A method and device for electroencephalogram signal identity recognition based on spatial-frequency domain features

Abstract

The application relates to a spatial-frequency domain feature-based electroencephalogram signal identity recognition method, which comprises the following steps: collecting electroencephalogram signals; pre-processing the collected electroencephalogram signals; extracting the co-space feature data of the pre-processed electroencephalogram signals; performing fractional Fourier transform on the co-space feature data to obtain spatial-frequency domain feature data; performing identity recognition classification on the spatial-frequency domain feature data; and outputting a recognition result. The application extracts the spatial domain feature of the electroencephalogram signal by using the co-space mode, then converts the spatial domain feature into the spatial-frequency domain feature by using the fractional Fourier transform, and realizes the identity recognition of different individuals by using the support vector machine for classification, so that the recognition rate can reach more than 85%.

Description

Technical Field

[0001] This invention belongs to the field of EEG signal recognition technology, and specifically relates to a method and device for EEG signal identity recognition based on spatial-frequency domain features. Background Technology

[0002] Biometric technology combines computers with advanced technologies such as optics, acoustics, biosensors, and biostatistics to identify individuals using inherent physiological characteristics (e.g., fingerprints, facial images, irises) and behavioral features (e.g., handwriting, voice, gait). Electroencephalograms (EEGs) are bioelectrical phenomena that reflect brain activity. Because no one can imitate another's thoughts, and because EEG signals differ between a person under duress or stress and a calm state, using EEGs for identification offers extremely high security and is an important supplement to traditional biometric identification systems.

[0003] Although many methods have been applied to the identification of individuals using EEG signals, such as short-time Fourier transform, wavelet transform, time-frequency analysis, deep learning, principal component analysis, and independent component analysis, it is difficult to achieve accurate identification of different individuals by using only a single method. Summary of the Invention

[0004] The purpose of this invention is to provide a method and device for identifying EEG signals based on spatial-frequency domain features, which can better separate and represent EEG signal components of different frequencies, thereby improving the accuracy of classification and identification;

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for identifying individuals using electroencephalogram (EEG) signals based on spatial-frequency domain features includes the following steps:

[0007] Collecting electroencephalogram (EEG) signals;

[0008] The collected EEG signals are preprocessed;

[0009] Extract cospatial feature data from preprocessed EEG signals;

[0010] Fractional Fourier transform is performed on the co-space feature data to obtain space-frequency domain feature data;

[0011] Perform identity recognition and classification on spatial-frequency domain feature data, and output the recognition results.

[0012] Preferably, the acquisition of EEG signals further includes the following step: acquiring EEG signals induced by emotional speech such as sadness, anger, happiness, and surprise.

[0013] Preferably, the EEG signal preprocessing includes the following steps: downsampling the EEG signal at a downsampling rate of 1000Hz; filtering the EEG signal with a filter at a rate of 0.1 to 30Hz; and removing artifacts from the EEG signal.

[0014] Preferably, the eigenvalue matrix of the co-space feature data

[0015] Z NxT =W NxN X NxT

[0016] Where X is the spatiotemporal signal matrix, N is the number of EEG channels, T is the number of sample points collected by each channel, and W is the projection matrix.

[0017] Preferably, the fractional Fourier transform is expressed as:

[0018]

[0019] Among them, K P (u,t)=A α exp[jπ(u 2 cotα-2utcscα+t 2 [cotα)] is the kernel function, α = pπ / 2, where p is 0.1 to 0.2.

[0020] Preferably, p = 0.2.

[0021] Preferably, the identity recognition classification includes the following steps: classifying spatial-frequency domain feature data using a support vector machine.

[0022] According to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, performs the steps of the above-described method for identifying EEG signals based on spatial-frequency domain features.

[0023] According to another aspect of the present invention, a brainwave signal identification device based on spatial-frequency domain features is also provided, comprising:

[0024] Memory, used to store software applications.

[0025] A processor is used to execute the software application, wherein each program of the software application correspondingly performs the steps in the above-described method for EEG signal identification based on spatial-frequency domain features.

[0026] This invention extracts spatial domain features of EEG signals using co-space patterns, then transforms these spatial domain features into spatial-frequency domain features using fractional Fourier transform, and finally uses support vector machines for classification to achieve identification of different individuals, with a recognition rate of over 85%. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the process of the present invention;

[0028] Figure 2 This is the EEG channel diagram selected in this embodiment.

[0029] Figure 3 This is a flowchart illustrating the shared space mode in this embodiment.

[0030] Figure 4 This is a schematic diagram of the identity recognition results when p is of order 0.2. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1 As shown, the present invention provides a method for identity recognition based on electroencephalogram (EEG) signals with spatial-frequency domain features, comprising the following steps:

[0033] S1. Acquire electroencephalogram (EEG) signals;

[0034] Specifically, in step S1, the acquisition of EEG signals further includes the following steps:

[0035] S11. Collect EEG signals induced by emotional speech such as sadness, anger, happiness and surprise.

[0036] In this embodiment, as Figure 2 As shown, due to the similarity between multichannel EEG and strong signal interference, it cannot be used during the acquisition process. Therefore, 16 channels were selected from six brain regions: the prefrontal (left / right), central (left / right), and parietal (left / right) regions, including Fz, FC3, FC1, FCz, FC4, FC2, C3, C1, CZ, C4, C2, CP3, CP1, CPZ, CP2, and CP4.

[0037] EEG data from 64 channels were acquired using a Dutch BIOSEMI 64-channel EEG acquisition system. Electrode caps were placed using the international 10-20 system. Data were collected from 20 subjects at a sampling rate of 1024 Hz, and electrode resistance was kept below 5 kΩ.

[0038] The audience sat in comfortable armchairs and were instructed to avoid making unnecessary movements. Four categories of EEG signals were collected from four emotional speech stimuli: sadness, anger, happiness, and surprise. Each emotional speech was played twice, once for emotional attention and once for semantic attention. When a speech was played a second time, the audience was asked to remember what they heard and repeat it. All of these tasks were designed to make the audience's attention more focused.

[0039] S2. Preprocess the collected EEG signals;

[0040] Specifically, in step S2, the preprocessing of the electroencephalogram (EEG) signal includes the following steps:

[0041] S21. Downsample the EEG signal at a rate of 1000Hz; apply high-pass and low-pass filters to the EEG signal; remove artifacts from the EEG signal.

[0042] Specifically, the downsampled EEG signal was subjected to a Butterworth filter for a 0.1Hz high-pass filter and a 30Hz low-pass filter.

[0043] Specifically, artifacts, including electromyography and electrooculography signals, are removed.

[0044] In this embodiment, the preprocessed EEG signal is segmented using a 500ms moving window (repeated 50%), and the 3000ms data is divided into 5 1000ms segments (1-1000ms, 500-1500ms, 1000-2000ms, ..., 2000-3000ms) to enhance the dataset.

[0045] S3. Extract the cospatial feature data of the preprocessed EEG signals;

[0046] In this embodiment, as Figure 3 As shown, the process of extracting cospatial feature data from EEG signals is as follows:

[0047] 1) After preprocessing, the spatiotemporal signal matrices X of the multi-channel induced signals in four categories are obtained. The dimensions of each matrix are N×T, where N is the number of EEG channels and T is the number of sample points collected for each channel.

[0048] All samples are represented by an N×T matrix X:

[0049]

[0050] The dimension of matrix X is 2400×16×5000, and the dimension of each trial is 16×5000, where 16 represents the number of EEG channels and 5000 represents the number of sample points collected by each channel.

[0051] 2) Normalize the matrix to obtain the covariance matrix for each category: R = XX T / trace(XX T )

[0052] Where T represents the transpose operator, trace(XX) T ) represents the sum of the diagonal elements of the matrix, and the dimension of matrix R is 16×16;

[0053] 3) Use them separately We have the mean covariance matrices for four types, and then we calculate the mixture covariance matrix for each type:

[0054] 4) Perform eigenvalue decomposition on the mixture covariance matrix R: R = UλU T , where U is the eigenvector matrix of matrix R, and λ is the diagonal matrix composed of the corresponding eigenvalues;

[0055] 5) Arrange the obtained eigenvalues ​​in descending order to obtain the whitening matrix: The dimension of matrix P is 16×16;

[0056] 6) Perform the following transformations on the covariance matrices R1, R2, R3, and R4:

[0057] S1=PR1P T S2 = PR2P T S3 = PR3P T S4 = PR4P T ;

[0058] 7) By performing an approximate joint diagonalization on matrices S1, S2, S3, and S4, we can obtain an orthogonal matrix and an approximate diagonal matrix:

[0059]

[0060] And it satisfies B1=B2=B3=B4=B、λ1+λ2+λ3+λ4=I;

[0061] 8) Use matrix B for classification. The projection matrix is ​​as follows: W = (B T P) T ;

[0062] 9) The projection matrix is ​​represented by an N×N matrix W: the dimension of matrix W is 16×16;

[0063]

[0064] The EEG data X is projected using a projection matrix to obtain the eigenvalue matrix Z. NxT =W NxN XNxT The dimension of matrix Z is 2400×16.

[0065] 10) Finally, you can choose Z. N×T The first m rows and the last m rows (2m < M) are used as features of EEG data.

[0066] S4. Perform a fractional Fourier transform on the cospace feature data to obtain the space-frequency domain feature data;

[0067] Specifically, in step S4, the fractional Fourier transform is expressed as:

[0068]

[0069] Among them, K P (u,t)=A α exp[jπ(u 2 cotα-2utcscα+t 2 [cotα)] is the kernel function, α = pπ / 2, where p is 0.1 to 0.2; preferably, p = 0.2.

[0070] In this embodiment, after the p-order fractional Fourier transform, the dimension of the feature data Z becomes 16×2400, and the dimension after transpose becomes 2400×16.

[0071] S5. Perform identity recognition and classification on the spatial-frequency domain feature data and output the recognition results.

[0072] Specifically, in step S5, the identity recognition and classification includes the following steps:

[0073] Support vector machines are used to classify spatial-frequency domain feature data;

[0074] In this embodiment, the feature vectors of 15 subjects are randomly selected from 20 subjects as training samples, and category labels are added to the last column. The dimension of the final training sample matrix Z is 2400×17.

[0075] Five-fold cross-validation was used in the training with 15 subjects, yielding good recognition results. At p=0.2, the average recognition rate was 85.7%. The recognition results for different individuals are as follows: Figure 4 As shown.

[0076] According to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, performs the steps of the above-described method for identifying EEG signals based on spatial-frequency domain features.

[0077] According to another aspect of the present invention, a brainwave signal identification device based on spatial-frequency domain features is also provided, comprising:

[0078] Memory, used to store software applications.

[0079] A processor is used to execute the software application, wherein each program of the software application correspondingly performs the steps in the above-described method for EEG signal identification based on spatial-frequency domain features.

Claims

1. A method for identity recognition based on electroencephalogram (EEG) signals with spatial-frequency domain features, characterized in that, Includes the following steps: Collecting electroencephalogram (EEG) signals; The collected EEG signals are preprocessed; Extract cospatial feature data from preprocessed EEG signals; Fractional Fourier transform is performed on the co-space feature data to obtain space-frequency domain feature data; Perform identity recognition and classification on spatial-frequency domain feature data, and output the recognition results; Wherein, the eigenvalue matrix of the shared spatial feature data ， in, It is a spatiotemporal signal matrix. Number of brainwave channels, The number of sample points collected for each channel. This is the projection matrix.

2. The EEG signal identification method based on spatial-frequency domain features according to claim 1, characterized in that, The acquisition of electroencephalogram (EEG) signals also includes the following steps: Collect EEG signals induced by emotional speech such as sadness, anger, happiness, and surprise.

3. The EEG signal identification method based on spatial-frequency domain features according to claim 1, characterized in that, The preprocessing of the electroencephalogram (EEG) signals includes the following steps: The EEG signal was downsampled at a rate of 1000 Hz; a filter was used to filter the EEG signal from 0.1 to 30 Hz; artifacts in the EEG signal were removed.

4. The EEG signal identification method based on spatial-frequency domain features according to claim 1, characterized in that, The fractional Fourier transform is expressed as: ， ， in, For kernel function, , p is 0.1~0.

2.

5. The EEG signal identification method based on spatial-frequency domain features according to claim 4, characterized in that, The value of p is 0.

2.

6. The EEG signal identification method based on spatial-frequency domain features according to claim 1, characterized in that, The identity recognition and classification includes the following steps: Support vector machines are used to classify spatial-frequency domain feature data.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it performs the steps of the EEG signal identification method based on spatial-frequency domain features as described in any one of claims 1 to 6.

8. A brainwave signal identification device based on spatial-frequency domain features, characterized in that, include: Memory, used to store software applications. A processor for executing the software application, wherein each program of the software application correspondingly performs the steps of the EEG signal identification method based on spatial-frequency domain features as described in any one of claims 1 to 6.

Images