A method for classifying EEG signals based on two-dimensional local mean decomposition

Abstract

The present invention discloses a method for classifying EEG signals based on two-dimensional local mean decomposition, comprising the following steps: collecting EEG signals of multiple channels obtained by an EEG acquisition device, and preprocessing the EEG signals; selecting channel signals representing left and right brain activities to construct two-dimensional signals, and obtaining product function components through two-dimensional local mean decomposition; performing Hilbert-Huang transform on the product function components of the two channels respectively to obtain an edge Hilbert-Huang spectrum; calculating the average band power of a specified frequency band in the edge Hilbert-Huang spectrum as a characteristic value; inputting the characteristic value into a classifier to classify EEG signals under different motor imagery tasks. Using the method of the present invention, feature extraction can be performed on EEG signals of fewer electrodes, and classification of EEG signals during different motor imagery tasks can be achieved, which is helpful to promote the research of online real-time brain-computer interaction control systems based on motor imagery.

Description

Technical Field

[0001] The invention belongs to the field of electroencephalogram signal processing, and in particular relates to an electroencephalogram signal classification method based on two-dimensional local mean decomposition. Background Art

[0002] Related studies have shown that imagining limb movements is similar to actually performing the movements, and both produce electrophysiological responses in the corresponding motor cortex areas of the brain, inducing action potentials. When preparing or planning specific limb movements, the activity state of a large number of central nervous cells in the cortical area will change, which will be reflected in the EEG signals, and the frequency components of certain specific frequency bands will be synchronously enhanced (event-related synchronization, ERS) or synchronously weakened (event-related desynchronization, ERD).

[0003] EEG signals play a very important role in many fields such as neuroscience, psychology, and biomedicine. At the same time, the analysis of EEG signals also plays an important role in the development of these disciplines. However, because the original EEG signals have large noise, low signal-to-noise ratio, and poor reliability, it is difficult to extract effective feature values ​​for analysis. Therefore, how to effectively and reliably extract EEG signal features has become a major core difficulty in the application of EEG signal analysis.

[0004] The most commonly used two-dimensional EEG signal decomposition method is the two-dimensional empirical mode decomposition method (BEMD). Through this method, the source signal is converted into an intrinsic mode function (IMF) containing the different time-frequency scale characteristics of the original EEG data signal, which can reflect the characteristics of EEG data at different time-frequency scales. It can decompose multi-channel data at the same time. Compared with the single-dimensional EMD algorithm, it can avoid the problem of mismatch in the number and frequency of IMFs decomposed from each variable, but it still has many problems such as mode aliasing and low accuracy.

[0005] Based on the above background, finding a method for feature extraction and classification of EEG data with high computational efficiency, high classification accuracy, and suitable for non-stationary, nonlinear and low signal-to-noise ratio is of great significance for EEG data analysis, and has very important practical value for further in-depth research on online real-time brain-computer interaction control systems based on motor imagery EEG. Summary of the invention

[0006] The present invention provides an EEG signal classification method based on two-dimensional local mean decomposition, which only requires fewer electrodes to extract the features of EEG signals and can adaptively decompose the signals into multiple intrinsic mode function signals according to the characteristics of the EEG signals of each subject, thereby solving the problems of mode aliasing and poor reliability of ordinary multidimensional empirical mode decomposition algorithms. The decomposition results are more interpretable, which greatly improves the classification accuracy of EEG signals.

[0007] A method for classifying electroencephalogram signals based on two-dimensional local mean decomposition comprises the following steps:

[0008] Step 1, collecting multi-channel EEG signals through an EEG acquisition device, and preprocessing the EEG signals;

[0009] Step 2, selecting channel signals representing left and right brain activities to construct a two-dimensional signal, and obtaining product function components through two-dimensional local mean decomposition;

[0010] Step 3, performing Hilbert-Huang transform on the product function components of the two channels respectively to obtain an edge Hilbert-Huang spectrum diagram;

[0011] Step 4, calculating the average band power of the specified frequency band in the edge Hilbert-Huang spectrum as the characteristic value;

[0012] Step 5: Input the feature values ​​into the classifier to classify the EEG signals under different motor imagery tasks.

[0013] By using the method of the present invention, the features of EEG data can be extracted to complete the effective classification of signals under different motor imagery tasks.

[0014] In step 1, the EEG signal is preprocessed by performing 0.2-50 Hz bandpass filtering on the EEG signal through a Butterworth filter to obtain a filtered EEG signal.

[0015] The specific process of step 2 is:

[0016] Step 2-1, select the C3 channel signal x1(t) and the C4 channel signal x2(t) representing the left and right brain activities respectively to construct a two-dimensional signal x(t) = [x1(t), x2(t)] as the input signal;

[0017] Step 2-2, taking a set of direction vectors evenly distributed in the two-dimensional space, projecting the input signal along the direction vectors, and obtaining projection signals along different directions;

[0018] Step 2-3, for each projection direction, according to the multi-dimensional signal extreme point solution method, the extreme points of the input signal in each projection direction are extracted, and the smoothed local mean estimation function in each projection direction is calculated. After integration, the overall local mean function of the input signal is obtained. The specific formula is as follows:

[0019]

[0020] Wherein, m(t) represents the overall local mean function of the input signal, represents a smoothed local mean estimation function of the input signal in each projection direction, and M represents the number of direction vectors;

[0021] Step 2-4, respectively extract the extreme values ​​of the two channel signals of the input signal, calculate the local amplitude, and smooth these local amplitudes to obtain the local amplitude functions of the two channels a(t)=[a1(t), a2(t)];

[0022] Step 2-5, calculate the FM signals of the two channels respectively according to the overall local mean function and the local amplitude function. The specific formula is as follows:

[0023]

[0024] Among them, s k (t) represents the frequency modulation signal of the kth channel, h k (t) represents the residual signal component obtained after subtracting the projection of the overall local mean function m(t) on the kth channel from the kth channel of the input signal;

[0025] Step 2-6, if the local amplitude functions of the two channels are not all constantly 1, then save the local amplitude functions, use the obtained two-dimensional signal constructed by the frequency modulation signals of the two channels as the input signal, and repeat steps 2-2 to 2-5 until the local amplitude functions of the two channels are all constantly 1;

[0026] Step 2-7, multiplying all the local amplitude functions saved in step 2-6 to obtain an integrated local amplitude function, and multiplying the integrated local amplitude function with the pure frequency modulation signal obtained in step 2-6 to obtain a product function component;

[0027] Step 2-8, subtract the product function component from the input signal to obtain the output signal; if the output signal is not a monotonic signal, use the output signal as a new input signal and repeat steps 2-2 to 2-7 until the output signal is a monotonic signal.

[0028] The specific process of step 3 is as follows:

[0029] Step 3-1, decompose the input signal into the sum of several product function components and a residual component through step 2. The specific formula is as follows:

[0030]

[0031] Where x(t) is the constructed two-dimensional signal, PF i (t) is the i-th product function component, r(t) is the residual component;

[0032] Step 3-2, perform Hilbert-Huang transform on each product function component. The specific formula is as follows:

[0033]

[0034] Step 3-3, construct the analytical signal of each product function component, and calculate the amplitude function and phase function of the analytical signal. The specific formula is as follows:

[0035]

[0036] Among them, x A (t) is the analytical signal, a i (t) and θ i (t) represent the amplitude function and phase function of the analytical signal respectively;

[0037] Step 3-4, deriving the phase function of the analytical signal to obtain the frequency function, and obtaining the Hilbert-Huang spectrum H(ω,t) of the analytical signal based on the frequency function and the amplitude function;

[0038] Step 3-5, integrate the Hilbert-Huang spectrum of the analytical signal with respect to time to obtain the edge Hilbert spectrum. The specific formula is as follows:

[0039] h(ω)=∫0 T H(ω,t)dt

[0040] In step 4, the average band power of the designated frequency band is the average band power of the 8-12 Hz frequency band, the 18-26 Hz frequency band, and the 0-50 Hz frequency band, which correspond to the Mu rhythm, the Beta rhythm, and the total frequency band, respectively.

[0041] In step 5, the classifier adopts a support vector machine classifier based on radial basis function, and the support vector machine classifier determines the classifier model parameters through a ten-fold cross-validation during the pre-training process.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. The present invention decomposes data of different frequency bands for feature extraction, which can retain the correlation information between different channels to a greater extent and is more reliable and stable.

[0044] 2. Compared with local mean decomposition, it can process two-dimensional signals and retain the correlation information between the two signal channels.

[0045] 3. Applied to actual EEG data, signal decomposition is more reliable and stable, and feature extraction is more accurate, so the classification accuracy is better than other existing feature value extraction algorithms based on time-frequency analysis.

[0046] 4. The present invention can extract the characteristics of EEG signals of different motor imagery tasks, providing an algorithmic basis for further in-depth research on online real-time brain-computer interaction control systems based on motor imagery EEG. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The overall experimental process of the motor imagery task in the embodiment of the present invention is as follows;

[0048] Figure 2 This is the experimental recording process of a single motor imagery task in an embodiment of the present invention;

[0049] Figure 3 This is a BLMD decomposition result diagram of the first set of experimental data in an embodiment of the present invention;

[0050] Figure 4 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0051] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be pointed out that the embodiments described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0052] The following takes the motor imagery classification problem of BCI Competition 2008—Graz data Set A as an example to illustrate the implementation plan and effect of the EEG signal classification method based on multidimensional local mean decomposition. This example is carried out in the Matlab2019a simulation environment.

[0053] The embodiment uses the data of experimenter 1 in BCI Competition 2008—Graz data Set A. The data of BCI Competition 2008—Graz data Set A consists of 9 healthy subjects (1-9) facing corresponding prompts on a computer screen, opening their eyes, closing their eyes, moving their eyes, and then performing the process of left hand / right hand / feet / tongue movement imagination.

[0054] Each set of data contains 22 channels of EEG signals. The same experiment was conducted on two days, with 288 experiments performed each day, for a total of 576 experimental data. Figure 1 As shown in the figure, each recording session was divided into four parts including 2 minutes of eyes open time, 2 minutes of eyes closed, 1 minute of eye movement and the rest of the motor imagery test. Figure 2 As shown in the figure, each motor imagery trial includes 2s of quiet preparation time, 1s of prompt time, 3s of motor imagery task execution time, and 2s of black screen rest time. The signal sampling frequency is 250Hz, and it is band-pass filtered from 0.05 to 200Hz.

[0055] like Figure 4 As shown, a method for classifying EEG signals based on two-dimensional local mean decomposition includes:

[0056] Step 1: Preprocess the collected EEG signals. Select two channels C3 and C4 from the 24 channels and perform 0.2-50 Hz bandpass filtering through a Butterworth filter. The filtered signal is x(t)=[x1(t),x2(t)]∈R 750×2 ; The total number of sample points is 750, the number of channels is 2, and t=1,2,…,750.

[0057] Step 2: Decompose the two-dimensional signal obtained from each experiment by two-dimensional local mean to obtain the intrinsic mode function signal and trend term. The decomposition results of the first set of experimental data are as follows: Figure 3 shown.

[0058] Step 3: Perform Hilbert-Huang transform on the intrinsic mode function signals of each layer of the two channels respectively to obtain the edge Hilbert-Huang spectrum.

[0059] Step 4, calculating the average band power of the 8-12 Hz frequency band (Mu rhythm), the 18-26 Hz frequency band (Beta rhythm) and the total frequency band (0-50 Hz) as the extracted feature value;

[0060] Step 5: Randomly select 60 groups of the 100 experimental data as training sets, and the remaining 40 groups as test sets. Input the feature vector extracted in step 4 into a support vector machine classifier using radial basis function for classification.

[0061] The classification accuracy of different motor imagery test sets is shown in Table 1, among which the left hand motor imagery and eye movement classification accuracy are the highest. The accuracy of this motor imagery EEG signal classification method can reach 87.5%.

[0062] Table 1 Comparison of BLMD and BEMD classification accuracy

[0063]

[0064] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for classifying electroencephalogram signals based on two-dimensional local mean decomposition, characterized in that: The following steps are involved: Step 1, collecting multi-channel EEG signals through an EEG acquisition device, and preprocessing the EEG signals; Step 2, select the channel signals representing the left and right brain activities to construct a two-dimensional signal, and obtain the product function components through two-dimensional local mean decomposition; the specific process is: Step 2-1, select the C3 channel signal x1(t) and the C4 channel signal x2(t) representing the left and right brain activities respectively to construct a two-dimensional signal x(t) = [x1(t), x2(t)] as the input signal; Step 2-2, taking a set of direction vectors evenly distributed in the two-dimensional space, projecting the input signal along the direction vectors, and obtaining projection signals along different directions; Step 2-3, for each projection direction, according to the multi-dimensional signal extreme point solution method, the extreme points of the input signal in each projection direction are extracted, and the smoothed local mean estimation function in each projection direction is calculated. After integration, the overall local mean function of the input signal is obtained. The specific formula is as follows: Wherein, m(t) represents the overall local mean function of the input signal, represents a smoothed local mean estimation function of the input signal in each projection direction, and M represents the number of direction vectors; Step 2-4, respectively extract the extreme values ​​of the two channel signals of the input signal, calculate the local amplitude, and smooth these local amplitudes to obtain the local amplitude functions of the two channels a(t)=[a1(t), a2(t)]; Step 2-5, calculate the FM signals of the two channels respectively according to the overall local mean function and the local amplitude function. The specific formula is as follows: Among them, s k (t) represents the frequency modulation signal of the kth channel, h k (t) represents the residual signal component obtained after subtracting the projection of the overall local mean function m(t) on the kth channel from the kth channel of the input signal; Step 2-6, if the local amplitude functions of the two channels are not all constantly 1, then save the local amplitude functions, use the obtained two-dimensional signal constructed by the frequency modulation signals of the two channels as the input signal, and repeat steps 2-2 to 2-5 until the local amplitude functions of the two channels are all constantly 1; Step 2-7, multiplying all the local amplitude functions saved in step 2-6 to obtain an integrated local amplitude function, and multiplying the integrated local amplitude function with the pure frequency modulation signal obtained in step 2-6 to obtain a product function component; Step 2-8, subtract the product function component from the input signal to obtain the output signal; if the output signal is not a monotonic signal, use the output signal as a new input signal and repeat steps 2-2 to 2-7 until the output signal is a monotonic signal; Step 3, performing Hilbert-Huang transform on the product function components of the two channels respectively to obtain an edge Hilbert-Huang spectrum diagram; Step 4, calculating the average band power of the specified frequency band in the edge Hilbert-Huang spectrum as the characteristic value; Step 5: Input the feature values ​​into the classifier to classify the EEG signals under different motor imagery tasks.

2. The method for classifying electroencephalogram signals based on two-dimensional local mean decomposition according to claim 1, characterized in that: In step 1, the EEG signal is preprocessed by performing 0.2-50 Hz bandpass filtering on the EEG signal through a Butterworth filter to obtain a filtered EEG signal.

3. The method for classifying EEG signals based on two-dimensional local mean decomposition according to claim 1, characterized in that: The specific process of step 3 is as follows: Step 3-1, decompose the input signal into the sum of several product function components and a residual component through step 2. The specific formula is as follows: Where x(t) is the constructed two-dimensional signal, PF i (t) is the i-th product function component, r(t) is the residual component; Step 3-2, perform Hilbert-Huang transform on each product function component. The specific formula is as follows: Step 3-3, construct the analytical signal of each product function component, and calculate the amplitude function and phase function of the analytical signal. The specific formula is as follows: Among them, x A (t) is the analytical signal, a i (t) and θ i (t) represent the amplitude function and phase function of the analytical signal respectively; Step 3-4, deriving the phase function of the analytical signal to obtain the frequency function, and obtaining the Hilbert-Huang spectrum H(ω,t) of the analytical signal based on the frequency function and the amplitude function; Step 3-5, integrate the Hilbert-Huang spectrum of the analytical signal with respect to time to obtain the edge Hilbert spectrum. The specific formula is as follows:

4. The method for classifying electroencephalogram signals based on two-dimensional local mean decomposition according to claim 1, characterized in that: In step 4, the average band power of the designated frequency band is the average band power of the 8-12 Hz frequency band, the 18-26 Hz frequency band, and the 0-50 Hz frequency band, which correspond to the Mu rhythm, the Beta rhythm, and the total frequency band, respectively.

5. The method for classifying electroencephalogram signals based on two-dimensional local mean decomposition according to claim 1, characterized in that: In step 5, the classifier adopts a support vector machine classifier based on radial basis function, and the support vector machine classifier determines the classifier model parameters through a ten-fold cross-validation during the pre-training process.

Images