CN110974221A - Mixed function correlation vector machine-based mixed brain-computer interface system - Google Patents
Mixed function correlation vector machine-based mixed brain-computer interface system Download PDFInfo
- Publication number
- CN110974221A CN110974221A CN201911323417.0A CN201911323417A CN110974221A CN 110974221 A CN110974221 A CN 110974221A CN 201911323417 A CN201911323417 A CN 201911323417A CN 110974221 A CN110974221 A CN 110974221A
- Authority
- CN
- China
- Prior art keywords
- kernel function
- brain
- computer interface
- mixed
- signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- 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]
- A61B5/377—Electroencephalography [EEG] using evoked responses
-
- 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]
- A61B5/377—Electroencephalography [EEG] using evoked responses
- A61B5/378—Visual stimuli
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
Abstract
The invention discloses a kernel function optimization method of a decision model of a non-invasive mixed normal brain-computer interface based on portable EEG equipment. Considering the mixed normal brain-computer interface combining several brain-computer interface normal forms, the mixed normal brain-computer interface can meet the requirement of multi-instruction output and has higher classification accuracy. The decision method uses a sparse Bayesian-based relevance vector machine classifier, needs more sparse relevance vectors, is shorter in training time, and is also shorter in testing time of the model. The model parameters are automatically determined. The kernel function is not limited by mercer theorem in selection, and a new kernel function can be arbitrarily constructed.
Description
Technical Field
The invention relates to a non-invasive EEG-based kernel function optimization method for a decision model of a hybrid normal brain-computer interface, and belongs to the technical field of brain-computer interfaces.
Background
The brain-computer interface is a new interactive mode that can realize direct communication between the human brain and the external environment (computer or external equipment). By collecting information of scalp potential activity of a human brain, different spontaneous brain potential activities or brain potential activities induced by external stimuli are reflected. The brain-computer interface has potential application value in the aspects of education, medical rehabilitation, disease diagnosis and the like. An Electroencephalogram (EEG) commonly used in a brain-machine interface is a widely used method in non-invasive brain-machine interface research due to convenient acquisition and low cost.
The brain-computer interface is usually tested and analyzed in a single paradigm, and the common brain-computer interface paradigm is: (1) a brain-machine interface based on motor imagery; (2) a P300-based brain-machine interface; (3) brain-machine interface based on steady-state visual evoked potentials (SSVEPs).
The single-model brain-computer interface can complete the output of control instructions of limited tasks, and the information transmission rate is low due to the fact that the number of categories of decision tasks is small and the classification accuracy is low. However, in order to meet the actual requirements, when the real-time application of the brain-computer interface is realized, the mixed normal brain-computer interface combining several brain-computer interface normal forms is considered, the mixed normal brain-computer interface can meet the requirement of multi-instruction output, and has higher classification accuracy.
The decision method uses a sparse Bayesian-based relevance vector machine classifier, needs more sparse relevance vectors, is shorter in training time, and is also shorter in testing time of the model. The model parameters are automatically determined. The kernel function is not limited by mercer theorem in selection, and a new kernel function can be arbitrarily constructed.
Based on a mixed normal brain-computer interface combining motor imagery and SSVEP, four types of imagery tasks are designed in the motor imagery normal form to achieve output of four control instructions, and four stimulation frequencies are designed in the SSVEP normal form to achieve output of the four control instructions. Flexible control of external devices can be achieved in a multiple instruction output system. Meets the requirements of real-time performance and high Information Transfer Rate (ITR).
Embodiments of the present invention seek to improve upon current brain-machine interface systems by providing a hybrid paradigm brain-machine interface system.
Content of patent
EEG signal acquisition: the electroencephalogram acquisition equipment is a Brianup device, and the left ear electrode and the right ear electrode are respectively used as a ground electrode and a reference electrode. Motor imagery paradigm EEG signal acquisition. The electrode position conforms to the 10-20 international standard electrode position. The sampling frequency of the signal is 250Hz, and the power frequency interference is removed through 50Hz notch filtering.
Signal preprocessing: for raw EEG signals, limited by the impact of the acquisition equipment and experimental environment, and the interference signals caused by blinking of the subject, pre-processing of the raw EEG signals is required to remove noise interference signals. A common approach is to have a low-pass Butterworth filter to select the desired frequency range signal.
And (3) signal feature extraction: aiming at the problems that the EEG signal channels are few and useful information is insufficient, a phase space reconstruction method is provided for reconstructing a one-dimensional EEG time sequence with few channels, reconstructing a one-dimensional EEG signal to a high-dimensional phase space, and observing richer dynamic characteristic information in the phase space. And simultaneously, the dimensionality and the channel number of the signal are increased. For a one-dimensional EEG time series of length N x (t) ═ x1,x2,…,xNIt can be reconstructed into m-dimensional phase space. The specific expression of the reconstructed phase space trajectory is as follows:
where M is N- (M-1) τ, τ is the delay time, and M is the embedding dimension. The delay time and the embedding dimension can be obtained by calculating the correlation integral by using a C-C method.
And extracting spatial features from the reconstructed EEG signals in the phase space by adopting a common spatial mode.
For two classes of EEG signals XiI ∈ {1, 2}, and the dimension of the signal is T × N, where T is the number of sample points in the time series and N is the number of channels.
By constructing the filter W, the filtered original EEG signal X results in a signature signal Z,
Z=WX (2)
the variance of the filtered feature signal Z is classified as a signal feature, in the following specific form,
the filter W maximizes the variance of one type of signal and minimizes the variance of the other type of signal by extracting the first m rows and the last m rows that best reflect the main differences of the two types of signals.
And (3) signal classification: a relevance vector machine classifier is adopted as a classifier, and the RVM is a machine learning model based on sparse Bayes. The method has the advantages that parameters can be automatically determined, the testing time is short, the probability output is obtained while the binary classification result is obtained, and compared with the SVM (support vector machine), the RVM is sparser, so that the testing time is shorter. RVMs have better generalization capability than SVMs. The training time for RVMs is shorter. The choice of kernel function is not limited by the mercer theorem.
Commonly used kernel functions include a local kernel function with an interpolation characteristic, a gaussian kernel function, and a global kernel function polynomial kernel function with an extrapolation characteristic. The mixed kernel function of the Gaussian kernel function and the polynomial kernel function is obtained through linear fitting, has good interpolation and extrapolation, and is a new kernel function with local and global characteristics. The new mixed kernel function RVM can improve classification performance when processing data.
Gaussian kernel function:
polynomial kernel function:
Kpoly(xi,xj)=(<xi,xj>+1)d(5)
mixed kernel function:
Khybrid=λ1Kgauss+λ2Kpoly(6)
the SSVEP signal is an induced signal of a specific stimulation frequency, an improved Filter band typical correlation analysis (FBCCA) is adopted, and by designing different pass frequency bands, the fundamental frequency, the second frequency and the third frequency of the stimulation frequency are sequentially filtered until the fifth frequency is reached, and finally, the discriminant analysis is carried out.
Wherein w (n) ═ n-1,n∈[1,N]. Obtaining a series of correlation coefficient valuesAnd selecting the maximum value, wherein the maximum value corresponds to the final classification category.
Drawings
FIG. 1 is a flow chart of a hybrid brain-computer interface algorithm
FIG. 2 is a Gaussian kernel function graph
FIG. 3 is a graph of a polynomial kernel function
FIG. 4 is a graph of a mixed kernel function
FIG. 5 SSVEP stimulation interface design
Detailed Description
The technical scheme of the invention is explained in detail in the following with reference to the attached drawings.
Fig. 1 is a flow chart of the algorithm of the hybrid paradigm brain-computer interface system of the present invention. Firstly, brain electroencephalogram signals are collected through electroencephalogram signal collecting equipment, and the collected electroencephalogram signals are processed by an EEG signal moving image (EEG) signal. After the signal is subjected to Butterworth low-pass filtering, phase space reconstruction parameters are calculated, and the embedding dimension m and the delay time t are calculated, so that the one-dimensional EEG signal is reconstructed into the phase space. A "one-to-one" co-spatial mode feature extraction of EEG signals is performed in phase space. And sending the extracted features into a related vector machine classifier for task classification and identification. The invention designs four motor imagery tasks (left hand, right hand, double feet and empty), wherein when the 'empty' is identified, the signal processing of the SSVEP paradigm is selected to enter. The SSVEP signals are obtained by the induction of stimulation interfaces with four flicker frequencies, and are classified and identified by a typical correlation analysis method based on a filter frequency band.
In the feature extraction part of the motor imagery EEG signal, the invention introduces a phase space reconstruction method, and rich dynamic features in the phase space can be found out by reconstructing the one-dimensional EEG signal into a high-dimensional phase space, and meanwhile, the information dimensionality is increased. By means of a one-to-one common space mode feature extraction method, four types of task signals are divided into six groups of features which are respectively used as feature vectors.
A machine learning model based on sparse Bayesian is adopted as a classifier. The method has the advantages that parameters can be automatically determined, the testing time is short, the probability output is obtained while the binary classification result is obtained, and compared with the SVM (support vector machine), the RVM is sparser, so that the testing time is shorter. RVMs have better generalization capability than SVMs. The training time for RVMs is shorter. The choice of kernel function is not limited by the mercer theorem.
Fig. 2 is a gaussian kernel function graph. The Gaussian kernel function is a local kernel function and has strong interpolation characteristics.
Fig. 3 is a polynomial kernel function graph. The polynomial kernel is a global kernel and has strong extrapolation characteristics.
Based on the characteristics of the Gaussian kernel function and the polynomial kernel function, the invention linearly combines the two kernel functions to obtain a mixed kernel function.
FIG. 4 is a mixing kernel function. A plot of a gaussian kernel and a polynomial kernel in different linear combinations. It can be seen that the larger the specific gravity of the gaussian kernel, the more pronounced the local nature of the mixed kernel. The larger the specific gravity of the polynomial kernel, the more apparent the global nature of the hybrid kernel.
And determining the optimal mixed kernel function correlation vector machine classifier by selecting a proper weight combination.
Fig. 5 is an SSVEP stimulation interface. By designing the stimulation interfaces of four different flicker frequencies, up, down, left, right, each of which is a black and white block of 100 x 100 pixels, the stimulation interfaces were obtained by the psychoolbox psychology toolbox design in MATLAB. And carrying out classified identification on the SSVEP signals by an FBCCA method.
Claims (4)
1. A kernel function optimization method of a decision model of a non-invasive mixed paradigm brain-machine interface.
The method is characterized in that: the mixed normal brain-computer interface combines several brain-computer interface normal forms, can meet the requirement of multi-instruction output, and has higher classification accuracy.
The first step is as follows: and (3) signal feature extraction: aiming at the problems that the channels of the electroencephalogram signals are few and useful information is insufficient, a phase space reconstruction method is provided for reconstructing the one-dimensional EEG time sequence with few channels;
the second step is that: the one-dimensional EEG signal is reconstructed to a high-dimensional phase space where more abundant kinetic feature information is observed. While increasing the signal dimension and number of channels.
2. The model function optimization method of claim 1, wherein:
the first step is as follows: and (3) signal classification: a relevance vector machine classifier is adopted as a classifier, and the RVM is a machine learning model based on sparse Bayes.
The second step is that: the method has the advantages that parameters are automatically determined, the testing time is short, the probability output is obtained while the binary classification result is obtained, and compared with the SVM, the RVM is sparser, so that the testing time is shorter.
The third step: the generalization ability is improved by RVM to make SVM better.
The fourth step: the shortened model training time by RVM is shorter.
3. The kernel function of claim 1, wherein the global kernel function:
the first step is as follows: commonly used kernel functions include a local kernel function with an interpolation characteristic, a gaussian kernel function, and a global kernel function polynomial kernel function with an extrapolation characteristic.
The second step is that: the mixed kernel function of the Gaussian kernel function and the polynomial kernel function is obtained through linear fitting, and the method has good interpolation and extrapolation.
The third step: a new kernel function is constructed having local and global properties.
The fourth step: the new mixed kernel function RVM can improve classification performance when processing data.
4. A portable EEG device based kernel function discrimination system according to claim 1, wherein: the SSVEP signal is an induced signal of a specific stimulation frequency, an improved Filter band typical correlation analysis (FBCCA) is adopted, and by designing different pass frequency bands, the fundamental frequency, the second frequency and the third frequency of the stimulation frequency are sequentially filtered until the fifth frequency is reached, and finally, the discriminant analysis is carried out. And obtaining a series of correlation coefficient values, and selecting the maximum value of the correlation coefficient values, wherein the maximum value corresponds to the final classification class.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911323417.0A CN110974221A (en) | 2019-12-20 | 2019-12-20 | Mixed function correlation vector machine-based mixed brain-computer interface system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911323417.0A CN110974221A (en) | 2019-12-20 | 2019-12-20 | Mixed function correlation vector machine-based mixed brain-computer interface system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110974221A true CN110974221A (en) | 2020-04-10 |
Family
ID=70073319
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911323417.0A Pending CN110974221A (en) | 2019-12-20 | 2019-12-20 | Mixed function correlation vector machine-based mixed brain-computer interface system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110974221A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111736690A (en) * | 2020-05-25 | 2020-10-02 | 内蒙古工业大学 | Motor imagery brain-computer interface based on Bayesian network structure identification |
CN114209343A (en) * | 2021-04-29 | 2022-03-22 | 上海大学 | Portable attention training system and method based on AR and SSVEP |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120172743A1 (en) * | 2007-12-27 | 2012-07-05 | Teledyne Licensing, Llc | Coupling human neural response with computer pattern analysis for single-event detection of significant brain responses for task-relevant stimuli |
WO2014069996A1 (en) * | 2012-10-29 | 2014-05-08 | Stichting Katholieke Universiteit | Method and system for a brain-computer interface |
CN104571533A (en) * | 2015-02-10 | 2015-04-29 | 北京理工大学 | Device and method based on brain-computer interface technology |
CN106874935A (en) * | 2017-01-16 | 2017-06-20 | 衢州学院 | SVMs parameter selection method based on the fusion of multi-kernel function self adaptation |
CN107656612A (en) * | 2017-09-06 | 2018-02-02 | 天津大学 | Big instruction set brain-machine interface method based on P300 SSVEP |
CN108459714A (en) * | 2018-03-05 | 2018-08-28 | 东南大学 | Few channel asynchronous controlling brain machine interface system based on the bis- normal forms of MI and SSVEP |
-
2019
- 2019-12-20 CN CN201911323417.0A patent/CN110974221A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120172743A1 (en) * | 2007-12-27 | 2012-07-05 | Teledyne Licensing, Llc | Coupling human neural response with computer pattern analysis for single-event detection of significant brain responses for task-relevant stimuli |
WO2014069996A1 (en) * | 2012-10-29 | 2014-05-08 | Stichting Katholieke Universiteit | Method and system for a brain-computer interface |
CN104571533A (en) * | 2015-02-10 | 2015-04-29 | 北京理工大学 | Device and method based on brain-computer interface technology |
CN106874935A (en) * | 2017-01-16 | 2017-06-20 | 衢州学院 | SVMs parameter selection method based on the fusion of multi-kernel function self adaptation |
CN107656612A (en) * | 2017-09-06 | 2018-02-02 | 天津大学 | Big instruction set brain-machine interface method based on P300 SSVEP |
CN108459714A (en) * | 2018-03-05 | 2018-08-28 | 东南大学 | Few channel asynchronous controlling brain machine interface system based on the bis- normal forms of MI and SSVEP |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111736690A (en) * | 2020-05-25 | 2020-10-02 | 内蒙古工业大学 | Motor imagery brain-computer interface based on Bayesian network structure identification |
CN114209343A (en) * | 2021-04-29 | 2022-03-22 | 上海大学 | Portable attention training system and method based on AR and SSVEP |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Nakisa et al. | Evolutionary computation algorithms for feature selection of EEG-based emotion recognition using mobile sensors | |
Liang et al. | Scalp EEG epileptogenic zone recognition and localization based on long-term recurrent convolutional network | |
Fang et al. | Extracting features from phase space of EEG signals in brain–computer interfaces | |
CN111329474B (en) | Electroencephalogram identity recognition method and system based on deep learning and information updating method | |
Hsu | Application of quantum-behaved particle swarm optimization to motor imagery EEG classification | |
Hsu | Application of competitive Hopfield neural network to brain-computer interface systems | |
KR102141185B1 (en) | A system of detecting epileptic seizure waveform based on coefficient in multi-frequency bands from electroencephalogram signals, using feature extraction method with probabilistic model and machine learning | |
Rejer | EEG feature selection for BCI based on motor imaginary task | |
CN110974221A (en) | Mixed function correlation vector machine-based mixed brain-computer interface system | |
Madanu et al. | Depth of anesthesia prediction via EEG signals using convolutional neural network and ensemble empirical mode decomposition | |
Tian et al. | Study on classification of left-right hands motor imagery EEG signals based on CNN | |
Christoforou et al. | Second-order bilinear discriminant analysis | |
CN110141211B (en) | Steady-state visual evoked potential classification method based on empirical mode decomposition | |
Dagdevir et al. | Determination of effective signal processing stages for brain computer interface on BCI competition IV data set 2b: a review study | |
Saidi et al. | FPGA implementation of EEG signal analysis system for the detection of epileptic seizure | |
Nguyen et al. | EEG data classification using wavelet features selected by Wilcoxon statistics | |
CN116421200A (en) | Brain electricity emotion analysis method of multi-task mixed model based on parallel training | |
Aldea et al. | Classifications of motor imagery tasks using k-nearest neighbors | |
KR101345216B1 (en) | Mehtod and apparatus for analyzing multi-subject electroencephalograph | |
Zhou et al. | Morphology-based wavelet features and multiple mother wavelet strategy for spike classification in EEG signals | |
Fauzi et al. | Automatic energy extraction methods for EEG channel selection | |
Liu et al. | Epilepsy EEG classification method based on supervised locality preserving canonical correlation analysis | |
Homri et al. | A hybrid cascade method for EEG classification | |
Manshouri et al. | Classification of human vision discrepancy during watching 2D and 3D movies based on EEG signals | |
Pratyusha et al. | Motor imagery based eeg signal classification using multi-scale cnn architecture |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200410 |