in the dictionary learning of the motor imagery electroencephalogram signals in the step (1), different thinking operations can establish a common learning dictionary phi, and in the class identification of the motor imagery electroencephalogram signals in the step (2), the class can be identified in the class identification process as long as the thinking operation appears in the dictionary learning part, so that the classification type of the algorithm is improved;
3. the motor imagery and the rest state can be detected: according to the electroencephalogram signal classification system and method based on wavelet transformation and sparse expression, a rest state is taken as a thinking operation, the state is taken into consideration when a dictionary phi is established, and in subsequent tests, when the rest state appears, the rest state of a user can be detected;
4. the classification speed is fast: according to the electroencephalogram signal classification system and method based on wavelet transformation and sparse expression, a common dictionary is established by different thinking operations, and classification results can be obtained at one time by the different thinking operations in a test.
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (3)
1. An electroencephalogram signal classification method based on wavelet transformation and sparse expression is characterized in that: the method comprises the following specific steps:
(1) dictionary learning of the motor imagery electroencephalogram signals in a dictionary learning module: the method comprises the steps that known motor imagery electroencephalogram signals collected by a brain-computer interface system are used as learning data to conduct feature extraction, dictionary learning and class histogram calculation, a common dictionary is established for different types of electroencephalogram signals in a dictionary learning module, the resting state of a user is considered to be a thinking operation, the learning dictionary and the class histogram of the learning data are obtained, and the learning dictionary and the class histogram of the learning data are transmitted to a class recognition module;
(2) and (3) identifying the category of the motor imagery electroencephalogram signal in a category identification module: performing feature extraction by taking unknown motor imagery electroencephalogram signals collected by a brain-computer interface system as classification data, obtaining sparse expression of the classification data by utilizing a learning dictionary in a training module, calculating a class histogram of the classification data, and performing class identification and classification on the classification data according to a comparison result of the class histogram of the learning data and the class histogram of the classification data;
the dictionary learning in the step (1) is to obtain sparse expression of the eigenvector matrix Y through compressive sensing according to the eigenvector matrix Y after wavelet transformation, and the specific steps are as follows:
obtaining a learning dictionary phi epsilon IR meeting the following formula by utilizing a K-SVD algorithm n×m M > n and sparse expressions
Wherein x is i Contains k or fewer non-zero elements:
wherein | · | purple sweet F Is Frobenius norm, | · | | purple 0 Is a 1 0 A half norm, which calculates non-zero elements contained in the vector; k is a radical of formula<<n；
The feature extraction comprises signal preprocessing, wavelet transformation and feature vector establishment;
the wavelet transformation firstly divides electroencephalogram signals of different motor imagings into five frequency bands: delta band, theta band, alpha band, beta band, gamma band;
wherein, the delta frequency band is less than or equal to 4Hz, the theta frequency band is 4-7 Hz, the alpha frequency band is 8-13 Hz, the beta frequency band is 14-25 Hz, and the gamma frequency band is more than 26 Hz;
aiming at the five frequency bands, performing wavelet transformation by adopting different wavelet transformation parameters;
the signal preprocessing comprises the following specific steps:
and (3) frequency domain filtering: performing frequency domain filtering by taking a known motor imagery electroencephalogram signal acquired by a brain-computer interface system as learning data, wherein the frequency range of the filtered motor imagery electroencephalogram signal is 0-40 Hz;
and
baseline wander removal: removing baseline drift of the data after frequency domain filtering, and removing the baseline drift by adopting a cubic spline difference method;
the specific steps of the wavelet transformation are as follows:
performing continuous wavelet transform on the motor imagery electroencephalogram signal f (t) after signal preprocessing:
wherein, W f (alpha, tau) is a motor imagery electroencephalogram signal after continuous wavelet transformation, psi (t) is a wavelet function, alpha is a scale factor, and alpha is>1, tau is a translation factor;
when α increases, it means to observe the entire f (t) with the stretched Ψ (t); conversely, when α decreases, then observe the part of f (t) with compressed Ψ (t);
determining wavelet transformation parameters of the motor imagery electroencephalogram signals of different frequency bands before wavelet transformation: center frequency f c And a bandwidth parameter f b ；
