CN103054574B

CN103054574B - Frequency identification method on basis of multivariate synchronous indexes

Info

Publication number: CN103054574B
Application number: CN201310003618.9A
Authority: CN
Inventors: 张杨松; 徐鹏; 尧德中
Original assignee: University of Electronic Science and Technology of China
Current assignee: University of Electronic Science and Technology of China
Priority date: 2013-01-06
Filing date: 2013-01-06
Publication date: 2014-07-09
Anticipated expiration: 2033-01-06
Also published as: CN103054574A

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 multivariate synchronous index

Technical Field

The invention belongs to the technical field of biomedical information, and particularly relates to a frequency identification method in a Brain-Computer Interface (BCI) system.

Background

The brain-computer interface can provide a direct online communication channel for human or animals and the external environment, and has important application value in nerve engineering and neuroscience because of not depending on the traditional peripheral nerve and muscle output channel.

When exposed to an external visual stimulus of constant frequency greater than 4Hz, the brain will produce a response equal to the frequency of the external stimulus or its harmonic frequencies, i.e., Regan D (1989) Human brain anatomical: exposed potentials and exposed magnetic fields in science and medicine: Elsevier. Since SSVEP is an endogenous response of the brain, such signals have high signal-to-noise ratio, strong robustness and less training, so that the SSVEP-based brain-computer interface (SSVEP-BCI) has high information transmission rate, and is an important direction for BCI online system research.

The SSVEP-BCI system comprises a signal acquisition module, a signal processing module, an application interface module and the like. The performance of the system depends mainly on the efficiency of the signal processing module. Therefore, fast and accurate signal processing methods are crucial. The amplitude, distribution, and available stimulation frequency of the SSVEPs of the different subjects vary greatly. When the current SSVEP-BCI system is used, parameter optimization such as electrode selection, data segment length and the like is required to obtain better performance, and particularly when the system uses a traditional signal processing method, the optimization processes are required.

In recent years, frequency identification methods based on multi-conductor signal detection have been proposed, which extract more useful information by combining and optimizing multi-conductor brain electrical signals, thereby improving identification accuracy and reducing optimization processes such as electrode selection. The Minimum Energy Combination (MEC) and the typical correlation analysis (CCA) are two frequency identification methods for detecting the multi-pilot signals used in the SSVEP-BCI system.

The detection method based on the minimum energy method is characterized in that the original multi-lead signals are projected by searching a spatial filter to obtain low-dimensional combined signals, so that noise signals and other artifact signals are weakened. The method can obtain high accuracy, does not need to carry out parameter optimization on pre-experimental data, and is successfully applied to an actual SSVEP-BCI system.

The canonical correlation analysis is a multivariate statistical method that maximizes the correlation between the multi-lead brain signal and the reference signal by finding a pair of linear projection vectors. The method has higher accuracy and robustness than the detection method based on the minimum energy method.

The algorithm with high speed and high accuracy is very important for the actual SSVEP-BCI system and is a core component for realizing a high-performance system. The multi-pilot detection algorithm has higher robustness on noise by optimally combining multi-pilot signals, so that the performance of the algorithm is improved; meanwhile, the algorithm almost does not need parameter optimization, thereby bringing more convenience in the actual implementation process. However, the SSVEP-BCI system with a high information transmission rate has a high requirement on the identification algorithm, and from the experimental result, the accuracy and performance obtained by the two methods (MEC and CCA) need to be further improved, so as to improve the performance of the SSVEP-BCI system.

Disclosure of Invention

The invention aims to solve the problems of the existing multi-derivative frequency detection method and provides a frequency identification method based on a multivariable synchronization index.

The technical scheme of the invention is as follows: a frequency identification method based on multivariate synchronous indexes specifically comprises the following steps:

step 1: according to the stimulation frequency f used by the SSVEP-BCI system₁，f₂,…，f_KConstructing a reference signal R corresponding to each frequency_f1,R_f2,…,R_fk；

Step 2: respectively calculating the synchronous index S between the multi-lead brain electrical signal and each reference signal₁,S₂,…,S_K；

And step 3: and finding out the reference signal with the maximum synchronous index with the electroencephalogram signal, and outputting the frequency corresponding to the reference signal as the identified frequency.

Further, the configuration in step 1 corresponds to the frequency f_iThe reference signal of (c) can be calculated as follows:

F_sfor the sampling rate, M is the number of samples.

Further, the specific process of calculating the synchronization index in step 2 is as follows:

and (3) setting the multi-lead brain electrical signal matrix as X and the reference signal as Y, calculating a joint correlation matrix of the X and the Y:

C = [\begin{matrix} C 11 & C 12 \\ C 21 & C 22 \end{matrix}]

wherein,

C 11 = \frac{1}{M} {XX}^{T}

C 22 = \frac{1}{M} {YY}^{T}

C 12 = C 21 = \frac{1}{M} {XY}^{T}

the following linear transformation is performed:

U = [\begin{matrix} {C 11}^{- \frac{1}{2}} & 0 \\ 0 & C 2 2^{- \frac{1}{2}} \end{matrix}]

obtaining:

wherein, I_N×NIs an N-dimensional unit square matrix,is 2N_hDimension unit matrix, N is number of electrodes, N_hIs the number of reference signal harmonics.

Decomposing the matrix R into eigenvalues to obtain the eigenvalue lambda of the matrix R₁,λ₂,…,λ_PAnd carrying out standardization:

wherein, P is N +2N_h；

Finally, the synchronization index between the multi-lead electroencephalogram signal and the reference signal can be calculated as:

the invention has the beneficial effects that: the invention provides a frequency identification method based on Multivariate Synchronization Index (MSI). in the method, the synchronization index of two multidimensional signals is used as a classification characteristic to carry out frequency identification on electroencephalogram signals in an SSVEP-BCI system. Compared with the existing main multi-pilot frequency detection method, the method has higher accuracy; and has the optimal performance under the conditions of less electrode derivative and shorter data length. The method of the invention can effectively accelerate the response speed of the SSVEP-BCI system and improve the performance of the system.

Drawings

Fig. 1 is a schematic flow diagram of a Multivariate Synchronization Index (MSI) based frequency identification method.

FIG. 2 is a schematic diagram showing the comparison result of the simulation experiment between the method of the present invention and the two conventional methods.

FIG. 3 is a schematic diagram showing the comparison result of the real electroencephalogram experiment between the method of the present invention and the two existing methods.

Detailed Description

The invention is further described with reference to the following figures and specific embodiments.

There are many methods for calculating signal synchronism, and the SSVEP-BCI system has a high requirement on the operation efficiency of the recognition algorithm (the algorithm must give the recognition result of the current electroencephalogram signal in less than 1 second). Therefore, in a Multivariate Synchronization Index (MSI) based frequency identification framework, an efficient frequency identification method is given as follows.

Suppose the EEG signal is X (NxM dimensional matrix) and the reference signal is Y (2N)_hX M dimensional matrix). Here, N is the number of electrodes, M is the number of samples, N_hIs the number of reference signal harmonics. Not in general, X and Y have been normalized to have zero mean unit variance. The implementation process of the frequency detection method based on the multivariate synchronous index is discussed in detail below:

first, a joint correlation matrix of X and Y is calculated

C = [\begin{matrix} C 11 & C 12 \\ C 21 & C 22 \end{matrix}] - - - (1)

Wherein,

C 11 = \frac{1}{M} {XX}^{T} - - - (2)

C 22 = \frac{1}{M} {YY}^{T} - - - (3)

C 12 = C 21 = \frac{1}{M} {XY}^{T} - - - (4)

c contains X, Y autocorrelation and X and Y cross-correlations, and in order to reduce the effect of autocorrelation on the synchronization index, a linear transformation is performed as follows:

U = [\begin{matrix} {C 11}^{- \frac{1}{2}} & 0 \\ 0 & C 2 2^{- \frac{1}{2}} \end{matrix}] - - - (5)

then the following results are obtained:

I_N×Nis an N-dimensional unit square matrix,

is 2N_hDimension unit square matrix.

Decomposing the matrix R into eigenvalues to obtain the eigenvalue lambda of the matrix R₁,λ₂,…,λ_PAnd is standardized

Here P = N +2N_h。

Finally, the synchronization index between the electroencephalogram signal and the reference signal can be calculated as:

suppose that the SSVEP-BCI system has K stimulation frequencies f₁，f₂,…，f_KThen corresponds to the frequency f_iThe reference signal of (c) can be calculated as follows:

F_sis the sampling rate.

According to (1) - (9), the synchronization indexes of all reference signals and electroencephalogram signals can be calculated, and then K synchronization indexes S are obtained₁,S₂,…,S_KThe final frequency identification is performed by the following formula:

namely, the frequency corresponding to the current electroencephalogram signal is the frequency of the reference signal with the maximum synchronization index with the electroencephalogram signal.

In order to more specifically explain the frequency identification method of the SSVEP-BCI system, the invention is further explained with reference to FIG. 1.

As shown in FIG. 1, the multi-lead brain electrical signal X is respectively associated with K reference signals R_f1,R_f2,…,R_fkAs input to the method of the invention, K synchronization indexes S are obtained₁,S₂,…,S_KThen, the maximum value of the K synchronization indexes is obtained. From this maximum value, the corresponding reference signal is found, the frequency used for the reference signal being the output result of the inventive method.

In order to verify the feasibility and the effect of the method, 3 groups of frequencies are adopted for simulation verification, and meanwhile, the method is compared with the existing minimum energy Method (MEC) based detection method and the typical correlation analysis (CCA) based method. The frequencies used were 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；

4 sinusoidal signals at each frequency are generated to simulate 4 electroencephalogram signals, the length of the signals is 10s, and the sampling rate is 250 Hz. Adding Gaussian white noise to each lead signal according to a certain signal-to-noise ratio to simulate a real electroencephalogram signal polluted by noise; then, carrying out frequency identification on the signals under each group of frequencies to obtain identification accuracy, wherein the length of the signals for frequency identification is 1 s; and repeating the operation for 50 times for each group of frequency, and taking the average identification accuracy of 50 results as an algorithm performance evaluation index under the signal-to-noise ratio ranging from-7 db to-20 db.

The signal-to-noise ratio is defined as follows:

<math> <mrow> <mi>SNR</mi> <mo>=</mo> <mn>10</mn> <mi>log</mi> <mfrac> <msub> <mi>P</mi> <mi>signal</mi> </msub> <msub> <mi>P</mi> <mi>noise</mi> </msub> </mfrac> <mo>=</mo> <mn>10</mn> <mi>log</mi> <mfrac> <mrow> <mo>(</mo> <mi>A</mi> <mo>/</mo> <msqrt> <mn>2</mn> </msqrt> <mo>)</mo> </mrow> <msup> <mi>σ</mi> <mn>2</mn> </msup> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>12</mn> <mo>)</mo> </mrow> </mrow> </math>

P_signalis the energy of the signal, P_noiseFor noise energy, A sine signal amplitude, σ²Is the noise variance.

The specific simulation results are shown in fig. 2, wherein (a) the frequencies used are 27Hz,29Hz,31Hz,33Hz,35Hz,37Hz,39Hz,41Hz and43 Hz; (b) the frequencies used are 8Hz,9Hz,10Hz,11Hz,12Hz,13Hz,14Hz,15 Hz; (c) the frequencies used were 6.7Hz,7.5Hz,8.6Hz,10Hz,12Hz,15 Hz.

Denotes significant difference in MSI and CCA + denotes significant difference in MSI and MEC under this condition, with p <0.05 using paired T test.

From the simulation results, the results of the inventive method are the best. For the first group of high-frequency sets and the second group of low-frequency sets, when the signal-to-noise ratio is greater than-12 db, the method provided by the invention has a significant difference compared with the two existing methods, and the method provided by the invention has stronger robustness on noise. In the third set of frequencies, all algorithms cannot achieve 100% accuracy because of the frequency components with harmonic relations, but the method of the present invention always remains significantly different from the two existing methods.

In addition, the validity of the algorithm is further verified by adopting the real electroencephalogram signals. In the experiment, an 8-lead electroencephalogram acquisition system with 4 frequencies of 7.5Hz,8.6Hz,10Hz and 12Hz is adopted to acquire the 30s electroencephalogram signals to be tested under each frequency. 11 subjects (21-28 years) were enrolled in the validation trial and the results are shown in FIG. 3. In the figure, (a) 4-lead brain, (b) 6-lead brain, and (c) 8-lead brain. Denotes significant difference in MSI and CCA + denotes significant difference in MSI and MEC under this condition, with p <0.05 using paired T test.

From the results, the method of the invention has better results than the two existing methods, and particularly under the condition that only 4 brain electrical signals are used and the signal length is 1s, the method of the invention has significant difference with the two existing methods. The electrode number is small, and more convenience can be brought to the application of the SSVEP-BCI system. More importantly, the shorter the signal length for frequency identification, the higher the accuracy of the algorithm can reduce the response time of the system and improve the response speed of the system, so that the method has greater potential to improve the performance of the SSVEP-BCI system.

In general, the simulation experiment and the real experiment result verify the effectiveness and feasibility of the scheme of the invention.

It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.

Claims

1. A frequency identification method based on multivariate synchronous indexes specifically comprises the following steps:

step 1: according to the stimulation frequency f used by the SSVEP-BCI system₁,f₂,…,f_KConstructing a reference signal R corresponding to each frequency_f1,R_f2,…,R_fk；

2. The frequency identification method according to claim 1, wherein the construction in step 1 corresponds to a frequency f_iThe reference signal of (c) can be calculated as follows:

F_sfor the sampling rate, M is the number of samples, N_hIs the number of reference signal harmonics.

3. The frequency identification method according to claim 2, wherein the specific process of calculating the synchronization index in step 2 is as follows:

and (3) setting a multi-lead electroencephalogram signal matrix as X and a reference signal matrix as Y, and calculating a joint correlation matrix of the X and the Y:

C = [\begin{matrix} C 11 & C 12 \\ C 21 & C 22 \end{matrix}]

wherein,

\begin{matrix} C 11 = \frac{1}{M} {XX}^{T} \\ C 22 = \frac{1}{M} {YY}^{T} \\ C 12 = C 21 = \frac{1}{M} {XY}^{T} \end{matrix}

the following linear transformation is performed:

U = [\begin{matrix} {C 11}^{- \frac{1}{2}} & 0 \\ 0 & {C 22}^{- \frac{1}{2}} \end{matrix}]

obtaining:

I_N×Nis an N-dimensional square matrix and is characterized in that,

is 2N_hDimensional matrix, N is number of electrodes;

wherein, P = N +2N_h；

Finally, the synchronous index between the multi-lead electroencephalogram signal and the reference signal can be obtained: