Single-channel electroencephalogram drowsiness identification method based on ICEEMDAN-PSD

Abstract

The invention discloses a single-channel electroencephalogram signal sleepiness identification method based on ICEEMDAN-PSD. The method comprises the steps that single-channel electroencephalogram signals of a subject are collected; the single-channel electroencephalogram signals are preprocessed, and the collected electroencephalogram signals are divided into a plurality of signal segments with fixed duration; time domain analysis of the electroencephalogram signals is analyzed through ICEEMDAN, and features are extracted; the electroencephalogram signals are analyzed through PSD, frequency domain analysis is carried out, and features are extracted; then, feature selection is carried out on the extracted time-frequency domain features through a ReilefF algorithm; and inputting the selected time-frequency domain features into a random forest algorithm to obtain a drowsiness identification result of the single-channel electroencephalogram signal.

Description

Technical Field

[0001] The present invention relates to the field of sleepiness recognition, and in particular to a single-channel electroencephalogram signal sleepiness recognition method based on ICEEMDAN-PSD. Background Art

[0002] Sleep is the most basic physiological need of human beings and is closely related to human health. With the increasing pressure of life, more and more people are troubled by sleep disorders. Sleep disorders not only cause decreased immunity, but also cause negative effects such as memory loss and lack of concentration. As the transition stage from wakefulness to sleep, sleepiness is a key process for the correct diagnosis and treatment of sleep disorders. Therefore, accurate detection of sleepiness is of great research significance.

[0003] Since the beginning of the last century, many experts and scholars have conducted research on sleepiness detection, but due to the limitations of the scientific and technological level at the time, there was no way to detect sleepiness through quantitative indicators. In the following decades, researchers found that the physiological signals of the human body contain a wealth of information, among which brain waves, as physiological signals containing a large amount of neural activity information, are the golden indicator of the human body's state of consciousness. After research, it was found that the EEG of the human body will undergo obvious changes when sleepiness occurs, and these changes can be used to accurately detect sleepiness. At present, many scholars have proposed sleepiness detection methods based on multi-channel EEG signals. Although high spatial resolution and rich information can be obtained based on multi-channel EEG signals, the wearing comfort of this acquisition method is poor, which will affect the test results. Although the single-channel acquisition method has less interference to the subjects, it contains relatively less information and the data quality is poor. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, the present invention provides a single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD, which can greatly improve the accuracy of sleepiness recognition.

[0005] To achieve the above purpose, the present invention adopts the following technical means:

[0006] The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD includes the following steps:

[0007] Step 1: pre-process the collected single-channel EEG signal and divide it into multiple signal segments of fixed duration;

[0008] Step 2: Use ICEEMDAN to perform time domain analysis on the preprocessed EEG signal segment X(n) to obtain local feature signals IMF at different time scales and a residual res representing the signal trend;

[0009] Step 3: Use the Pearson correlation coefficient method to calculate the correlation coefficient between each IMF and the original signal and retain the first five IMF components with the largest correlation coefficient;

[0010] Step 4: Extract characteristic parameters of the retained IMF components as time domain characteristic parameters of the single-channel EEG signal. The time domain characteristic parameters include sample entropy and Hjorth parameter of each IMF; perform power spectral density analysis on each signal segment, and calculate the power spectral density and power spectral entropy of each signal segment;

[0011] Step 5: Use the Welch algorithm to find the PSD distribution curve of the EEG signal in different states after preprocessing, calculate the corresponding power spectrum density and power spectrum entropy, and the average power spectrum density, band energy and relative power of different sub-bands as frequency domain features;

[0012] Step 6: Use the ReliefF algorithm to perform feature importance analysis on the time domain features and frequency domain features obtained in steps 4 and 5, and select the top 20 features as input features of the classification model;

[0013] Step 7: Input the feature parameters selected in step 6 into the random forest classification algorithm to obtain the sleepiness recognition result; that is, the single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD is completed.

[0014] Preferably, in step one, a single-channel EEG signal acquisition device is used to acquire the subject's single-channel EEG signal at a sampling frequency of 500 Hz, and the acquired single-channel EEG signal is preprocessed.

[0015] Preferably, in step one, the preprocessing includes: first, filtering with a bandpass filter of 0.5 Hz to 30 Hz to filter out high-frequency signals; second, segmenting the filtered complete single-channel EEG signal into multiple segment signals with a certain duration.

[0016] Preferably, in step 4, the method for extracting sample entropy is:

[0017] Where: N represents the number of data in the time series {x(n)}, m represents the reconstruction dimension, r represents the threshold value, C m Represents the number of data whose distance between two m-dimensional reconstructed vectors is less than or equal to r.

[0018] Preferably, in step 4, the method for extracting the Hjorth parameter is:

[0019] Hjorth parameters include Hjorth activity, Hjorth mobility and Hjorth complexity:

[0020] The Hjorth activity formula is as follows:

[0021] AcTivity=σ 2 , where: σ is the standard deviation of the signal;

[0022] The Hjorth mobility formula is as follows:

[0023] Where: σ' is the standard deviation of the first-order difference signal;

[0024] The Hjorth complexity formula is as follows:

[0025] Where: σ” is the standard deviation of the second-order difference signal.

[0026] Preferably, in step 4, the power spectrum density is calculated as follows:

[0027] Divide the data x(n) with length N, n=0,1,L,N-1 into L segments, each segment has M data, and the i-th segment data is represented as: x i (n)=x(n+iM-M),0≤n≤M,1≤i≤L;

[0028] Then add the window function w(n) to each segment of data and find the periodogram of each segment. The periodogram of the i-th segment is: Where U is a factor used to reduce the influence of the window function on the power spectrum estimation.

[0029] Considering each periodogram segment as unrelated, the power spectral density estimated by the Welch method is:

[0030] Preferably, in step 4, the power spectrum entropy is calculated by:

[0031] Shannon entropy is defined as: Among them, x i is an event that follows a random process, P(x i ) for each event x i Probability of occurrence;

[0032] According to the definition of Shannon entropy, the power spectrum entropy H of the signal x(n) can be obtained x for: in: Represents the power spectral density.

[0033] Preferably, in step five, the average power spectral density is calculated by adding the power spectral densities of the rhythmic waves and then dividing by the number of frequency bands to obtain the average power spectral density.

[0034] Preferably, in step 5, the method for calculating the frequency band energy is:

[0035] Use Fourier transform to convert the signal from time domain to frequency domain, and then calculate the energy, the calculation formula is: Where X(f) is the frequency spectrum of the signal, f 1 、f 2 is the frequency range.

[0036] Preferably, in step 5, the relative power is calculated as follows:

[0037] First calculate the total power of the signal: Then calculate the power of the signal in a specific frequency band: Finally, calculate the relative power:

[0038] Beneficial effects:

[0039] 1. The present invention proposes a single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD. The method integrates the time domain analysis, frequency domain analysis and feature parameter selection method of time-frequency domain analysis of single-channel EEG signals, and combines the random forest classifier to realize sleepiness recognition, thereby effectively solving the problem that efficient sleepiness recognition cannot be performed through single-channel EEG signals.

[0040] 2. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD provided by the present invention uses ICEEMDAN to perform modal decomposition on the EEG signal and extract the 5 IMFs with the highest autocorrelation coefficients, and then extracts their sample entropy and Hjorth parameters as time domain features, and uses power spectral density to perform frequency domain analysis on the EEG signal, and then solves its power spectral entropy, average power, and the frequency band energy and relative power of four basic filters as frequency domain features, and then uses the ReliefF algorithm to select the feature parameters, and finally uses the random forest algorithm to perform sleepiness recognition, achieving a high recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 It is a flowchart of a single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD in a specific embodiment of the present invention;

[0043] Figure 2 This is an ICEEMDAN decomposition effect diagram in a specific implementation manner of the present invention;

[0044] Figure 3 This is a diagram showing the ranking of feature importance in a specific implementation manner of the present invention;

[0045] Figure 4 This is a random forest recognition effect diagram in a specific implementation manner of the present invention. DETAILED DESCRIPTION

[0046] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Example:

[0048] like Figure 1-4 As shown, the single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD includes the following steps:

[0049] Step S10, using a single-channel EEG signal acquisition device to acquire a single-channel EEG signal of the subject at a sampling frequency of 500 Hz;

[0050] Step S20, preprocessing the acquired single-channel EEG signal, the preprocessing includes: first, filtering with a bandpass filter of 0.5 Hz to 30 Hz to filter high-frequency signals; second, segmenting the filtered complete single-channel EEG signal into multiple segment signals with a certain duration; the duration of the segment signal is 30 seconds;

[0051] Step S30, using ICEEMDAN to decompose each acquired segment signal to obtain a local feature signal IMF containing different time scales and a residual res representing a signal trend;

[0052] Step S40, using the Pearson correlation coefficient method to obtain the correlation between the local feature signal IMF of different time scales and the original signal and retaining the first five IMF components with the largest correlation coefficient;

[0053] Step S50: Extract characteristic parameters of the retained IMF components as time domain characteristic parameters of the single-channel EEG signal, mainly including sample entropy and Hjorth parameter of each IMF. The following will introduce the extraction method of each time domain characteristic parameter in detail:

[0054] Sample entropy:

[0055] Sample entropy is a method to detect the complexity of a time series by measuring the probability of generating new patterns in a signal. The calculation formula is as follows: Where: N represents the number of data in the time series {x(n)}, m represents the reconstruction dimension, r represents the threshold value, C m Indicates the number of data whose distance between two m-dimensional reconstructed vectors is less than or equal to r;

[0056] Hjorth parameters:

[0057] Hjorth parameters include Hjorth activity, Hjorth mobility and Hjorth complexity:

[0058] The Hjorth activity formula is as follows:

[0059] AcTivity=σ 2 , where: σ is the standard deviation of the signal;

[0060] The Hjorth mobility formula is as follows:

[0061] Where: σ' is the standard deviation of the first-order difference signal;

[0062] The Hjorth complexity formula is as follows:

[0063] Where: σ” is the standard deviation of the second-order difference signal;

[0064] S60, performing power spectrum density analysis on each signal segment, and calculating the power spectrum density and power spectrum entropy of each signal segment. The calculation method will be described in detail below:

[0065] Power Spectral Density:

[0066] Divide the data x(n) with length N, n=0,1,L,N-1 into L segments, each segment has M data, and the i-th segment data is represented as: x i (n)=x(n+iM-M),0≤n≤M,1≤i≤L;

[0067] Then add the window function w(n) to each segment of data and find the periodogram of each segment. The periodogram of the i-th segment is: Where U is a factor used to reduce the influence of the window function on the power spectrum estimation.

[0068] Considering each periodogram segment as unrelated, the power spectral density estimated by the Welch method is:

[0069] Power spectral entropy:

[0070] Shannon entropy is defined as: Among them, x i is an event that follows a random process, P(x i ) for each event x i Probability of occurrence;

[0071] According to the definition of Shannon entropy, the power spectrum entropy H of the signal x(n) can be obtained x for: in: represents the power spectral density;

[0072] Step S70, extract the δ rhythmic waves of 0.5Hz-4Hz, theta of 4Hz-8Hz, alpha of 8Hz-14Hz, and beta of 14Hz-30Hz of the EEG signal through a bandpass filter, calculate their average power spectrum density and the band energy and relative power of different frequency bands, and use the power spectrum density and power spectrum entropy of each signal segment and the average power spectrum density of the rhythmic waves of different frequency bands and the band energy and relative power of different frequency bands as the frequency domain feature parameters of the EEG signal. The following will introduce the extraction method of each frequency domain feature parameter in detail:

[0073] Average power spectral density:

[0074] After obtaining the power spectral density of each rhythm wave, add them up and divide them by the number of frequency bands to get the average power spectral density:

[0075] Band Energy:

[0076] When calculating the energy of a frequency band, the energy can be calculated by the spectrum of the signal. Usually, Fourier transform is used to convert the signal from the time domain to the frequency domain, and then the energy is calculated. The calculation formula is: Where X(f) is the frequency spectrum of the signal, f 1 、f 2 is the frequency range;

[0077] Relative power of the signal:

[0078] The relative power of a signal usually refers to the ratio of the power of a signal in a certain frequency band to the total power of the signal. It is used to describe the relationship between the strength of a signal in a specific frequency band and the total power. It is often used in EEG analysis to describe the distribution of relative strengths in different frequency ranges. The calculation formula is as follows: First, calculate the total power of the signal: Then calculate the power of the signal in a specific frequency band: Finally, calculate the relative power:

[0079] Step S80, sorting the feature importance by the time-frequency domain feature parameters of the ReliedfF algorithm, and selecting the top 20 features as the features of the single-channel EEG;

[0080] Step S90: Inputting these characteristic parameters into the constructed random forest algorithm can obtain the sleepiness recognition result of the single-channel EEG signal.

[0081] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0082] In addition, it should be understood that although the present specification is described according to embodiments, not every embodiment contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation methods that those skilled in the art can understand.

