CN108470182B - Brain-computer interface method for enhancing and identifying asymmetric electroencephalogram characteristics - Google Patents

Abstract

The present invention relates to a brain-computer interface method for asymmetric electroencephalographic feature enhancement and recognition, comprising the following steps: Step 1, establishing a training set X _k , a training sample Y ₁ and a test sample through a brain-computer interface system. The EEG signal module of Y; Step 2, perform frequency domain filtering and down-sampling data processing on the test sample _Y in the EEG signal module; Calculate the spatial projection matrix W; step 4, perform DSP spatial filtering on the training set X _k and test sample Y in the EEG signal module according to the following formula to obtain

and W ^T Y eigenvectors; step five, according to

Perform CCA spatial filtering with the ^WTY eigenvectors to construct projection matrices _Uk and _Vk ; obtain the eigenvectors by

W ^T Y, projection matrix U _k and V _k carry out template matching to generate eigenvector p _l according to the following formula; adopt different classifier models to identify the feature vector p _l and output; this method improves the signal-to-noise ratio of the EEG signal itself to improve Classification and recognition efficiency of signal features.

Description

A brain-computer interface method for asymmetric EEG feature enhancement and recognition

technical field

The invention relates to the technical field of brain-computer interface systems, in particular to a brain-computer interface method for asymmetric electroencephalographic feature enhancement and recognition.

Background technique

Brain-Computer Interface (BCI) is a system that directly converts the activities of the central nervous system into artificial output, which can replace, repair, enhance, supplement or improve the normal output of the central nervous system, thereby improving the central nervous system. Interaction with the internal and external environment. By collecting and analyzing the EEG signals of subjects under different stimuli, certain engineering techniques are used to establish communication and control channels between the human brain and computers or other electronic devices. BCI technology realizes a new way of information interaction and control, which can provide a way for disabled people, especially those patients with impaired basic limb motor function but normal thinking, to communicate and control information with the outside world, so that they do not need to use language Or body movements can communicate with the outside world or manipulate external equipment. To this end, BCI technology is also getting more and more attention. P300-speller based on P300 characteristics in Event-Related Potential (ERP) and SSVEP-BCI based on Steady-State Visual Evoked Potential (SSVEP) are widely used visual stimulation-evoked brain - The machine interface system, the related technology of which has been developed relatively stable and mature.

For real-time data acquisition systems, in order to eliminate interference signals, it is usually necessary to perform digital filtering on the collected data. Traditional filtering methods usually filter out specific frequency bands, such as: low-pass filtering, high-pass filtering, band-pass filtering, notch, etc. . The EEG signal has the characteristics of nonlinearity and non-stationarity. In the research of the brain-computer interface system, how to process and analyze the collected EEG signal, extract the weak EEG signal characteristics from the complicated background EEG, and analyze it? Classification and identification of different features is a key factor in determining the performance of the BCI system. Due to the frequency characteristics of EEG signals, filtering methods are often used in the processing and analysis of EEG signals. Usually, the filter frequency band is adjusted according to different EEG characteristics. After filtering, the traditional EEG signal classification and identification methods include Linear Discriminant Analysis (LDA), Common Spatial Pattern (CSP), Support Vector Machine (SVM), Canonical Correlation Analysis ( Canonical Correlation Analysis, CCA) and other methods. These methods all contain the idea of spatial filtering, that is, selecting one or several classification planes in a high-dimensional space, and using its vector as a spatial filter to spatially filter the signal. Classification. The canonical correlation analysis algorithm is currently widely used in the SSVEP-BCI system, and some studies have further improved the algorithm, that is, the template matching principle is used in the process of EEG information processing to introduce the subject's own signal, which improves the system's performance. The identification accuracy rate and information transmission rate have laid a strong foundation for the further transformation of BCI technology into application results.

SUMMARY OF THE INVENTION

The object of the present invention is to overcome the defects existing in the above-mentioned background technology, and to provide a brain-computer interface method for asymmetric EEG feature enhancement and recognition, which is a feature classification method combining discriminant pattern spatial filtering and template matching principles, On the basis of the existing template matching CCA classification strategy, the DSP spatial filtering method is introduced, and different decoding templates are constructed according to the coding strategies of different stimulation paradigms to improve the signal-to-noise ratio of the EEG signal itself, thereby improving the classification and recognition efficiency of signal features. .

The present invention adopts the following technical solutions to be implemented:

A brain-computer interface method for asymmetric EEG feature enhancement and recognition, comprising the following steps:

Step 1: Establish an EEG data set including a training set X _k , a training sample Y _l and a test sample Y through the brain-computer interface system

Step 2, performing frequency domain filtering and downsampling data processing on the test sample Y in the EEG signal data set;

Step 3, based on Fisher's linear discriminant criterion, calculate the training set X _k in the EEG signal module to obtain the spatial projection matrix W;

和W^TY特征向量；Step 4: Perform DSP spatial filtering on the training set X _k and the test sample Y in the EEG signal data set according to the following formula to obtain:

and W ^T Y eigenvectors;

S _B =∑ ₁₁ +∑ ₂₂ -∑ ₁₂ -∑ ₂₁

S _w =σ ₁ ² +σ ₂ ²

和W^TY特征向量采用如下公式进行CCA空间滤波构建投影矩阵U_k和V_k；Step five, according to

and W ^T Y eigenvectors adopt the following formulas to carry out CCA spatial filtering to construct projection matrices U _k and V _k ;

W^TY、投影矩阵U_k和V_k按照如下公式进行模板匹配生成特征向量ρ_l；Step 6, by obtaining the feature vector

W ^T Y, projection matrix U _k and V _k perform template matching according to the following formula to generate eigenvector ρ _l ;

Step 7: Use different classifier models to identify and output the feature vector _ρl .

k表示两类特征，即k＝1，2；所述训练样本

所述测试样本

其中Nc表示采集脑电的通道数，N_t表示截取信号长度，N_s表示训练集样本个数。the training set

k represents two types of features, that is, k=1, 2; the training samples

the test sample

Among them, Nc represents the number of channels for collecting EEG, N _t represents the length of the intercepted signal, and N _s represents the number of training set samples.

Compared with the prior art, the present invention has the advantages:

1. The present invention is used for the classification and recognition of asymmetric EEG features, which can effectively improve the signal-to-noise ratio of the recognition signal and improve the classification accuracy.

2. In the brain-computer interface system for asymmetric EEG feature control of the present invention, the average classification accuracy of asymmetric EEG features is increased by 17.88% compared with the traditional classification method, which proves that this method can further improve the brain-computer interface. Interface technology to promote the transformation of this technology to application results; the application range is wide.

3. The present invention has been applied to a brain-computer interface system based on asymmetric EEG feature control, designed and implemented BCI-speller offline and online brain-computer interface system experiments with instruction set 32; further research can obtain a perfect brain-computer interface system. The machine interface system is expected to obtain considerable social and economic benefits.

Description of drawings

FIG. 1 is a flowchart of a brain-computer interface method for asymmetric EEG feature enhancement and recognition according to the present invention.

FIG. 2 is a schematic structural diagram of a brain-computer interface system including 32 instruction sets according to the present invention.

Detailed ways

The present invention will be further described below through specific embodiments and accompanying drawings. The embodiments of the present invention are for better understanding of the present invention by those skilled in the art, and do not limit the present invention.

As shown in FIG. 1 , the present invention provides a brain-computer interface method for asymmetric EEG feature enhancement and recognition;

Step 1 101: Establish an EEG data set including a training set X _k , a training sample Y ₁ and a test sample Y through the brain-computer interface system

为训练集，k表示两类特征，即k＝1，2，

为两类训练样本，l＝1，2，

为测试样本，其中N_c表示采集脑电的通道数，N_t表示截取信号长度，N_s表示训练集样本个数。训练集和测试集都在时间尺度上进行了零均值处理，即每一个时间点的数值s_t都减去时间窗[t₁，t₂]内的时间平均值

如公式(1)所示：assumed

For the training set, k represents two types of features, that is, k=1, 2,

For two types of training samples, l=1, 2,

is the test sample, where N _c represents the number of EEG channels, N _t represents the length of the intercepted signal, and N _s represents the number of samples in the training set. Both the training set and the test set are processed with zero mean on the time scale, that is, the value s _t at each time point is subtracted from the time mean within the time window [t ₁ , t ₂ ]

As shown in formula (1):

表示。两类模板之间的协方差矩阵

表示为：The average value of all samples in the training set is obtained to obtain the template signal of class k, which is given by

express. Covariance matrix between two types of templates

Expressed as:

The variances of the two types of signals X ₁ and X ₂ are expressed as:

测试样本进行频域滤波和降采样数据处理；Step 2 102, select from the EEG signal data set

The test samples are subjected to frequency domain filtering and downsampling data processing;

Step 3 103, based on Fisher's linear discriminant criterion, calculate the training set X _k in the EEG signal module to obtain the spatial projection matrix W;

和W^TY特征向量Step 4 104: Perform DSP spatial filtering on the training set X _k and the test sample Y in the EEG signal data set according to the following formula to obtain:

and W ^T Y eigenvectors

Based on Fisher's linear criterion, the DSP algorithm obtains a projection matrix W so that the two types of eigensignals have greater separability after being projected. This matrix W can be regarded as a spatial filter, and its solution method is:

S _B =∑ ₁₁ +∑ ₂₂ -∑ ₁₂ -∑ ₂₁ (6)

S _w =σ ₁ ² +σ ₂ ² (7)

和W^TY之间的相关性，CCA空间滤波器U_k和V_k由下述公式(8)计算得到。where λ _i is the eigenvector of the i-th column in the matrix W, and N _W represents the number of selected spatial filters. The common mode signal between the two types of signals can be filtered out by W spatial filtering, and the CCA algorithm can be used to calculate the DSP spatial filtering by constructing two projection matrices U _k and V _k .

and W ^T Y, the CCA spatial filters U _k and V _k are calculated by the following equation (8).

和W^TY特征向量采用如下公式进行CCA空间滤波构建投影矩阵U_k和V_k Step five 105, according to

and W ^T Y eigenvectors use the following formulas for CCA spatial filtering to construct projection matrices U _k and V _k

表示数学期望。典型相关分析是衡量两个多维变量之间的线性相关关系的统计分析方法。区别于在线性回归中利用直线来拟合样本点，CCA是将多维特征向量都看作一个整体，利用数学方法寻求一组最优解，使得两个整体之间有最大关联的权重，即令公式(8)计算得到的数值最大，这就是典型相关分析的目的。in,

Indicates mathematical expectations. Canonical correlation analysis is a statistical analysis method that measures the linear correlation between two multidimensional variables. Different from the use of straight lines to fit sample points in linear regression, CCA treats multi-dimensional eigenvectors as a whole, and uses mathematical methods to find a set of optimal solutions, so that the two wholes have the greatest correlation weight, that is, let the formula (8) The calculated value is the largest, which is the purpose of the canonical correlation analysis.

W^TY、投影矩阵_Uk和V_k按照如下公式进行模板匹配生成特征向量ρ_l；Step 6 106, by obtaining the feature vector

W ^T Y, projection matrix _U k and V _k perform template matching according to the following formula to generate eigenvector ρ _l ;

In the template matching process, a template is constructed from the training set data, and the template construction can also be adjusted according to the different stimulation methods. Taking the classification of asymmetric EEG characteristic signals as an example, the vector ρ _k shown in formula (9) represents the similarity between the training template and the training sample signal l.

之间的相关性越大。连接ρ_1·，l和ρ_2·，l即可得到训练模板和训练样本经特征提取后得到的特征向量ρ_M，由公式(10)所示：where corr(*) represents the Pearson correlation coefficient and dist(*) represents the Euclidean distance. If ρ _k1 , ρ _k2 , ρ _k3 , ρ _k4 and ρ _k5 are larger, it means that Y _l and

the greater the correlation between them. Connecting ρ _{1 , l} and ρ _{2 , l} can obtain the feature vector ρ _M obtained from the training template and the training sample after feature extraction, which is shown by formula (10):

ρ _l =(ρ _1·l , ρ _2.l ), l=1, 2 (10)

Step 7 107: Use different classifier models to identify and output the feature vector _ρ1 .

Different classifier models of different pattern recognition algorithms such as Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) are established according to the feature vector ρ _l , and the test sample Y is sent to the input after preprocessing and feature extraction. The classifier performs pattern recognition, and then predicts the category of the sample and outputs the result, as shown in Figure 1.

FIG. 2 is a schematic diagram showing the structure of a brain-computer interface system including 32 instruction sets applied by the algorithm of the present invention. The system includes a liquid crystal display stimulation interface, an EEG acquisition system such as EEG electrodes and EEG amplifiers, and a computer processing platform. The system uses the visual stimulation paradigm to induce two types of asymmetric EEG features, adopts the EEG digital acquisition system produced by NeuroScan Company to collect EEG signals, amplifies and filters the signals through the EEG amplifier, and then inputs them into the computer. EEG features are classified, and finally the EEG signals are decoded and converted into BCI instructions for output. Stimulus presentation and data processing and analysis were completed based on the Matlab platform.

The signal-to-noise rate (SNR) of the two types of asymmetric EEG characteristic signals are calculated to be -17.98dB and -14.90dB, respectively. SNR is defined as the ratio of signal energy to noise energy. The calculation formula is:

where AMP _i represents the average amplitude of the signal in the i-th trial time window, and N represents the number of trials.

The BCI system to which the algorithm of the present invention is applied was tested on 12 subjects. The experimental results show that after applying the algorithm of the present invention, the average classification accuracy rate of the 12 subjects is increased by 17.88%, which is significantly improved (paired T test results are: t ₁₁ =-8.91, p < 0.01), and the two types of characteristics The signal-to-noise ratio of the signal is improved to -9.71dB and -8.68dB respectively after DSP spatial filtering.

It should be understood that the embodiments and examples discussed here are only for illustration, and for those skilled in the art, improvements or changes may be made, and all these improvements and changes should fall within the protection scope of the appended claims of the present invention.

Claims (2)

1. a brain-computer interface method for asymmetric EEG feature enhancement and identification, is characterized in that, comprises the steps: Step 1, establishing an EEG signal data set including a training set X _k , a training sample Y ₁ and a test sample Y through the brain-computer interface system; Step 2, performing frequency domain filtering and downsampling data processing on the test sample Y in the EEG signal data set; Step 3, based on Fisher's linear discriminant criterion, calculate the training set X _k in the EEG signal module to obtain the spatial projection matrix W; Step 4: Perform DSP spatial filtering on the training set X _k and the test sample Y in the EEG signal data set according to the following formula to obtain:

and W ^T Y eigenvectors;

S _B =∑ ₁₁ +∑ ₂₂ -∑ ₁₂ -∑ ₂₁ S _w =σ ₁ ² +σ ₂ ² Step five, according to

and W ^T Y eigenvectors adopt the following formulas to carry out CCA spatial filtering to construct projection matrices U _k and V _k ;

Step 6, by obtaining the feature vector

W ^T Y, projection matrix U _k and V _k perform template matching according to the following formula to generate eigenvector ρ _l ;

Where: corr(*) represents the Pearson correlation coefficient, dist(*) represents the Euclidean distance; Step 7: Use different classifier models to identify and output the feature vector _ρl . 2. A brain-computer interface method for asymmetric EEG feature enhancement and recognition according to claim 1, characterized in that: the training set

k represents two types of features, that is, k=1, 2; the training samples

the test sample

Among them, N _c represents the number of channels for collecting EEG, N _t represents the length of the intercepted signal, and N _s represents the number of training set samples.

Images