Fatigue electroencephalogram feature extraction method based on wavelet packet and sample entropy

Abstract

The invention discloses a fatigue electroencephalogram feature extraction method based on a wavelet packet and sample entropy. The fatigue electroencephalogram feature extraction method comprises the following steps: firstly, carrying out wavelet packet transformation on a preprocessed electroencephalogram signal; signals of different frequency bands are obtained, and local time-frequency characteristics are obtained; frequency band energy of the four basic rhythms after wavelet packet decomposition is calculated, and relative energy values and energy ratio characteristics of the frequency band energy are determined; the sample entropy of the electroencephalogram signals is calculated, and the complexity of the electroencephalogram signals is quantified; combining the rhythm relative energy value, the energy ratio feature and the sample entropy into a comprehensive feature vector, and verifying the generalization ability of the model by using methods such as cross validation; and inputting the constructed feature vectors into a support vector machine for classification, and verifying the feature extraction performance and the fatigue state recognition accuracy of the cross fusion algorithm. According to the method, the precision and the resolution of signal analysis are improved, and the distinguishing capability and the classification performance of the features are enhanced, so that remarkable advantages are brought to electroencephalogram signal classification and fatigue detection application.

Description

Technical Field

[0001] This invention relates to the field of fatigue detection in electroencephalogram (EEG) signal processing, and more particularly to a method for extracting fatigue EEG features based on wavelet packets and sample entropy. Background Technology

[0002] Electroencephalography (EEG), as a unique bioelectrical signal in the human body, contains rich information about brain activity and can reflect an individual's physical and mental state. In recent years, with the widespread application of EEG signal detection technology in the field of fatigue monitoring, how to efficiently and accurately extract fatigue characteristics from EEG signals has become a key research area.

[0003] Existing technologies primarily employ time-domain analysis, frequency-domain analysis, time-frequency analysis, and nonlinear dynamics analysis to analyze electroencephalogram (EEG) signals. While each method has its advantages, the complexity and non-stationarity of EEG signals often make it difficult to comprehensively capture their dynamic changes. Particularly in the identification of fatigue states, subtle changes in EEG signals are frequently overlooked, limiting detection accuracy and real-time performance.

[0004] Wavelet packet transform (WPT), a commonly used multi-scale signal analysis method, excels in characterizing local signal features. It adaptively selects the corresponding frequency bands of the signal to match its spectrum, extracting the time-frequency features of the EEG signal by decomposing it into the four basic rhythms. Compared to wavelet transform, wavelet packet transform overcomes the shortcomings of poor frequency resolution in the high-frequency domain and poor time resolution in the low-frequency domain, providing better time-frequency characteristics and ensuring no signal loss. However, when used alone, wavelet packet transform's ability to analyze spatial dimensions is relatively insufficient, making it difficult to fully reveal the spatiotemporal characteristics of the EEG signal. Summary of the Invention

[0005] The purpose of this invention is to provide a fatigue EEG feature extraction method based on wavelet packets and sample entropy, thereby solving the aforementioned problems existing in the prior art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A fatigue EEG feature extraction method based on wavelet packet and sample entropy includes the following steps:

[0008] S100. First, wavelet packet transform is performed on the preprocessed EEG signal to obtain signals of different frequency bands and acquire local time-frequency features.

[0009] S200. Calculate the frequency band energy of the four basic rhythms after wavelet packet decomposition, and determine their relative energy values ​​and energy ratio characteristics.

[0010] S300. Calculate the sample entropy of the EEG signal and quantify the complexity of the EEG signal;

[0011] S400. Combine the rhythm relative energy value, energy ratio feature and sample entropy into a comprehensive feature vector, and use methods such as cross-validation to verify the generalization ability of the model.

[0012] S500. Input the constructed feature vectors into the support vector machine for classification to verify the feature extraction performance and fatigue state recognition accuracy of the cross-fusion algorithm.

[0013] In some specific embodiments, the specific method of step S100 is as follows:

[0014] Let the original EEG signal be f(t). The wavelet packet decomposition coefficients at the j-th layer and k-th point after wavelet packet transform can be expressed by equations (1) and (2):

[0015]

[0016] Where h(n) and g(n) are the low-pass filter and the high-pass filter, respectively.

[0017] In some specific embodiments, the specific method of step S200 is as follows:

[0018] Extract the four EEG rhythm signals δ (0-4Hz), θ (4-8Hz), α (8-13Hz), and β (13-30Hz), and calculate the energies of the four rhythms Eδ, θ, α, and β, respectively. δ E θ E α E β And calculate the total energy E of the four basic rhythms. total ;

[0019]

[0020] E total =E δ +E θ +E α +E β (16)

[0021] Where E k x(k) represents the energy value of a specific rhythm among the four rhythms δ, θ, α, and β. n E represents the amplitude at each sampling point of the rhythm. total It represents total energy.

[0022] In some specific embodiments, the method of step S200 further includes:

[0023] The relative energy values of the four rhythms of δ, θ, α, and β calculated according to the rhythm wave energy are P δ , P θ , P α , P β , and calculate the energy ratio H:

[0024] P k = E k / E total (17)

[0025]

[0026] where P k represents the relative energy value of a certain rhythm among the four rhythms of δ, θ, α, and β.

[0027] In some specific embodiments, the specific method of step S300 is as follows:

[0028] Extract the sample entropy feature of the EEG signal, use the sample entropy to quantify the complexity of the EEG signal; construct a sequence vector with an embedding dimension of m:

[0029] X m (i)=[x(i), x(i + 1), …, x(i + m - 1)] (19)

[0030] where 1 ≤ i ≤ N - m + 1;

[0031] Define the distance between the vectors X m (i) and X m (j) as d[X m (i), X m (j)], calculate the distance between the sequence vectors, that is

[0032] d[X m (i), X m (j)] = max|x(i + k) - x(j + k)|. (20)

[0033] In some specific embodiments, the specific method of step S400 is as follows:

[0034] Given a threshold r (r > 0), count the ratio of the number of d[X m (i), X m (j)] < r to the total number of vectors N - m - 1, denoted as

[0035]

[0036] Find its average value for all i:

[0037]

[0038] Increment the dimension by 1, resulting in an m+1 dimensional vector. Repeat the steps from formula (7) to formula (10) to obtain B. m+1 (r); The sample entropy of the sequence is:

[0039]

[0040] When N takes finite values, the estimated value of the sample entropy is:

[0041]

[0042] In some specific embodiments, the specific method of step S500 is as follows:

[0043] The calculated relative energies of each frequency band, δ / (δ+θ+α+β), θ / δ+θ+α+β, α / δ+θ+α+β, and β / δ+θ+α+β, along with the energy ratio H and sample entropy SampEn, are used as feature vectors and fused together. The final feature vector is expressed as follows:

[0044] F = [P] δ P θ P α P β H, SampEn]. (13)

[0045] The beneficial effects of this invention are:

[0046] This invention discloses a fatigue EEG feature extraction method based on wavelet packet and sample entropy, comprising the following steps: First, wavelet packet transform is performed on the preprocessed EEG signal to obtain signals in different frequency bands and acquire local time-frequency features; the frequency band energy of the four basic rhythms after wavelet packet decomposition is calculated, and their relative energy values ​​and energy ratio features are determined; the sample entropy of the EEG signal is calculated to quantify the complexity of the EEG signal; the rhythm relative energy value, energy ratio features, and sample entropy are combined into a comprehensive feature vector, and the generalization ability of the model is verified using methods such as cross-validation; the constructed feature vector is input into a support vector machine for classification to verify the feature extraction performance of the cross-fusion algorithm and the accuracy of fatigue state recognition. Wavelet packet transform (WPT) can perform orthogonal decomposition of signals across the entire frequency band. It provides a more refined signal analysis method and can adaptively select the corresponding frequency band according to the signal characteristics to match the signal spectrum, thereby improving the time-frequency resolution. It is very suitable for the analysis and processing of non-stationary EEG signals. Wavelet packet transform not only decomposes the low-frequency approximation of a signal but also its high-frequency details, preserving more of the original signal's characteristics. The decomposed signal more realistically and precisely reflects the signal's essence. While inheriting the essence of wavelet transform, it also compensates for its shortcomings, improving the accuracy of EEG signal analysis. Due to its excellent time-frequency characteristics, wavelet packet transform is often used to decompose the four basic rhythms of EEG signals to extract time-frequency features. Under different fatigue states, the relative energy values ​​of the four basic rhythms of EEG signals after wavelet packet transform show significant differences. Sample entropy (SE) is an improved complexity measurement method proposed based on approximate entropy. It not only possesses all the advantages of approximate entropy but also avoids the problem of inconsistent statistics in approximate entropy. As a nonlinear analysis method, sample entropy can effectively measure the complexity of EEG signals and reflect their nonlinear characteristics; sample entropy values ​​also show significant differences under different fatigue states. By combining the rhythmic relative energy value, energy ratio, and sample entropy after wavelet packet transform, a more effective comprehensive vector is generated, enabling efficient extraction of multi-dimensional features from EEG signals. This vector is then input into an SVM classifier for training and testing. This method not only improves the accuracy and resolution of signal analysis but also enhances the feature discrimination ability and classification performance, thus bringing significant advantages to EEG signal classification and fatigue detection applications. Attached Figure Description