The motor imagery electroencephalogram signal is divided into a delta frequency band, a theta frequency band, an alpha frequency band, a beta frequency band and a gamma frequency band;
the frequency of the delta band is set to be not more than 4 Hz;
the frequency of the theta frequency band is set to be greater than 4Hz and less than 7 Hz;
the frequency of the alpha band is set to be more than 8Hz and less than 13 Hz;
the frequency of the beta band is set to be greater than 14Hz and less than 25 Hz;
the frequency of the gamma band is set to be greater than 26 Hz;
the specific steps of establishing the feature vector are as follows:
performing wavelet analysis on the motor imagery electroencephalogram signals after wavelet transformation: decomposing the motor imagery electroencephalogram signal into wavelets of different levels according to a wavelet decomposition algorithm, wherein the number of layers of wavelet decomposition depends on the useful components and sampling frequency of each node motor imagery electroencephalogram signal for extracting features; the wavelet coefficient after the decomposition of the motor imagery electroencephalogram expresses the energy distribution of the wavelet coefficient in the time domain and the frequency domain;
selecting a frequency range of the ERD/ERS of the motor imagery electroencephalogram signal, selecting the decomposed motor imagery electroencephalogram signal for further wavelet analysis, and extracting three statistics of a mean value, an energy mean value and a mean square error of a wavelet coefficient as a feature vector;
the specific steps of obtaining sparse expression calculation of the classification data by using the learning dictionary in the training module in the step (2) are as follows:
combining a feature vector matrix Q obtained by feature extraction of the classified data in the step (2) with the learning dictionary phi obtained in the step (1),
according to the following steps:
obtaining sparse representation X of classified data Q ；
The class histogram of the learning data in the step (1) and the class histogram of the classification data in the step (2) are respectively based on:
obtaining a class histogram of the learning data in the step (1) as h i And the class histogram of the learning data in the step (2) is h Q ；
The step (2) of classifying the data specifically comprises the following steps:
according to the following steps:
determining the category of the classification data.
2. The electroencephalogram signal classification method based on wavelet transformation and sparse representation as claimed in claim 1, wherein the applied electroencephalogram signal classification system based on wavelet transformation and sparse representation comprises a dictionary learning module and a category identification module; the method is characterized in that:
the dictionary learning module is configured to take known motor imagery electroencephalogram signals collected by the brain-computer interface system as learning data to perform feature extraction, dictionary learning and class histogram calculation to obtain a learning dictionary and a class histogram of the learning data; the dictionary learning module transmits the obtained learning dictionary and the class histogram of the learning data to the class identification module; the dictionary learning module establishes a common dictionary for different types of electroencephalogram signals, and considers the rest state of a user as a thinking operation;
the classification identification module is configured as a module which is used for extracting the characteristics of unknown motor imagery electroencephalogram signals collected by the brain-computer interface system as classification data, obtaining sparse expression of the classification data by utilizing a learning dictionary in the dictionary learning module, calculating a classification histogram of the classification data, and identifying and classifying the classification of the classification data according to the comparison result of the classification histogram of the learning data and the classification histogram of the classification data;
the dictionary learning module is used for obtaining sparse expression of the characteristic vector matrix Y through compressive sensing according to the characteristic vector matrix Y after wavelet transformation in the dictionary learning process, and the specific steps are as follows:
obtaining a learning dictionary phi epsilon IR meeting the following formula by utilizing a K-SVD algorithm n×m M > n and sparse expressions
Wherein x is i Contains k or fewer non-zero elements:
wherein | · | purple sweet F Is Frobenius norm, | · | | purple 0 Is a 0 A half norm, which is used for calculating non-zero elements contained in the vector; k is a radical of formula<<n；
The feature extraction comprises signal preprocessing, wavelet transformation and feature vector establishment;
the wavelet transformation firstly divides electroencephalogram signals of different motor imagings into five frequency bands: delta band, theta band, alpha band, beta band, gamma band;
wherein, the delta frequency band is less than or equal to 4Hz, the theta frequency band is 4-7 Hz, the alpha frequency band is 8-13 Hz, the beta frequency band is 14-25 Hz, and the gamma frequency band is more than 26 Hz;
for these five bands, wavelet transform is performed using different parameters of wavelet transform.