Claims

1. A single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD, characterized by: The following steps are involved: Step 1: pre-process the collected single-channel EEG signal and divide it into multiple signal segments of fixed duration; Step 2: Use ICEEMDAN to perform time domain analysis on the preprocessed EEG signal segment X(n) to obtain local feature signals IMF at different time scales and a residual res representing the signal trend; Step 3: Use the Pearson correlation coefficient method to calculate the correlation coefficient between each IMF and the original signal and retain the first five IMF components with the largest correlation coefficient; Step 4: Extract characteristic parameters of the retained IMF components as time domain characteristic parameters of the single-channel EEG signal. The time domain characteristic parameters include sample entropy and Hjorth parameter of each IMF; perform power spectral density analysis on each signal segment, and calculate the power spectral density and power spectral entropy of each signal segment; Step 5: Use the Welch algorithm to find the PSD distribution curve of the EEG signal in different states after preprocessing, calculate the corresponding power spectrum density and power spectrum entropy, and the average power spectrum density, band energy and relative power of different sub-bands as frequency domain features; Step 6: Use the ReliefF algorithm to perform feature importance analysis on the time domain features and frequency domain features obtained in steps 4 and 5, and select the top 20 features as input features of the classification model; Step 7: Input the feature parameters selected in step 6 into the random forest classification algorithm to obtain the sleepiness recognition result; that is, the single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD is completed.

2. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD according to claim 1 is characterized in that: In the step 1, a single-channel EEG signal acquisition device is used to acquire the single-channel EEG signal of the subject at a sampling frequency of 500 Hz, and the acquired single-channel EEG signal is preprocessed.

3. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD according to claim 2 is characterized in that: In the step 1, the preprocessing includes: first, filtering with a bandpass filter of 0.5 Hz to 30 Hz to filter out high-frequency signals; second, segmenting the filtered complete single-channel EEG signal into multiple segment signals with a certain duration.

4. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD according to claim 1, characterized in that: In step 4, the method for extracting sample entropy is: Where: N represents the number of data in the time series {x(n)}, m represents the reconstruction dimension, r represents the threshold value, C m Represents the number of data whose distance between two m-dimensional reconstructed vectors is less than or equal to r.

5. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD according to claim 4 is characterized in that: In step 4, the method for extracting the Hjorth parameter is: Hjorth parameters include Hjorth activity, Hjorth mobility and Hjorth complexity: The Hjorth activity formula is as follows: AcTivity=σ 2 , where: σ is the standard deviation of the signal; The Hjorth mobility formula is as follows: Where: σ' is the standard deviation of the first-order difference signal; The Hjorth complexity formula is as follows: Where: σ” is the standard deviation of the second-order difference signal.

6. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD according to claim 5, characterized in that: In step 4, the power spectrum density is calculated as follows: Divide the data x(n) with length N, n=0,1,L,N-1 into L segments, each segment has M data, and the i-th segment data is represented as: x i (n)=x(n+iM-M),0≤n≤M,1≤i≤L; Then add the window function w(n) to each segment of data and find the periodogram of each segment. The periodogram of the i-th segment is: Where U is a factor used to reduce the influence of the window function on the power spectrum estimation. Considering each periodogram segment as unrelated, the power spectral density estimated by the Welch method is:

7. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD according to claim 6, characterized in that: In the step 4, the power spectrum entropy is calculated as follows: Shannon entropy is defined as: Among them, x i is an event that follows a random process, P(x i ) for each event x i Probability of occurrence; According to the definition of Shannon entropy, the power spectrum entropy H of the signal x(n) can be obtained x for: in: Represents the power spectral density.

8. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD according to claim 1, characterized in that: In the step 5, the average power spectrum density is calculated by adding the power spectrum densities of the rhythmic waves and then dividing by the number of frequency bands to obtain the average power spectrum density.

9. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD according to claim 8, characterized in that: In step 5, the calculation method of the frequency band energy is: Use Fourier transform to convert the signal from time domain to frequency domain, and then calculate the energy, the calculation formula is: Where X(f) is the spectrum of the signal, and f1 and f2 are the frequency ranges.

10. The single-channel EEG signal sleepiness recognition method based on ICEEMDAN-PSD according to claim 9, characterized in that: In step 5, the relative power is calculated as follows: First calculate the total power of the signal: Then calculate the power of the signal in a specific frequency band: Finally, calculate the relative power:

Images