[0047] Figure 1 This is a flowchart of the WPT-SE feature extraction process of the present invention;

[0048] Figure 2 This is a schematic diagram of the three-layer wavelet packet decomposition structure of the present invention;

[0049] Figure 3This is a schematic diagram of the four basic rhythmic signals in the awake and fatigued states after wavelet packet transformation according to the present invention;

[0050] Figure 4 These are the entropy values ​​of two states for the 16 channels of this invention;

[0051] Figure 5 This is a schematic diagram of wavelet packet energy feature extraction according to the present invention;

[0052] Figure 6 This is a schematic diagram of sample entropy feature extraction according to the present invention;

[0053] Figure 7 This invention relates to wavelet packet and sample entropy fusion feature extraction;

[0054] Figure 8 This is a flowchart of a fatigue EEG feature extraction method based on wavelet packets and sample entropy according to the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0056] The technology of this invention aims to improve the feature recognition rate of electroencephalogram (EEG) signals, so as to capture important information in EEG signals more comprehensively and accurately, and provide a more reliable processing solution for neuroscience research and other brain science research fields.

[0057] The unique feature of this invention is that it combines the relative energy values, energy ratios and sample entropy of the four basic rhythms after wavelet packet transform, which brings many advantages to the feature extraction of EEG signals.

[0058] Reference Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 and Figure 8 The fatigue EEG feature extraction method shown includes the following steps:

[0059] S100. First, wavelet packet transform is performed on the preprocessed EEG signal to obtain signals in different frequency bands and acquire local time-frequency features. Wavelet packet transform can decompose the signal into various frequency bands and obtain the coefficients of different frequency bands.

[0060] S200. Calculate the frequency band energy of the four basic rhythms after wavelet packet decomposition, and determine their relative energy values ​​and energy ratio characteristics.

[0061] S300. Calculate the sample entropy of the EEG signal and quantify the complexity of the EEG signal.

[0062] S400. Combine the rhythm relative energy value, energy ratio feature and sample entropy into a comprehensive feature vector, and use methods such as cross-validation to verify the generalization ability of the model.

[0063] S500. Input the constructed feature vectors into the support vector machine for classification to verify the feature extraction performance and fatigue state recognition accuracy of the cross-fusion algorithm.

[0064] In some specific embodiments, the specific method of step S100 is as follows:

[0065] Let the original EEG signal be f(t). The wavelet packet decomposition coefficients at the j-th layer and k-th point after wavelet packet transform can be expressed by equations (1) and (2):

[0066]

[0067] Where h(n) and g(n) are the low-pass filter and the high-pass filter, respectively.

[0068] In some specific embodiments, the specific method of step S200 is as follows:

[0069] Extract the four EEG rhythm signals δ (0-4Hz), θ (4-8Hz), α (8-13Hz), and β (13-30Hz), and calculate the energies of the four rhythms Eδ, θ, α, and β, respectively. δ E θ E α E β And calculate the total energy E of the four basic rhythms. total ;

[0070]

[0071] E total =E δ +E θ +E α +E β (28)

[0072] Where E k x(k) represents the energy value of a specific rhythm among the four rhythms δ, θ, α, and β. n E represents the amplitude at each sampling point of the rhythm. total It represents total energy.

[0073] In some specific embodiments, the method of step S200 further includes:

[0074] The relative energy values of the four rhythms of δ, θ, α, and β calculated based on the rhythm wave energy are P δ , P θ , P α , P β , and calculate the energy ratio H:

[0075] P k =E k / E total (29)

[0076]

[0077] where P k represents the relative energy value of a certain rhythm among the four rhythms of δ, θ, α, and β.

[0078] In some specific embodiments, the specific method of step S300 is as follows:

[0079] Extract the sample entropy feature of the EEG signal, and use the sample entropy to quantify the complexity of the EEG signal; construct a sequence vector with an embedding dimension of m:

[0080] X m (i)=[x(i),x(i + 1),…,x(i + m - 1)] (31)

[0081] where 1≤i≤N - m + 1;

[0082] Define the distance between the vectors X m (i) and X m (j) as d[X m (i),X m (j)], calculate the distance between the sequence vectors, that is

[0083] d[X m (i),X m (j)] = max|x(i + k)-x(j + k)|. (32)

[0084] In some specific embodiments, the specific method of step S400 is as follows:

[0085] Given a threshold r (r>0), count the ratio of the number of d[X m (i),X m (j)]<r to the total number of vectors N - m - 1, denoted as B i m (r):

[0086]

[0087] Find its average value for all i:

[0088]

[0089] Increment the dimension by 1, resulting in an m+1 dimensional vector. Repeat the steps from formula (7) to formula (10) to obtain B. m+1 (r); The sample entropy of the sequence is:

[0090]

[0091] When N takes finite values, the estimated value of the sample entropy is:

[0092]

[0093] In some specific embodiments, the specific method of step S500 is as follows:

[0094] The calculated relative energies of each frequency band, δ / (δ+θ+α+β), θ / δ+θ+α+β, α / δ+θ+α+β, and β / δ+θ+α+β, along with the energy ratio H and sample entropy SampEn, are used as feature vectors and fused together. The final feature vector is expressed as follows:

[0095] F = [P] δ P θ P α P β H, SampEn]. (13)

[0096] The EEG signal feature extraction process of this invention combines the advantages of two algorithms: wavelet packet transform (WPT) and sample entropy (SE). Wavelet packet transform highlights the time-frequency characteristics of the signal, while sample entropy captures the nonlinear characteristics of the EEG signal. The combination of the two can extract fatigue-related feature information more comprehensively.

[0097] In the feature integration stage, rhythmic energy, energy ratios of different frequency bands, and sample entropy of the entire signal are first extracted to construct feature vectors. Then, the feature vectors are normalized, and feature selection and standardization ensure data consistency. Next, principal component analysis (PCA) is used to reduce the dimensionality of the integrated feature matrix, simplifying the high-dimensional feature space and reducing data complexity. Finally, the integrated feature vectors are input into a support vector machine (SVM) classifier to achieve effective classification of EEG signals.

[0098] This comprehensive feature extraction method, which integrates wavelet packets and sample entropy, allows for a more complete extraction of key features from EEG signals. It not only reveals the multi-domain characteristics of the signal but also identifies changes in the EEG signal across different time scales, significantly improving feature discrimination. This method provides more discriminative input for subsequent classification models, comprehensively revealing the temporal and frequency characteristics of the signal and offering strong support for in-depth research into EEG signal analysis.

[0099] Brainwave rhythms are crucial in EEG signal analysis, with delta, theta, alpha, and beta being the most prevalent rhythmic signals. These four basic rhythms effectively reflect a subject's mental fatigue state. Extracting the relative energy characteristics and energy ratios of these four basic rhythms allows for effective differentiation of the fatigue state represented by the EEG signals. Wavelet packet decomposition and reconstruction algorithms can accurately decompose these four basic rhythmic components. Figure 3 It is a three-layer wavelet packet decomposition.

[0100] From an energy perspective, after wavelet packet decomposition, the energy of an EEG signal is distributed across different time and frequency domains. Since the integral of the squared amplitude after wavelet packet decomposition is proportional to the energy of the original EEG signal, the magnitude of the wavelet packet decomposition coefficients can effectively represent the strength of the EEG signal. Therefore, using the sum of the squares of the wavelet packet decomposition coefficients to represent the energy of the EEG signal is reasonable. Each frequency band can be considered as a wavelet packet subband. By selecting specific wavelet packet subbands for signal reconstruction, the wavelet packet decomposition coefficients corresponding to the four basic rhythm waves can be extracted, thus obtaining the four basic rhythm signals after wavelet packet decomposition and reconstruction. Figure 3 The diagram shows the four basic rhythmic signals under awake and fatigued states after wavelet packet transform.

[0101] Sample entropy can be used for fatigue detection. Theoretically, the sample entropy value of a subject in a conscious state is significantly higher than that in a fatigued state. The value of sample entropy is mainly determined by three parameters: dimension m, threshold r, and number of sampling points M. A fatigue EEG signal dataset containing 12 subjects was selected as input data, with m = 2, r = 0.2std (std is the standard deviation of the original sequence), and M = 1000. In the experiment, sample entropy was extracted from 16 channels for each subject. Then, 300 entropy feature points were selected as the training set, and another 300 were selected as the test set. In each training or test set, fatigue and conscious samples accounted for half. Figure 4 The sample entropy values ​​of the 16 channels in two states are displayed.

[0102] Finally, 80% of the sample data from the fatigue group and the non-fatigue group were randomly selected as training samples, and the remaining 20% ​​were used as test samples. The classification performance of three feature extraction methods—wavelet packet energy, sample entropy, and wavelet packet and sample entropy fusion feature extraction—was compared, and the data were input into an SVM classifier for classification. These feature vectors were transformed into energy-entropy ratios and normalized using a logistic regression model based on Newton's iteration method, ultimately transforming them into a two-dimensional feature point map for easy intuitive analysis and comparison. Experimental results show that when using the combined wavelet packet and sample entropy feature extraction method, compared to using any single feature extraction strategy, this method exhibits superior performance in classification tasks, verifying the effectiveness and practicality of the proposed feature fusion strategy. A comparison of the feature extraction results of the single algorithm and the fusion algorithm is shown in the figure below. Figures 5-7 .

[0103] By adopting the above-disclosed technical solution of this invention, the following beneficial effects are obtained:

[0104] This invention discloses a fatigue EEG feature extraction method based on wavelet packet and sample entropy, comprising the following steps: First, wavelet packet transform is performed on the preprocessed EEG signal to obtain signals in different frequency bands and acquire local time-frequency features; the frequency band energy of the four basic rhythms after wavelet packet decomposition is calculated, and their relative energy values ​​and energy ratio features are determined; the sample entropy of the EEG signal is calculated to quantify the complexity of the EEG signal; the rhythm relative energy value, energy ratio features, and sample entropy are combined into a comprehensive feature vector, and the generalization ability of the model is verified using methods such as cross-validation; the constructed feature vector is input into a support vector machine for classification to verify the feature extraction performance of the cross-fusion algorithm and the accuracy of fatigue state recognition. Wavelet packet transform (WPT) can perform orthogonal decomposition of signals across the entire frequency band. It provides a more refined signal analysis method and can adaptively select the corresponding frequency band according to the signal characteristics to match the signal spectrum, thereby improving the time-frequency resolution. It is very suitable for the analysis and processing of non-stationary EEG signals. Wavelet packet transform not only decomposes the low-frequency approximation of a signal but also its high-frequency details, preserving more of the original signal's characteristics. The decomposed signal more realistically and precisely reflects the signal's essence. While inheriting the essence of wavelet transform, it also compensates for its shortcomings, improving the accuracy of EEG signal analysis. Due to its excellent time-frequency characteristics, wavelet packet transform is often used to decompose the four basic rhythms of EEG signals to extract time-frequency features. Under different fatigue states, the relative energy values ​​of the four basic rhythms of EEG signals after wavelet packet transform show significant differences. Sample entropy (SE) is an improved complexity measurement method proposed based on approximate entropy. It not only possesses all the advantages of approximate entropy but also avoids the problem of inconsistent statistics in approximate entropy. As a nonlinear analysis method, sample entropy can effectively measure the complexity of EEG signals and reflect their nonlinear characteristics; sample entropy values ​​also show significant differences under different fatigue states. By combining the rhythmic relative energy value, energy ratio, and sample entropy after wavelet packet transform, a more effective comprehensive vector is generated, enabling efficient extraction of multi-dimensional features from EEG signals. This vector is then input into an SVM classifier for training and testing. This method not only improves the accuracy and resolution of signal analysis but also enhances the feature discrimination ability and classification performance, thus bringing significant advantages to EEG signal classification and fatigue detection applications.

[0105] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A fatigue EEG feature extraction method based on wavelet packets and sample entropy, characterized in that: The following steps are involved: S100, firstly, performing wavelet packet transform on the pre-processed EEG signal; To obtain signals in different frequency bands and obtain local time-frequency features; S200, calculating the frequency band energy of the four basic rhythms after the wavelet packet decomposition, and determining the relative energy value and energy ratio characteristics thereof; S300, calculating the sample entropy of the EEG signal to quantify the complexity of the EEG signal; S400, combining the rhythm relative energy value, energy ratio feature and sample entropy into a comprehensive feature vector, and using methods such as cross-validation to verify the generalization ability of the model; S500, input the constructed feature vector into a support vector machine for classification, and verify the feature extraction performance and fatigue state recognition accuracy of the cross-fusion algorithm.

2. The fatigue EEG feature extraction method based on wavelet packet and sample entropy according to claim 1 is characterized in that: The specific method of step S100 is: Assume that the original EEG signal is f(t), and the wavelet packet decomposition coefficients at the jth layer and the kth point after wavelet packet transformation can be expressed by equations (1) and (2): Among them, h(n) and g(n) are low-pass filter and high-pass filter respectively.

3. The fatigue EEG feature extraction method based on wavelet packet and sample entropy according to claim 2 is characterized in that: The specific method of step S200 is: Extract the four EEG rhythm signals of δ (0-4Hz), θ (4-8Hz), α (8-13Hz) and β (13-30Hz), and calculate the energy of the four rhythms of δ, θ, α and β as E δ , E θ , E α , E β , and calculate the total energy E of the four basic rhythms total ; AND total =And δ +E θ +E α +E β (4) Where E k represents the energy value of a rhythm among the four rhythms of δ, θ, α and β, x(k) n is the amplitude of each sampling point of the rhythm, E total Represents total energy.

4. The fatigue EEG feature extraction method based on wavelet packet and sample entropy according to claim 3 is characterized in that: The specific method of step S200 also includes: According to the rhythm wave energy, the relative energy values ​​of the four rhythms of δ, θ, α and β are calculated as P δ , P θ , P α , P β , and calculate the energy ratio H: P k =And k / AND total (5) Where P k Represents the relative energy value of one of the four rhythms: δ, θ, α and β.

5. The fatigue EEG feature extraction method based on wavelet packet and sample entropy according to claim 4 is characterized in that: The specific method of step S300 is: Extract the sample entropy features of the EEG signal and use the sample entropy to quantify the complexity of the EEG signal; construct a sequence vector with an embedding dimension of m: X m (i)=[x(i),x(i+1),…,x(i+m-1)] (7) Where, 1≤i≤N-m+1; Define vector X m (i) and X m The distance between (j) is d[X m (i),X m (j)], calculate the distance between sequence vectors, that is, d[X m (i),X m (j)]=max|x(i+k)-x(j+k)|。 (8) 6. The fatigue EEG feature extraction method based on wavelet packet and sample entropy according to claim 5 is characterized in that: The specific method of step S400 is: Given a threshold r (r > 0), count the ratio of the number of d[X m (i), X m (j)] < r to the total number N - m - 1 of vectors, denoted as Find the average value for all i: Add 1 to the dimension, and it is now an m+1-dimensional vector. Repeat the steps from formula (7) to formula (10) above to obtain B m+1 (r); the sample entropy of the sequence is: When N takes a finite value, the estimated value of sample entropy is:

7. The fatigue EEG feature extraction method based on wavelet packet and sample entropy according to claim 6 is characterized in that: The specific method of step S500 is: The calculated relative energy of each frequency band, δ / (δ+θ+α+β) relative energy, θ / δ+θ+α+β relative energy, α / δ+θ+α+β relative energy, β / δ+θ+α+β relative energy, energy ratio H and sample entropy SampEn are fused as feature vectors, and the final feature vector is expressed as: F [P δ 、P θ 、P α 、P β H, SampEn, (13).

Images