3. The electroencephalogram signal classification method based on wavelet transformation and sparse representation as claimed in claim 2, which is characterized in that: the dictionary learning module comprises a first electroencephalogram signal storage module, and the first electroencephalogram signal storage module stores motor imagery electroencephalogram signals collected by the brain-computer interface system as learning data;
the category identification module comprises a second electroencephalogram signal storage module, and the second electroencephalogram signal storage module stores motor imagery electroencephalogram signals collected by the brain-computer interface system as category data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710244077.7A CN107423668B (en) | 2017-04-14 | 2017-04-14 | Electroencephalogram signal classification system and method based on wavelet transformation and sparse expression |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710244077.7A CN107423668B (en) | 2017-04-14 | 2017-04-14 | Electroencephalogram signal classification system and method based on wavelet transformation and sparse expression |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107423668A CN107423668A (en) | 2017-12-01 |
| CN107423668B true CN107423668B (en) | 2022-09-27 |
Family
ID=60424128
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710244077.7A Active CN107423668B (en) | 2017-04-14 | 2017-04-14 | Electroencephalogram signal classification system and method based on wavelet transformation and sparse expression |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107423668B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109044351A (en) * | 2018-08-03 | 2018-12-21 | 周文芳 | Logical thinking training system and method based on EEG feedback |
| CN110674738A (en) * | 2019-09-23 | 2020-01-10 | 长沙理工大学 | Method for identifying left and right hand electroencephalogram signals and related device |
| CN114259240A (en) * | 2021-09-28 | 2022-04-01 | 大连大学 | Electroencephalogram signal dictionary learning method based on improved experience wavelet transformation |
Family Cites Families (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8396310B1 (en) * | 2009-09-30 | 2013-03-12 | Rockwell Collins, Inc. | Basis learning for sparse image representation and classification and low data rate compression |
| US8364254B2 (en) * | 2009-01-28 | 2013-01-29 | Brainscope Company, Inc. | Method and device for probabilistic objective assessment of brain function |
| GB0906029D0 (en) * | 2009-04-07 | 2009-05-20 | Nat Univ Ireland Cork | A method of analysing an electroencephalogram (EEG) signal |
| CN101558997A (en) * | 2009-05-27 | 2009-10-21 | 江西蓝天学院 | Electroencephalograph signal identification method based on second-order blind identification |
| CN102024152B (en) * | 2010-12-14 | 2013-01-30 | 浙江大学 | Method for recognizing traffic sings based on sparse expression and dictionary study |
| CN102133100A (en) * | 2011-03-04 | 2011-07-27 | 上海交通大学 | Sparse representation-based electroencephalogram signal detection method |
| KR101380964B1 (en) * | 2011-05-09 | 2014-04-04 | 광주과학기술원 | Apparatus of brain-computer interface, and classification method thereof |
| CN102488516A (en) * | 2011-12-13 | 2012-06-13 | 湖州康普医疗器械科技有限公司 | Nonlinear electroencephalogram signal analysis method and device |
| CN103425249A (en) * | 2013-09-06 | 2013-12-04 | 西安电子科技大学 | Electroencephalogram signal classifying and recognizing method based on regularized CSP and regularized SRC and electroencephalogram signal remote control system |
| CN107438398A (en) * | 2015-01-06 | 2017-12-05 | 大卫·伯顿 | Portable wearable monitoring system |
| CN105468138A (en) * | 2015-07-15 | 2016-04-06 | 武汉理工大学 | Intelligent vehicle obstacle avoidance and navigation method based on brain-computer interface technology and lidar |
| CN105022486B (en) * | 2015-07-17 | 2018-01-19 | 西安交通大学 | EEG signals discrimination method based on the driving of different expressions |
| CN105574534B (en) * | 2015-12-17 | 2019-03-26 | 西安电子科技大学 | Conspicuousness object detection method based on sparse subspace clustering and low-rank representation |
| CN105595961A (en) * | 2015-12-21 | 2016-05-25 | 天津大学 | Alzheimer's disease detecting system and method based on electroencephalogram signals |
| CN105740772A (en) * | 2016-01-22 | 2016-07-06 | 江苏师范大学 | Real time monitoring method of cortical somatosensory evoked potential of transform joint sparse model |
| CN105809124B (en) * | 2016-03-06 | 2018-06-01 | 北京工业大学 | The feature extracting method of Mental imagery EEG signals based on DWT and Parametric t-SNE |
| CN105956624B (en) * | 2016-05-06 | 2019-05-21 | 东南大学 | Mental imagery brain electricity classification method based on empty time-frequency optimization feature rarefaction representation |
| CN106446762A (en) * | 2016-07-07 | 2017-02-22 | 哈尔滨工程大学 | Motion imagination EEG classification processing method based on sparse representation classification algorithm |
| CN106388814B (en) * | 2016-10-11 | 2019-06-18 | 天津大学 | Epileptic EEG Signal recognition methods based on optimal core time-frequency distributions Visual Graph |
-
2017
- 2017-04-14 CN CN201710244077.7A patent/CN107423668B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN107423668A (en) | 2017-12-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Removal of muscle artifacts from the EEG: A review and recommendations | |
| Kachenoura et al. | ICA: a potential tool for BCI systems | |
| Phinyomark et al. | Application of wavelet analysis in EMG feature extraction for pattern classification | |
| CN110269609B (en) | Method for separating ocular artifacts from electroencephalogram signals based on single channel | |
| Somers et al. | Removal of eye blink artifacts in wireless EEG sensor networks using reduced-bandwidth canonical correlation analysis | |
| CN114533086B (en) | Motor imagery brain electrolysis code method based on airspace characteristic time-frequency transformation | |
| CN106991409B (en) | Motor imagery electroencephalogram signal feature extraction and classification system and method | |
| CN107423668B (en) | Electroencephalogram signal classification system and method based on wavelet transformation and sparse expression | |
| Tanaka et al. | SSVEP frequency detection methods considering background EEG | |
| Ramos-Aguilar et al. | Analysis of EEG signal processing techniques based on spectrograms | |
| Liu et al. | EEG classification for multiclass motor imagery BCI | |
| Gu et al. | AOAR: an automatic ocular artifact removal approach for multi-channel electroencephalogram data based on non-negative matrix factorization and empirical mode decomposition | |
| CN111820876A (en) | Dynamic construction method of electroencephalogram spatial filter | |
| Geng et al. | A fusion algorithm for EEG signal processing based on motor imagery brain-computer interface | |
| Rajashekhar et al. | Electroencephalogram (EEG) signal classification for brain–computer interface using discrete wavelet transform (DWT) | |
| Maleki et al. | Application of PLSR with a comparison of MATLAB classification learner app in using BCI | |
| Liu et al. | Identification of anisomerous motor imagery EEG signals based on complex algorithms | |
| Zhao et al. | Incremental common spatial pattern algorithm for BCI | |
| Xu et al. | Feature extraction and classification of EEG for imaging left-right hands movement | |
| Zhu et al. | Evaluating functional connectivity in alcoholics based on maximal weight matching | |
| Christoforou et al. | The bilinear brain: towards subject-invariant analysis | |
| Hsu | Wavelet-coherence features for motor imagery EEG analysis posterior to EOG noise elimination | |
| CN116150646A (en) | Motor imagery electroencephalogram signal classification algorithm based on non-uniform frequency band MFMDRM | |
| Ghane et al. | Robust understanding of EEG patterns in silent speech | |
| Alomari et al. | Eeg-based classification of imagined fists movements using machine learning and wavelet transform analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |