CN103054574A - Frequency identification method on basis of multivariate synchronous indexes - Google Patents

Abstract

The invention discloses a frequency identification method on the basis of multivariate synchronous indexes. The frequency identification method particularly includes constructing a reference signal corresponding to each frequency according to a frequency of an SSVEP-BCI (steady state visual evoked potential-brain computer interface) system; respectively computing the synchronization indexes among multi-order brain electrical signals and the various reference signals; and finding out the reference signal with the maximum synchronization index with the brain electrical signals, and outputting the frequency corresponding to the reference signal as an identified frequency. The synchronous indexes among the brain electrical signals and the different reference signals constructed on the basis of stimulation on the system are computed, the reference signal with the maximum synchronization index with the brain electrical signals is found out according to magnitudes of the synchronization indexes, and the frequency of the reference signal is outputted as an identification result. Compared with a multi-order frequency detection method mainly used at preset, the frequency identification method is high in accuracy and optimal in performance under the conditions that electrode order number is low and data are short.

Description

Frequency identification method based on the multivariate sync index

Technical field

The invention belongs to the biomedical information technical field, be specifically related to the frequency identification method in brain-computer interface (Brain Computer Interface, the BCI) system.

Background technology

Brain-computer interface can provide direct online communication passage for human or animal and external environment, owing to do not rely on traditional nervus peripheralis and muscle output channel, thereby in neural engineering and neuroscience, have significant application value.

When being subject to greater than the constant extraneous visual stimulus of 4Hz frequency, brain will produce and environmental stimuli frequency or the identical response of its harmonic frequency, i.e. (Regan D (1989) Human brain electrophysiology:evoked potentials and evoked magnetic fields in science and medicine:Elsevier.).Because SSVEP is the endogenous reaction of brain, this class signal has very high signal to noise ratio, very strong robustness and less training, so that based on the brain-computer interface (SSVEP-BCI) of SSVEP higher rate of information transmission is arranged, it is an important directions of BCI on-line system research always.

The SSVEP-BCI system comprises signals collecting, signal processing, several main modular such as application interface.The performance of system depends primarily on the efficient of signal processing module.Therefore, signal processing method is vital fast and accurately.There are very large difference in the amplitude of the SSVEP that difference is tested, distribution and available stimulus frequency.When using present SSVEP-BCI system, in order to obtain preferably performance, must carry out the parameter optimizations such as electrode selection, data segment, length, when especially using traditional signal processing method in system, these optimizing process are necessary.

In recent years, proposed some based on the frequency identification method of leading signal detection, these methods extract more useful informations by Combinatorial Optimization from lead EEG signals more, have improved accuracy of identification, reduce simultaneously the optimizing process such as electrode selection more.Detection method (Minimum Energy Combination based on minimizing energy method, MEC) with based on the method (Canonical CorrelationAnalysis, CCA) of canonical correlation analysis be the employed frequency identification method of leading signal detection two kinds in the present SSVEP-BCI system more.

, lead signal will be original more and carry out projection by seeking spatial filter based on the detection method of minimizing energy method, obtain the low-dimensional composite signal, thus attenuating noise signal and other artefact signals.The method can obtain high accuracy rate, does not need the preliminary experiment data to carry out parameter optimization, has been successfully applied to actual SSVEP-BCI system.

Canonical correlation analysis is a kind of multivariable statistical method, and the method by finding a pair of linear projection vector, so that lead the relevant maximum between EEG signals and the reference signal more.The method has higher accuracy rate and vigorousness than the detection method based on minimizing energy method.

Fast and the high algorithm of accuracy be very important to the SSVEP-BCI system of reality, be the core component of realization high performance system.Lead detection algorithm more and lead signal by optimum organization more, psophometer is revealed larger robustness, thereby improve algorithm performance; This class algorithm needs to carry out parameter optimization hardly simultaneously, thereby has brought more convenience in actual implementation process.But the SSVEP-BCI system of high rate of information transmission requires highly to recognizer, and from experimental result, above-mentioned two kinds of methods (MEC and CCA) resulting accuracy rate and performance all remain further to be improved, to promote the performance of SSVEP-BCI system.

Summary of the invention

The objective of the invention is to have proposed a kind of frequency identification method based on the multivariate sync index in order to solve the problems referred to above of existing many setting frequencies detection method existence.

Technical scheme of the present invention is: a kind of frequency identification method based on the multivariate sync index specifically comprises the steps:

Step 1: according to the employed stimulus frequency f of SSVEP-BCI system ₁, f ₂..., f _K, construct reference signal R corresponding to each frequency _F1, R _F2..., R _Fk

Step 2: calculate respectively the sync index S that leads between EEG signals and each reference signal more ₁, S ₂..., S _K

Step 3: find out the reference signal with EEG signals sync index maximum, the frequency that it is corresponding is as the frequency output of identification.

Further, construct corresponding to frequency f in the step 1 _iReference signal can be calculated as follows:

$R_{\mathrm{fi}}=\left(\begin{array}{c}\sin (2\mathrm{\π}\·f_i\·t)\\ \cos (2\mathrm{\π}\·f_i\·t)\\ \·\·\·\\ \sin (2\mathrm{\π}\·N_h\·f_i\·t)\\ \cos (2\mathrm{\π}\·N_h\·f_i\·t)\end{array}\right),$ $t=\frac{1}{\mathrm{Fs}},\frac{2}{\mathrm{Fs}},\·\·\·,\frac{M}{\mathrm{Fs}}$

F _sBe sample rate, M is sample number.

Further, the detailed process of calculating sync index is as follows in the step 2:

Be X if lead the EEG signals matrix, reference signal is Y more, calculates the associating correlation matrix of X and Y:

$C=\left[\begin{array}{cc}C11& C12\\ C21& C22\end{array}\right]$

Wherein,

$C11=\frac{1}{M}{\mathrm{XX}}^T$

$C22=\frac{1}{M}{\mathrm{YY}}^T$

$C12=C21=\frac{1}{M}{\mathrm{XY}}^T$

Carry out following linear transformation:

$U=\left[\begin{array}{cc}{C11}^{-\frac{1}{2}}& 0\\ 0& \mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA00002707373100021.png"\; state="\{\{state\}\}">C</figure-callout>}22^{-\frac{1}{2}}\end{array}\right]$

Obtain:

$R={\mathrm{UCU}}^T=\left[\begin{array}{cc}{I}_{N\&times;N}& {C11}^{-\frac{1}{2}}C12{C22}^{-\frac{1}{2}}\\ {C22}^{-\frac{1}{2}}{C21C11}^{-\frac{1}{2}}& I_{2N_h\&times;{2N}_h}\end{array}\right]$

Wherein, I _{N * N}Be N dimension unit square formation,

Be 2N _hDimension unit square formation, N is electrode number, N _hReference signal harmonic wave quantity.

Matrix R is carried out Eigenvalues Decomposition, obtain its eigenvalue λ ₁, λ ₂..., λ _P, the column criterion of going forward side by side:

${\mathrm{\&lambda;}}_i^{\&prime;}=\frac{{\mathrm{\&lambda;}}_i}{\underset{i=1}{\overset{P}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i}=\frac{{\mathrm{\&lambda;}}_i}{\mathrm{tr}\left(R\right)}$

Wherein, P=N+2N _h

Last sync index of leading between EEG signals and this reference signal can be calculated as more:

Beneficial effect of the present invention: the present invention proposes a kind of based on multivariate sync index (Multivarite Synchronization Index, MSI) frequency identification method, in the method, adopt the sync index of two multidimensional signals as characteristic of division, EEG signals in the SSVEP-BCI system is carried out frequency identification, its core be calculate EEG signals and different reference signals that the stimulus frequency that uses based on system constructs between sync index, size according to sync index, find out the reference signal with EEG signals sync index maximum, with the result's output as identification of the frequency of this reference signal.Compare with existing employed main many setting frequencies detection method, have higher accuracy rate; And few at the electrode derivative, under the shorter condition of data length, have optimum performance.Method of the present invention can be accelerated the response speed of SSVEP-BCI system effectively, improves the performance of system.

Description of drawings

Fig. 1 is based on the schematic flow sheet of the frequency identification method of multivariate sync index (MSI).

The emulation experiment comparing result schematic diagram of Fig. 2 method of the present invention and existing two kinds of methods.

The true brain electricity experiment comparing result schematic diagram of Fig. 3 method of the present invention and existing two kinds of methods.

The specific embodiment

The present invention is described further below in conjunction with the drawings and specific embodiments.

The signal calculated synchronicity can have a lot of methods, and the SSVEP-BCI system is to the operational efficiency requirement very high (algorithm must provide the recognition result of current EEG signals within less than 1 second time) of recognizer.Therefore, in the frequency identification framework based on multivariate sync index (MSI), provide following a kind of efficient frequency identification method.

Suppose that EEG signals is that X(N * M ties up matrix), reference signal is Y(2N _h* M ties up matrix).Here, N is electrode number, and M is sample number, N _hReference signal harmonic wave quantity.Be not general, standardization of X and Y has the zero-mean unit variance.Below discuss in detail implementation process based on the frequency detecting method of multivariate sync index:

At first, calculate the associating correlation matrix of X and Y

$C=\left[\begin{array}{cc}C11& C12\\ C21& C22\end{array}\right]---\left(1\right)$

Wherein,

$C11=\frac{1}{M}{\mathrm{XX}}^T---\left(2\right)$

$C22=\frac{1}{M}{\mathrm{YY}}^T---\left(3\right)$

$C12=C21=\frac{1}{M}{\mathrm{XY}}^T---\left(4\right)$

C comprises X, Y self correlation and X and Y cross-correlation, in order to weaken self correlation to the impact of sync index, carries out following linear transformation:

$U=\left[\begin{array}{cc}{C11}^{-\frac{1}{2}}& 0\\ 0& \mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA00002707373100021.png"\; state="\{\{state\}\}">C</figure-callout>}22^{-\frac{1}{2}}\end{array}\right]---\left(5\right)$

Then obtain:

$R={\mathrm{UCU}}^T=\left[\begin{array}{cc}{I}_{N\&times;N}& {C11}^{-\frac{1}{2}}C12{C22}^{-\frac{1}{2}}\\ {C22}^{-\frac{1}{2}}{C21C11}^{-\frac{1}{2}}& I_{2N_h\&times;{2N}_h}\end{array}\right]---\left(6\right)$

I _{N * N}Be N dimension unit square formation,

Be 2N _hDimension unit square formation.

Matrix R is carried out Eigenvalues Decomposition, obtain its eigenvalue λ ₁, λ ₂..., λ _P, the column criterion of going forward side by side

${\mathrm{\&lambda;}}_i^{\&prime;}=\frac{{\mathrm{\&lambda;}}_i}{\underset{i=1}{\overset{P}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i}=\frac{{\mathrm{\&lambda;}}_i}{\mathrm{tr}\left(R\right)}---\left(7\right)$

Here P=N+2N _h

Sync index between last EEG signals and the reference signal can be calculated as:

$S=1+\frac{\underset{i=1}{\overset{P}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i^{\&prime;}\log \left({\mathrm{\&lambda;}}_i^{\&prime;}\right)}{\log \left(P\right)}---\left(9\right)$

Suppose that there be K stimulus frequency f in the SSVEP-BCI system ₁, f ₂..., f _K, then corresponding to frequency f _iReference signal can be calculated as follows:

$R_{\mathrm{fi}}=\left(\begin{array}{c}\sin (2\mathrm{\&pi;}\&CenterDot;f_i\&CenterDot;t)\\ \cos (2\mathrm{\&pi;}\&CenterDot;f_i\&CenterDot;t)\\ \&CenterDot;\&CenterDot;\&CenterDot;\\ \sin (2\mathrm{\&pi;}\&CenterDot;N_h\&CenterDot;f_i\&CenterDot;t)\\ \cos (2\mathrm{\&pi;}\&CenterDot;N_h\&CenterDot;f_i\&CenterDot;t)\end{array}\right),$ $t=\frac{1}{\mathrm{Fs}},\frac{2}{\mathrm{Fs}},\&CenterDot;\&CenterDot;\&CenterDot;,\frac{M}{\mathrm{Fs}}---\left(10\right)$

F _sBe sample rate.

According to (1)-(9), can calculate the sync index of all reference signals and EEG signals, and then obtain K sync index S ₁, S ₂..., S _K, carry out last frequency identification by following formula:

$T=\underset{i}{\max }{S}_i,i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,K---\left(\text{11}\right)$

Be the corresponding frequency of current EEG signals for and EEG signals between have the frequency of the reference signal of maximum sync index.

For the more specifically mentioned SSVEP-BCI system frequency recognition methods of explanation invention, further specify in conjunction with Fig. 1.

As shown in Figure 1, lead more EEG signals X respectively with K reference signal R _F1, R _F2..., R _FkAs the input of the inventive method, obtain K sync index S ₁, S ₂..., S _K, then obtain maximum in K the sync index.According to this maximum, find corresponding reference signal, the employed frequency of this reference signal is as the Output rusults of the inventive method.

Be feasibility and the effect of verifying method of the present invention, adopt 3 class frequencys to carry out simulating, verifying, compare with existing detection method based on minimizing energy method (MEC) with based on the method (CCA) of canonical correlation analysis simultaneously.The frequency that adopts is as follows:

A）27Hz,29Hz,31Hz,33Hz,35Hz,37Hz,39Hz,41Hz?and43Hz；

B）8Hz,9Hz,10Hz,11Hz,12Hz,13Hz,14Hz,15Hz；

C）6.7Hz,7.5Hz,8.6Hz,10Hz,12Hz,15Hz；

Simulate 4 with the sinusoidal signal under 4 these frequencies of each frequency generation and lead EEG signals, Chief Signal Boatswain 10s, sample rate is 250Hz.Simulate the true EEG signals that is subjected to sound pollution to whenever leading signal by certain signal to noise ratio interpolation white Gaussian noise; Then the signal under every class frequency is carried out frequency identification, obtain recognition accuracy, the signal length that is used for frequency identification is 1s; Every class frequency repeats 50 such operations, and as the algorithm performance evaluation index under this signal to noise ratio, the signal to noise ratio scope is from-7db to-20db with 50 results' average recognition accuracy.

The defined formula of signal to noise ratio is as follows:

$\mathrm{SNR}=10\log \frac{P_{\mathrm{signal}}}{P_{\mathrm{noise}}}=10\log \frac{(A/\sqrt{2})}{{\mathrm{\&sigma;}}^2}---\left(12\right)$

P _SignalBe the energy of signal, P _NoiseBe noise energy, A sinusoidal signal amplitude, σ ²Be noise variance.

Concrete simulation result as shown in Figure 2, wherein, (a) employed frequency is 27Hz, 29Hz, 31Hz, 33Hz, 35Hz, 37Hz, 39Hz, 41Hz and43Hz; (b) employed frequency is 8Hz, 9Hz, 10Hz, 11Hz, 12Hz, 13Hz, 14Hz, 15Hz; (c) employed frequency is 6.7Hz, 7.5Hz, 8.6Hz, 10Hz, 12Hz, 15Hz.

* represent that MSI and CCA have significant difference under this condition, MSI has significant difference with MEC under this condition of+expression, and the result uses pairing T to check p＜0.05.

From simulation result, the result of the inventive method is best.For first group of high frequency frequency sets and the set of second group of Frequency, when signal to noise ratio during greater than-12db, method of the present invention is compared with existing two kinds of methods has significant difference, illustrates that method of the present invention has stronger robustness to noise.In the 3rd class frequency, because have the frequency content of harmonic relationships, all algorithms all can not reach 100% accuracy rate, but method of the present invention and existing two kinds of methods remain significant difference.

In addition, adopt the effectiveness of the further verification algorithm of true EEG signals.In experiment, adopt 8 eeg collection systems that lead, 4 kinds of frequency 7.5Hz, 8.6Hz, 10Hz, 12Hz gathers tested 30s EEG signals under every kind of frequency.11 tested (21-28 years) have participated in demonstration test, and experimental result as shown in Figure 3.Among the figure, (a) 4 lead the brain electricity, (b) 6 lead the brain electricity, (c) 8 lead the brain electricity.* represent that MSI and CCA have significant difference under this condition, MSI has significant difference with MEC under this condition of+expression, and the result uses pairing T to check p＜0.05.

From the result, methods and results of the present invention is all better than existing two kinds of methods, is especially only leading EEG signals with 4, and signal length is in the 1s situation, and there are significant difference in method of the present invention and existing two kinds of methods and resultses.Electrodeplate is few, brings more convenience can for the application of SSVEP-BCI system.What is more important, the signal length that is used for frequency identification is shorter, and the algorithm that accuracy rate is higher more can reduce the response time of system, improves the response speed of system, and therefore method of the present invention has the performance that larger potential goes to improve the SSVEP-BCI system.

On the whole, emulation experiment and true experiment show the present invention program's effectiveness and feasibility.

Those of ordinary skill in the art will appreciate that, embodiment described here is in order to help reader understanding's principle of the present invention, should to be understood to that protection scope of the present invention is not limited to such special statement and embodiment.Those of ordinary skill in the art can make various other various concrete distortion and combinations that do not break away from essence of the present invention according to these technology enlightenments disclosed by the invention, and these distortion and combination are still in protection scope of the present invention.

Claims (3)

1. the frequency identification method based on the multivariate sync index specifically comprises the steps:

Step 1: according to the employed stimulus frequency f of SSVEP-BCI system ₁, f ₂..., f _K, construct reference signal R corresponding to each frequency _F1, R _F2..., R _Fk

Step 2: calculate respectively the sync index S that leads between EEG signals and each reference signal more ₁, S ₂..., S _K

Step 3: find out the reference signal with EEG signals sync index maximum, the frequency that it is corresponding is as the frequency output of identification.

2. frequency identification method according to claim 1 is characterized in that, structure is corresponding to frequency f in the step 1 _iReference signal can be calculated as follows:

$R_{\mathrm{fi}}=\left(\begin{array}{c}\sin (2\mathrm{\&pi;}\&CenterDot;f_i\&CenterDot;t)\\ \cos (2\mathrm{\&pi;}\&CenterDot;f_i\&CenterDot;t)\\ \&CenterDot;\&CenterDot;\&CenterDot;\\ \sin (2\mathrm{\&pi;}\&CenterDot;N_h\&CenterDot;f_i\&CenterDot;t)\\ \cos (2\mathrm{\&pi;}\&CenterDot;N_h\&CenterDot;f_i\&CenterDot;t)\end{array}\right),$ $t=\frac{1}{\mathrm{Fs}},\frac{2}{\mathrm{Fs}},\&CenterDot;\&CenterDot;\&CenterDot;,\frac{M}{\mathrm{Fs}}$

F _sBe sample rate, M is sample number.

3. frequency identification method according to claim 1 and 2 is characterized in that, the detailed process of calculating sync index in the step 2 is as follows:

Be X if lead the EEG signals matrix, the reference signal matrix is Y more, calculates the associating correlation matrix of X and Y:

$C=\left[\begin{array}{cc}C11& C12\\ C21& C22\end{array}\right]$

Wherein,

$C11=\frac{1}{M}{\mathrm{XX}}^T$

$C22=\frac{1}{M}{\mathrm{YY}}^T$

$C12=C21=\frac{1}{M}{\mathrm{XY}}^T$

Carry out following linear transformation:

$U=\left[\begin{array}{cc}{C11}^{-\frac{1}{2}}& 0\\ 0& C22^{-\frac{1}{2}}\end{array}\right]$

Obtain:

$R={\mathrm{UCU}}^T=\left[\begin{array}{cc}{I}_{N\&times;N}& {C11}^{-\frac{1}{2}}C12{C22}^{-\frac{1}{2}}\\ {C22}^{-\frac{1}{2}}{C21C11}^{-\frac{1}{2}}& I_{2N_h\&times;{2N}_h}\end{array}\right]$

I _{N * N}Be N dimension square formation,

Be 2N _hThe dimension square formation, N is electrode number, N _hReference signal harmonic wave quantity;

Matrix R is carried out Eigenvalues Decomposition, obtain its eigenvalue λ ₁, λ ₂..., λ _P, the column criterion of going forward side by side:

${\mathrm{\&lambda;}}_i^{\&prime;}=\frac{{\mathrm{\&lambda;}}_i}{\underset{i=1}{\overset{P}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i}=\frac{{\mathrm{\&lambda;}}_i}{\mathrm{tr}\left(R\right)}$

Wherein, P=N+2N _h

Can obtain at last leading the sync index between EEG signals and this reference signal more:

Images