CN114861738B - Electroencephalogram tracing and dipole selection-based motor imagery classification method - Google Patents

Electroencephalogram tracing and dipole selection-based motor imagery classification method Download PDF

Info

Publication number
CN114861738B
CN114861738B CN202210783534.0A CN202210783534A CN114861738B CN 114861738 B CN114861738 B CN 114861738B CN 202210783534 A CN202210783534 A CN 202210783534A CN 114861738 B CN114861738 B CN 114861738B
Authority
CN
China
Prior art keywords
dipole
electroencephalogram
motor imagery
signal
tracing
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210783534.0A
Other languages
Chinese (zh)
Other versions
CN114861738A (en
Inventor
陈昆
魏欣
马力
刘泉
艾青松
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University of Technology WUT
Original Assignee
Wuhan University of Technology WUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Wuhan University of Technology WUT filed Critical Wuhan University of Technology WUT
Priority to CN202210783534.0A priority Critical patent/CN114861738B/en
Publication of CN114861738A publication Critical patent/CN114861738A/en
Application granted granted Critical
Publication of CN114861738B publication Critical patent/CN114861738B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/369—Electroencephalography [EEG]
    • A61B5/372—Analysis of electroencephalograms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Pathology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Psychiatry (AREA)
  • Psychology (AREA)
  • Biophysics (AREA)
  • Signal Processing (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention provides a motor imagery classification method based on electroencephalogram tracing and dipole selection, which comprises the following steps: acquiring electroencephalogram signals based on multi-class motor imagery; tracing the source of the multi-channel electroencephalogram signal to obtain a signal of a cortical neural activity source; carrying out dipole channel selection on a source space dipole, taking the energy of each dipole electroencephalogram signal as a search strategy for selecting and deleting a dipole channel set, taking an improved F-score value of each motor imagery category and the energy of the rest categories of electroencephalograms as an optimal dipole channel selection evaluation criterion, and extracting electroencephalogram data of the source space selection dipole; inputting the electroencephalogram data into a common spatial mode filter for feature extraction; and inputting the common spatial mode features into a support vector machine classifier to realize classification of the motor imagery electroencephalogram signals. According to the invention, on the basis of exploring the motor imagery electroencephalogram rule, the research on motor imagery electroencephalogram signal processing, feature extraction and classification methods is developed, and the classification accuracy is effectively improved.

Description

Electroencephalogram tracing and dipole selection-based motor imagery classification method
Technical Field
The invention relates to the technical field of electroencephalogram signal processing, in particular to a motor imagery classification method based on electroencephalogram tracing and dipole selection, which is used for improving the accuracy of motor imagery electroencephalogram signal task decoding.
Background
The brain-computer interface is a direct connection established between the brain of a human or animal and external equipment, and realizes information exchange between the brain and the equipment. The intention of a user is judged by analyzing the electroencephalogram signals and detecting and identifying the activation effects of different brain areas, the information processing process of the human brain is known, and then the communication and the control between the brain and the external equipment are realized.
The brain-computer interface based on motor imagery is an important component in a plurality of brain-computer interface paradigms, which need to collect electroencephalogram signals of a subject when the subject executes a specific motor imagery task, identify motor imagery contents according to the electroencephalogram signals, and then convert an identification result into a command to control external equipment. However, the electroencephalogram signal is low in spatial resolution due to the influence of brain volume conduction effect, so that the key to increasing the spatial resolution of the electroencephalogram signal and extracting effective distinguishing features is the success of a motor imagery brain-computer interface.
At present, common methods for improving the spatial resolution of electroencephalogram signals are mainly divided into three categories. The first category is a method introducing electroencephalogram imaging technology, mainly comprising fMRI technology, but the brain-computer interface based on fMRI is not suitable for daily use due to large time delay. The second type is invasive electroencephalogram signal acquisition, but the method easily causes immunoreaction and callus of a patient, and further causes the decline and disappearance of signal quality. The third category is blind source separation (BS) methods, which separate the mixed signal and decompose the signal into linear combinations of sources of different signals, under the condition that neither the source signal nor the transmission system characteristics are known. And EEG Source Imaging (ESI) used for the focal range of brain diseases such as epilepsy can provide noninvasive imaging of brain nerve activity, and EEG activity signals with high time resolution and high spatial resolution are estimated by solving the problem of EEG inverse.
In the prior art, the accuracy of multiple classes of motor imagery is to be improved urgently, and the analysis process of the electroencephalogram signals of the motor imagery comprises the following steps: the method comprises the steps of signal acquisition, preprocessing, feature extraction and selection and pattern recognition, wherein the electroencephalogram signals detected by scalp electrodes are weak in amplitude and easy to be interfered by noise, the EEG is poor in spatial resolution, and the difference between subjects is obvious, so that the research of motor imagery multi-class classification needs to be further developed.
Disclosure of Invention
The invention aims to provide a motion imagery classification method based on electroencephalogram tracing and dipole selection, aiming at the technical problem that the recognition method in the prior art is low in accuracy of multiple classifications of motion imagery, and the accuracy of multiple classifications of electroencephalogram signals can be improved while characteristic dimensions are effectively reduced, redundant channels are eliminated, and the calculated amount is reduced.
In order to achieve the purpose, the invention designs a motor imagery classification method based on electroencephalogram tracing and dipole selection, which is characterized by comprising the following steps:
s1: acquiring electroencephalogram signals based on multi-class motor imagery, and preprocessing the electroencephalogram signals through a band-pass filter and a common average reference;
s2: calculating a head model, a source model, a SourceModel and a lead field;
s3: taking the standardized low-resolution brain magnetoelectric tomography sLORETA as a constraint condition for calculating a source space electroencephalogram signal, and tracing the acquired electroencephalogram signal and a lead field to obtain the source space electroencephalogram signal;
s4: calculating the energy of each category of motor imagery electroencephalogram signals in each dipole of a source space, obtaining a screened source space dipole subset by adopting an improved F-score method and setting a threshold value as a strategy for selecting source space dipoles, and extracting electroencephalogram data of the source space dipole subset;
s5: extracting signal characteristics of the extracted electroencephalogram data by adopting a common space mode CSP;
s6: and inputting the CSP features into a support vector machine for classification to obtain a classification result.
Preferably, the source model Sourcemodel in step S2 is a source model obtained by calculation according to the estimated number, position and direction of dipole sources, the number of dipoles is set to 15002 in step S2, and the dipoles are uniformly distributed on the cortex, so that the accuracy of electroencephalogram tracing is ensured under the condition that the calculation amount is not large.
Preferably, when the head model Headmodel is calculated in step S2, the scalp-cerebrospinal fluid-skull boundary element model BEM is used, and because the electrical conductivity of each structure in the brain is different, the scalp is set in step S2: cerebrospinal fluid: the conductivity of the skull is 1.
Preferably, the specific steps of step S4 include:
s4.1: all data for source space electroencephalography signals after tracingS(t)∈4N×C×TTo find the leftAverage signal of various signals of hand, right hand, tongue and feet
Figure 312599DEST_PATH_IMAGE001
，
Figure 869482DEST_PATH_IMAGE002
In whichＮRepresenting the experimental times of the brain electrical signals of each motor imagery category,Cthe number of dipoles in the source space is indicated,Trepresenting the sample points in the time domain,ithe element belongs to {1,2,3 and 4}, and represents motor imagery electroencephalogram signals of a left hand, a right hand, a tongue and two feet respectively;
s4.2: calculating the energy of each kind of average signal in the source space in each dipoleP i (c)∈C×1，
Figure 870936DEST_PATH_IMAGE003
Whereini∈{1,2,3,4}；
S4.3: obtaining improved F-score value of each kind of signal energy and other kinds of signal energy in each dipoleF ir (c) And respectively setting threshold values, selectingF ir (c) Dipoles larger than the threshold value to obtain the selected dipole subsetLExtracting source space in dipole subsetLThe brain electrical signal ofV(t)。
Preferably, step S4.3 specifically comprises:
s4.3.1: calculating improved F-score values in all dipoles of source space of each category of motor imagery electroencephalogram signal energy and other categories of signal energyF ir (c)；
S4.3.2: the F-score values of each classF ir (c) Of maximum value ofx% is set as threshold, dipoles with improved F-score values larger than the threshold are selected, and the dipoles are added into the selected dipole set;
s4.3.3: using the common set of each category selected dipole set as the final selected dipole setLIf, ifLIn which the selected dipoles are too large orToo small, return to step S4.3.2 for threshold value thresholdx% adjustment to extract source spatial data dipolesLCorresponding data as the data after channel selectionV(t)，V(t)∈4N×L×T. When the number of training samples input into the classifier is equal to the number of sample features, the obtained classification effect is the best. Therefore, the threshold value is set so that the number of dipoles in the selected dipole set is equal to the number of dipoles in the selected dipole set, and the CSP feature quantity extracted in the step S5 is ensured not to be too much or too little so as to cause under-fitting or over-fitting of the classification model training, so that the best classification effect is achieved.
Preferably, the modified F-score valueF ir (c) In the calculation formula, an OVR (One-overturs-Rest) strategy is adopted to popularize the binary problem into the multi-classification problem, a certain class sample is classified into One class, and all other samples are classified into One class, so that the discrimination capability of the dipole on a plurality of class signals is calculated. The complexity of the OVR strategy is small, and the calculation speed is high.F ir (c) In the calculation formula, the numerator is the square of the difference value of different signal energies of the same dipole, the denominator is the sum of the different signal energies of the same dipole, the numerator calculates the discrimination capacity, and the denominator pair performs data normalization operation:
Figure 928891DEST_PATH_IMAGE004
wherein, the first and the second end of the pipe are connected with each other,X r =1/3P rest ，X i =P i ，P rest means for removingiThe sum of the source spatial mean signal energies of the other three classes outside the class.
Preferably, in step S5, the source space signal of the extracted electroencephalogram data is subjected to logarithm operation first, and then CSP signal feature extraction is performed, and the logarithm operation is performed to make the data conform to normal distribution, so that the absolute value of the data is reduced, and the variables of the scale are compressed, so that the data is more stable.
Preferably, the CSP signal feature extraction in step S5 adopts OTO-CSP method to extract features of multi-class tasks. The OTO-CSP method is used to enable the distribution distance of a plurality of different types of features to be larger, and the improvement of the classification precision is facilitated.
Preferably, each class F-score valueF ir (c) Of maximum value ofx% set range of [0.15,0.35]. The different thresholds set by each subject are used for adapting to the difference of electroencephalogram signals of different subjects so as to ensure the classification effect.
The invention also provides computer equipment which comprises a memory, a processor and program instructions stored in the memory and run by the processor, wherein the processor executes the program instructions to realize the motor imagery classification method based on electroencephalogram tracing and dipole selection.
The invention provides a motor imagery classification method based on electroencephalogram tracing and dipole selection, which comprises the steps of firstly preprocessing electroencephalogram signals generated based on motor imagery through a band-pass filter and a common average reference; then tracing the multi-channel motor imagery electroencephalogram signals by adopting an sLORETA method, obtaining electroencephalogram signals of all dipole channels in a source space, wherein the overlarge number of dipoles in the source space can cause overlarge calculated amount and reduction of classification precision, the optimal leads of each subject are different due to individual difference of the subjects, calculating the energy of each category of electroencephalogram signals in each dipole channel in the source space, taking the energy of each dipole electroencephalogram signal as a search strategy for selecting and deleting a dipole channel set, selecting the dipole channels of the whole source space dipole, taking an improved F-score value of each motor imagery category and the energy of the rest category of electroencephalogram signals as an evaluation criterion for selecting and deleting the optimal dipole channels, and effectively selecting the optimal dipole channel subset by utilizing an improved F-score method and setting different thresholds for each subject. And extracting the electroencephalogram signal of the optimal dipole channel, performing log operation, extracting the CSP (chip size Package) characteristics of the electroencephalogram signal subjected to the log operation, and inputting the characteristics into an SVM (support vector machine) classifier to obtain a classification result. The dipole channel selection method for the source space can effectively reduce the characteristic dimension, eliminate redundant channels, reduce the calculated amount and improve the multi-classification accuracy of the electroencephalogram signals.
The invention has the following beneficial effects:
1) F-score is an effective feature selection method, which can measure the distinguishing capability of a feature between two categories, and the larger the F-score value of a feature is, the larger the distinguishing capability of the feature is. The method uses the characteristic selection method for channel selection, judges the distinguishing capability of each dipole on different types of signals in a source space, selects the dipole with higher distinguishing capability, reduces data redundancy and improves the classification accuracy.
2) When the F-score is used as a feature selection method, the F-score of the feature average value of the positive sample and the negative sample is calculated to obtain the distinguishing capability of each feature. When the channel selection is carried out, the F-score value of each category of average signal energy of all dipoles in the source space is calculated. This is because different motor imagery tasks activate different areas of the cerebral cortex, the energy of the brain areas activated is increased, and therefore the energy F-score value is chosen as a criterion for selecting dipoles to have greater discriminatory power.
3) The invention improves the original F-score calculation formula, the numerator is the square of the difference value of different signal energies of the same dipole, the denominator is the sum of the different signal energies of the same dipole, the numerator calculates the discrimination capacity, and the denominator pair carries out data normalization operation. Compared with the original F-score method, the improved method reduces the calculation complexity and improves the calculation speed.
4) The invention adopts OVR (One-overturs-Rest) strategy to popularize the F-score method two-classification problem to the multi-classification problem, and classifies a certain class of samples into One class and all other samples into One class, thereby calculating the discrimination capability of the dipole on a plurality of class signals. Compared with an OTO (One-To-One) strategy, the OVR strategy has the advantages of smaller complexity and higher calculation speed.
Drawings
Fig. 1 is a flowchart of a motor imagery classification method based on electroencephalogram tracing and dipole selection in an embodiment of the present invention.
FIG. 2 is a result diagram of tracing different electroencephalograms.
Fig. 3 is a detailed flow chart of source space dipole channel selection.
Fig. 4 is a CSP feature map extracted after extracting the source spatial signal.
Fig. 5 is a diagram of the confusion matrix resulting from the classification.
FIG. 6 is a graph showing the classification accuracy of each subject and the average classification accuracy.
Detailed Description
In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is given with reference to the accompanying drawings and specific examples, but the following examples are only illustrative, and the scope of the present invention is not limited by these examples.
The embodiment of the invention provides a motor imagery signal classification method based on electroencephalogram tracing and dipole selection, which is characterized in that source space dipole channels are selected by adopting an improved F-score method after multi-channel data are traced to complete the identification of four categories of motor imagery electroencephalogram signals including a left hand, a right hand, a tongue and two feet, and the technical effect of improving the classification accuracy of the motor imagery electroencephalogram signals can be realized.
In order to achieve the technical effects, the general idea of the invention is as follows:
firstly, electroencephalogram signals based on multi-class motor imagery are collected, preprocessing is carried out through a band-pass filter and a common average reference, then the preprocessed multi-channel electroencephalogram signals are traced to obtain source data in a source space, the energy of each dipole electroencephalogram signal in the source space is selected as a feature, then an improved F-score method is used as a search strategy, different thresholds are set for different subjects as criteria for selecting and deleting dipole sets, CSP feature extraction is carried out after log operation is carried out on extracted dipole data, the extracted features are input into a support vector machine to be classified to obtain classification results, and finally motor imagery electroencephalogram signal classification is achieved according to the classification results.
As shown in fig. 1, the motor imagery classification method based on electroencephalogram tracing and dipole selection provided by the invention comprises the following steps:
s1, electroencephalogram signals based on multi-class motor imagery are collected and preprocessed through a band-pass filter and a common average reference to obtain multi-channel electroencephalogram signals. The electroencephalogram signals are subjected to 8-30Hz band-pass filtering and 50Hz power frequency removal processing, clutter and interference of other frequency bands can be eliminated, common average reference is adopted, the signal-to-noise ratio is improved, and the multichannel electroencephalogram signals are obtained.
S2, a head model, a source model, a SourceModel and a lead field are calculated.
Specifically, a head model Headcodel is computed from the ICBM152, sourceModel is set based on the estimated number, position, and orientation of the sources, and a lead field Leadfield is computed from the 3D electrode positions, headcodel, and SourceModel.
S3, the standardized low-resolution brain magnetoelectric tomography sLORETA is used as a constraint condition for calculating the source space electroencephalogram signal, and the source space electroencephalogram signal is obtained by tracing the acquired electroencephalogram signal and the lead field.
Specifically, the tracing is to estimate the position, direction and intensity information of the neural activity source in the brain by inversion according to the potential signal measured by the head table. Because the number of scalp electrodes is far less than the number of dipoles in source space, a unique solution cannot be obtained, the L2 norm is mathematically constrained by adopting the sLORETA, and the source space electroencephalogram signal under constraint is obtained. The traced signals are shown in fig. 2, which shows that different brain areas are activated by different brain electrical signals.
S4, calculating the energy of each category of motor imagery electroencephalogram signals in each dipole of the source space, adopting an improved F-score method and setting a threshold value as a strategy for selecting the source space dipoles, obtaining a screened source space dipole subset, and extracting electroencephalogram data of the source space dipole subset.
The F-score method can measure the discrimination capability of the sample characteristics between two types of samples, and the calculation formula of the F-score characteristic selection is as follows:
Figure 332190DEST_PATH_IMAGE005
。
the traditional F-score feature selection method is improved, the categories are expanded from two categories to a plurality of categories, and the method is used for channel selection. Calculating the energy of the average signal of each EEG category in each dipole channel of the source space, calculating an improved F-score value of the energy of each category signal of each dipole channel, setting different thresholds for each subject, selecting the dipole channel with the improved F-score value larger than the threshold, generating a dipole subset, and extracting the EEG signals of the dipole subset in the source space. The specific flow is shown in fig. 3.
Since different motor imagery actions in the source space activate different brain regions, the energy of each dipole is chosen as the basis for calculating the F-score value to discriminate between the classes separability.S(t) Is a source spatial signal of 4N×C×TTo be finally obtainedCIs a subset ofLThe method and the device prevent overfitting while reducing the calculation amount, and improve the classification accuracy of the multi-class-ratio signals.
The specific steps of step S4 include:
s4.1: for all data of source space after tracingS(t)∈4N×C×TCalculating the average signal of each signal type
Figure 794396DEST_PATH_IMAGE006
，
Figure 407780DEST_PATH_IMAGE007
In whichNRepresenting the experimental times of the brain electrical signals of each motor imagery category,Cthe number of dipoles in the source space is indicated,Trepresenting the sample points in the time domain,iand E {1,2,3,4}, which respectively represent the motor imagery signals of the left hand, the right hand, the tongue and the feet.
S4.2: calculating the energy of each kind of average signal in the source space in each dipoleP i (c)∈C×1，
Figure 879212DEST_PATH_IMAGE008
In whichi∈{1,2,3,4}。
S4.3: evaluating various classes of informationModified F-score values of number and other class signals in each dipoleF ir (c) And setting a threshold value, selectingF ir (c) Dipoles larger than the threshold value to obtain the selected dipole subsetLExtracting source space in dipole subsetLThe brain electrical signal ofV(t)。
For classification, the larger the inter-class distance is, and the smaller the intra-class distance is, the stronger the discrimination ability of classification is. For channel selection, the conventional feature selection method F-score is modified and used for channel selection to maximize the spacing between different classes of signals in the same dipole.
In one embodiment, step S4.3 specifically includes:
s4.3.1: the F-score formula is improved, an OVR (One-overturs-Rest) strategy is adopted to expand the two-classification problem into a multi-classification problem, and the improved F-score formula is
Figure 402597DEST_PATH_IMAGE009
WhereinX r =1/3P rest ，X i =P i ，P rest Means for removingiAnd calculating the improved F-score value of all dipoles of the source space of the motor imagery electroencephalogram signals and the signals of other categories.
S4.3.2: taking the maximum value of the F-score values of each classx% is set as threshold, and F-score values of various categories are selectedF ir (c) Dipole above thresholdcAdded to a subset of the class-chosen dipoles. Since the distribution of EEG varies greatly among subjects due to the differences in cognitive level, neural activity and fatigue state among subjects, the initial threshold value of each subject is set according to the maximum value of the improved F-score values of the signals of the respective classesx%，xHas a value range of [0.15,0.35 ]]。
S4.3.3: using the common set of each category selected dipole set as the final selected dipole setClosing boxLIf, ifLIf the number of dipoles selected in the dipole pair is too large or too small, the step S4.3.2 pairs is repeatedxAdjusting to extract source space data dipoleLCorresponding data as the data after channel selectionV(t)，V(t) ∈4N×L×T. By pairsxIs adjusted so thatLNumber of dipoles inNAnd the similarity is close to prevent the phenomenon of under-fitting or over-fitting when the classifier is trained in the later period.
And S5, performing log operation on the extracted electroencephalogram data, and extracting signal features by adopting the CSP.
The common space mode (CSP) is the most common feature of motor imagery electroencephalogram classification, is a spatial filtering feature extraction algorithm under a two-classification task, and adopts an OVO-CSP method to extract the features of a multi-classification task, and the obtained features are shown in FIG. 4.
And S6, inputting the extracted CSP features into a support vector machine for classification to obtain a classification result.
In this embodiment, a multi-class support vector machine classification method is adopted to obtain a classification result, and an obtained confusion matrix is shown in fig. 5. The classification accuracy of the electroencephalogram signal of the classification motor imagery of the four proposed feature extraction method is 93.76%, and the classification accuracy of each subject and the average classification accuracy are shown in fig. 6.
The invention also provides a computer device comprising a memory, a processor and program instructions stored in the memory for execution by the processor, wherein the processor executes the program instructions to implement the method described above.
The invention provides a multi-class signal source space dipole selection method on the basis of tracing electroencephalogram signals to source space, aiming at the problem that the classification accuracy of multi-class electroencephalogram signals is not high. The method effectively reduces the calculated amount, prevents the over-fitting phenomenon during classification, and improves the classification accuracy of the motor imagery electroencephalogram signals.
The present invention has been described in terms of specific examples, which are provided to aid understanding of the invention and are not intended to be limiting. For a person skilled in the art to which the invention pertains, several simple deductions, modifications or substitutions may be made according to the idea of the invention. It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and any modifications, equivalents and improvements made within the spirit and scope of the present invention should be included.
Those not described in detail in this specification are well within the skill of the art.

Claims (9)

1. A motor imagery classification method based on electroencephalogram tracing and dipole selection is characterized by comprising the following steps: the method comprises the following steps:
s1: acquiring electroencephalogram signals based on multi-class motor imagery, and preprocessing the electroencephalogram signals through a band-pass filter and a common average reference;
s2: calculating a head model, a source model, a SourceModel and a lead field;
s3: taking the standardized low-resolution brain magnetoelectric tomography sLORETA as a constraint condition for calculating a source space electroencephalogram signal, and tracing the acquired electroencephalogram signal and a lead field to obtain the source space electroencephalogram signal;
s4: calculating the energy of each category of motor imagery electroencephalogram signals in each dipole of a source space, obtaining a screened source space dipole subset by adopting an improved F-score method and setting a threshold value as a strategy for selecting source space dipoles, and extracting electroencephalogram data of the source space dipole subset; the method comprises the following specific steps:
s4.1: obtaining the average signal of each category of signals of the left hand, the right hand, the tongue and the feet for all the data S (T) of the source space brain electrical signals after tracing to the source, wherein the data S (T) belongs to 4N multiplied by C multiplied by T
Figure FDA0003801916520000011
Figure FDA0003801916520000012
Wherein N represents the experimental times of the electroencephalogram signals of each motor imagery class, and C represents the source spaceThe number of dipoles, T represents a time domain sampling point, i belongs to {1,2,3 and 4}, and represents motor imagery electroencephalogram signals of a left hand, a right hand, a tongue and two feet respectively;
s4.2: calculating the energy P of each kind of average signal in the source space in each dipole i (c)∈C×1，
Figure FDA0003801916520000013
Wherein i ∈ {1,2,3,4};
s4.3: obtaining the improved F-score value F of each kind of signal energy and other kinds of signal energy in each dipole ir (c) And setting the threshold values respectively, selecting F ir (c) Obtaining a selected dipole subset L by using dipoles larger than a threshold value, and extracting an electroencephalogram signal V (t) of a source space in the dipole subset L;
s5: extracting signal features of the extracted electroencephalogram data by adopting a common space mode CSP;
s6: and inputting the CSP features into a support vector machine for classification to obtain a classification result.
2. The motor imagery classification method based on electroencephalogram tracing and dipole selection as claimed in claim 1, wherein the motor imagery classification method comprises the following steps: the source model Sourcemodel in the step S2) is a source model calculated according to the estimated number, position and direction of the dipole sources, and the number of dipoles set in the step S2 is 15002, and the dipoles are uniformly distributed on the cortex.
3. The motor imagery classification method based on electroencephalogram tracing and dipole selection as claimed in claim 1, wherein the motor imagery classification method comprises the following steps: when the head model Headmodel is calculated in step S2), a scalp-cerebrospinal fluid-skull boundary element model BEM is used, and because the electrical conductivity of each structure in the brain is different, the scalp is set in step S2: cerebrospinal fluid: the conductivity of the skull was 1.0125 and the boundary element model used the ICBM152 model.
4. The motor imagery classification method based on electroencephalogram tracing and dipole selection as claimed in claim 1, wherein the motor imagery classification method comprises the following steps: step S4.3 specifically includes:
s4.3.1: calculating improved F-score values F in all dipoles of source space of each category of motor imagery electroencephalogram signal energy and other categories of signal energy ir (c)；
S4.3.2: the F-score values F of the various classes ir (c) Setting x% of the maximum value of the N-type dipoles as threshold, selecting dipoles with improved F-score values larger than the threshold, and adding the dipoles into a set of the selected dipoles;
s4.3.3: and taking the common set of the selected dipole sets in each category as a finally selected dipole set L, if the number of the dipoles selected in the set L is too large or too small, returning to the step S4.3.2 to adjust x% in the threshold value threshold, and taking the data corresponding to the dipole L of the extracted source space data as the data V (T) after channel selection, wherein the V (T) belongs to 4 NxLxT.
5. The motor imagery classification method based on electroencephalogram tracing and dipole selection as claimed in claim 4, wherein the motor imagery classification method comprises the following steps: the improved F-score value F ir (c) In the calculation formula, the numerator is the square of the difference value of different signal energies of the same dipole, the denominator is the sum of the different signal energies of the same dipole, the numerator calculates the discrimination capacity, and the denominator pair performs data normalization operation:
Figure FDA0003801916520000031
wherein X r ＝1/3P rest ，X i ＝P i ，P rest Representing the sum of the source spatial mean signal energies of the other three classes except the i class.
6. The motor imagery classification method based on electroencephalogram tracing and dipole selection as claimed in claim 1, wherein the motor imagery classification method comprises the following steps: and S5) carrying out logarithm operation on the source space signal of the extracted electroencephalogram data, then carrying out CSP signal feature extraction, carrying out logarithm operation to enable the data to conform to normal distribution, reducing the absolute numerical value of the data, and compressing the variable of the scale to enable the data to be more stable.
7. The motor imagery classification method based on electroencephalogram tracing and dipole selection as claimed in claim 6, wherein: and in the step S5, the CSP signal characteristics are extracted by adopting an OTO-CSP method to extract the characteristics of the multi-class tasks.
8. The motor imagery classification method based on electroencephalogram tracing and dipole selection as claimed in claim 4, wherein the motor imagery classification method comprises the following steps: respective class F-score value F ir (c) Is set in the range of [0.15,0.35 ] to x% of the maximum value of]。
9. A computer device comprising a memory, a processor, and program instructions stored in the memory for execution by the processor, wherein the processor executes the program instructions to implement the method of any of claims 1 to 8.
CN202210783534.0A 2022-07-05 2022-07-05 Electroencephalogram tracing and dipole selection-based motor imagery classification method Active CN114861738B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210783534.0A CN114861738B (en) 2022-07-05 2022-07-05 Electroencephalogram tracing and dipole selection-based motor imagery classification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210783534.0A CN114861738B (en) 2022-07-05 2022-07-05 Electroencephalogram tracing and dipole selection-based motor imagery classification method

Publications (2)

Publication Number Publication Date
CN114861738A CN114861738A (en) 2022-08-05
CN114861738B true CN114861738B (en) 2022-10-04

Family

ID=82625724

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210783534.0A Active CN114861738B (en) 2022-07-05 2022-07-05 Electroencephalogram tracing and dipole selection-based motor imagery classification method

Country Status (1)

Country Link
CN (1) CN114861738B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116369950B (en) * 2023-05-25 2024-01-26 武汉理工大学 Target detection method based on electroencephalogram tracing and multi-feature extraction

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109199376B (en) * 2018-08-21 2021-04-09 北京工业大学 Decoding method of motor imagery electroencephalogram signal based on OA-WMNE brain source imaging
CN109199414B (en) * 2018-10-30 2020-11-17 武汉理工大学 Audio-visual evoked emotion recognition method and system based on electroencephalogram signals
CN110531861B (en) * 2019-09-06 2021-11-19 腾讯科技（深圳）有限公司 Method and device for processing motor imagery electroencephalogram signal and storage medium
CN112861778A (en) * 2021-03-05 2021-05-28 南京邮电大学 Multi-mode fusion based emotion classification recognition method

Also Published As

Publication number Publication date
CN114861738A (en) 2022-08-05

Similar Documents

Publication Publication Date Title
Wang et al. Detection analysis of epileptic EEG using a novel random forest model combined with grid search optimization
Zhang et al. Bayesian learning for spatial filtering in an EEG-based brain–computer interface
Gibson et al. Technology-aware algorithm design for neural spike detection, feature extraction, and dimensionality reduction
Baskar et al. Hybrid fuzzy based spearman rank correlation for cranial nerve palsy detection in MIoT environment
Eltrass et al. Automated ECG multi-class classification system based on combining deep learning features with HRV and ECG measures
Fu et al. Automatic detection of epileptic seizures in EEG using sparse CSP and fisher linear discrimination analysis algorithm
Prasanth et al. Deep learning for interictal epileptiform spike detection from scalp EEG frequency sub bands
Alvi et al. Developing a deep learning based approach for anomalies detection from EEG data
CN114861738B (en) Electroencephalogram tracing and dipole selection-based motor imagery classification method
Zhang et al. Channel selection in motor imaginary-based brain-computer interfaces: a particle swarm optimization algorithm
Lian et al. Spatial enhanced pattern through graph convolutional neural network for epileptic EEG identification
Mohammadi et al. Cursor movement detection in brain-computer-interface systems using the K-means clustering method and LSVM
Sartipi et al. Diagnosis of schizophrenia from R-fMRI data using Ripplet transform and OLPP
Sharma et al. A fractal based machine learning method for automatic detection of epileptic seizures using EEG
Liu et al. Epilepsy EEG classification method based on supervised locality preserving canonical correlation analysis
CN113057654B (en) Memory load detection and extraction system and method based on frequency coupling neural network model
Gu et al. Heterogeneous classifier ensembles for EEG-based motor imaginary detection
Sood et al. Design and Development of Prediction Model to Detect Seizure Activity Utilizing Higher Order Statistical Features of EEG signals.
Reaj et al. Emotion recognition using EEG-based brain computer interface
Wang et al. Classification of Single-Trial EEG based on support vector clustering during finger movement
Jayashekar et al. Hybrid Feature Extraction for EEG Motor Imagery Classification Using Multi-Class SVM.
Hu et al. Character encoding-based motor imagery EEG classification using CNN
Fouad et al. Attempts towards the first brain-computer interface system in INAYA Medical College
Karimoi et al. EEG signal classification using Bayes and Naïve Bayes Classifiers and extracted features of Continuous Wavelet Transform
Yu et al. PGMM—pre-trained Gaussian mixture model based convolution neural network for electroencephalography imagery analysis

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant