A coupling feature extraction method of multi-channel non-stationary signals

Abstract

This invention relates to a method for extracting coupling features from multi-channel non-stationary signals, comprising: dividing the multi-channel non-stationary signal to be tested into several time segments according to a preset duration, obtaining a multi-channel non-stationary signal for each time segment; decomposing the multi-channel non-stationary signal of each time segment into multiple intrinsic mode functions (IMFs), obtaining multiple IMFs for each time segment; analyzing the phase-amplitude coupling degree between different frequency bands using the multiple IMFs of each time segment to obtain the cross-frequency coupling values ​​between different frequency bands within each time segment; statistically analyzing the global distribution of the cross-frequency coupling values ​​between different frequency bands within all time segments, and setting a baseline value based on the global distribution; comparing the cross-frequency coupling values ​​between different frequency bands within each time segment with the baseline value, and using the cross-frequency coupling values ​​exceeding the baseline value as the final frequency domain features. This method can achieve accurate and high-resolution analysis of non-stationary signals in both the time and frequency domains.

Description

Technical Field

[0001] This invention relates to feature extraction methods, and more particularly to a method for extracting coupled features from multi-channel non-stationary signals. Background Technology

[0002] Electroencephalogram (EEG) signals, as typical non-stationary signals, have frequency information closely related to specific physiological functions. Cross-frequency coupling between different frequency bands has been proven to be closely related to cognitive function, emotional state, and disease characteristics in animals.

[0003] However, traditional event-related potential (ERP) analysis methods rely on known external event markers to align cross-frequency coupling analysis with specific stimulus moments, which limits the application of this method in exploring unknown intrinsic brain state changes.

[0004] Furthermore, traditional time-frequency conversion methods are limited by the Heisenberg uncertainty principle, making it impossible to simultaneously and accurately acquire time and frequency domain information. High-resolution time and frequency domain analysis is particularly important when processing short-duration, non-stationary signals. While existing Empirical Mode Decomposition (EMD) methods can alleviate this problem to some extent, single-channel EMD suffers from mode aliasing and inter-channel mode misalignment when processing multi-channel signals, affecting the accuracy of subsequent cross-frequency coupling analysis. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for extracting coupling features of multi-channel non-stationary signals that can improve the accuracy of cross-frequency coupling analysis, in contrast to the above-mentioned prior art.

[0006] The technical solution adopted by this invention to solve the above-mentioned technical problems is as follows: a method for extracting coupling features of multi-channel non-stationary signals, characterized by comprising the following steps:

[0007] Step 1: Divide the multi-channel non-stationary signal to be tested into several time segments according to the preset duration to obtain the multi-channel non-stationary signal of each time segment;

[0008] Step 2: Decompose the multi-channel non-stationary signal of each time segment into multiple intrinsic mode functions to obtain multiple intrinsic mode functions for each time segment;

[0009] Step 3: Analyze the phase-amplitude coupling between different frequency bands using multiple intrinsic mode functions of each time segment to obtain the cross-frequency coupling values ​​between different frequency bands within each time segment;

[0010] Step 4: Calculate the global distribution of cross-frequency coupling values ​​between different frequency bands within all time segments, and set the baseline value based on the global distribution;

[0011] Step 5: Compare the cross-frequency coupling values ​​between different frequency bands within each time segment with the baseline values. The cross-frequency coupling values ​​that exceed the baseline values ​​are taken as the final frequency domain features. The final frequency domain features are the extracted coupling features.

[0012] Specifically, the method for obtaining multiple intrinsic mode functions for a certain time segment in step 2 is as follows:

[0013] Using the number of channels of the non-stationary signal as the number of projection directions, project the signal along each projection direction and calculate the local extremum point of each projected signal;

[0014] Subsequently, the local extrema of each projected signal are mapped back to the original multi-channel space to generate multidimensional envelopes in each projection direction;

[0015] Calculate the mean of the multidimensional envelopes in all projection directions, and subtract the mean of the multidimensional envelopes in all projection directions from the original multi-channel signal of the current time segment to obtain the first intrinsic mode function.

[0016] Subtract the first intrinsic mode function from the original multi-channel signal of the current time segment to obtain the new multi-channel signal of the current time segment. The second, third, ..., Nth intrinsic mode functions can be obtained in the same way as above, where N is the total number of intrinsic mode functions.

[0017] Preferably, the multidimensional envelopes in each projection direction are generated by a multidimensional interpolation method based on radial basis functions.

[0018] Preferably, the specific method for obtaining the cross-frequency coupling value in step 3 is as follows:

[0019] The instantaneous amplitude and instantaneous phase of each intrinsic mode function are obtained. The instantaneous phase corresponding to one low-frequency intrinsic mode function is selected as the phase signal, and the instantaneous amplitude corresponding to one high-frequency intrinsic mode function is selected as the amplitude signal. The coupling strength between the phase and amplitude is quantized by modulation exponent to obtain the cross-frequency coupling value.

[0020] Preferably, a Hilbert transform is performed on each intrinsic mode function to construct an analytic signal; the instantaneous amplitude and instantaneous phase of each intrinsic mode function are solved using the analytic signal.

[0021] Preferably, the specific processing procedure in step 4 is as follows: statistically analyze the cross-frequency coupling values ​​between different frequency bands in all time segments, calculate the mean and standard deviation, and the baseline value is the mean plus k times the standard deviation, where k is a preset positive number.

[0022] Preferably, the non-stationary signal is an electroencephalogram (EEG) signal.

[0023] Compared with existing technologies, the advantages of this invention are as follows: By decomposing multi-channel non-stationary signals into multiple intrinsic mode functions, it solves the problems of mode aliasing and mode misalignment in the multi-channel signal decomposition of traditional EMD, thereby achieving accurate high-resolution analysis of non-stationary signals in both the time and frequency domains and improving the accuracy of cross-frequency coupling analysis. Furthermore, by statistically analyzing the global distribution of cross-frequency coupling values ​​between different frequency bands across all time segments and setting a baseline value based on this global distribution, this method can automatically identify statistically significant cross-frequency coupling mode changes in signals without relying on external event markers. This makes it suitable for scenarios such as exploring unknown intrinsic brain state changes, enhancing the automation and applicability of cross-frequency coupling analysis. Attached Figure Description

[0024] Figure 1 This is a flowchart of the coupling feature extraction method for multi-channel non-stationary signals in an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram illustrating the decomposition of EEG signals into different frequency bands in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0027] like Figure 1 As shown, the coupling feature extraction method for multi-channel non-stationary signals in this embodiment includes the following steps:

[0028] Step 1: Divide the multi-channel non-stationary signal to be tested into several time segments according to the preset duration to obtain the multi-channel non-stationary signal of each time segment;

[0029] In this embodiment, the multi-channel non-stationary signal to be tested is divided into several time segments of equal length in chronological order, and the last part that is not long enough is discarded.

[0030] Step 2: Decompose the multi-channel non-stationary signal of each time segment into multiple intrinsic mode functions to obtain multiple intrinsic mode functions for each time segment;

[0031] Step 3: Analyze the phase-amplitude coupling between different frequency bands using multiple intrinsic mode functions of each time segment to obtain the cross-frequency coupling values ​​between different frequency bands within each time segment;

[0032] Step 4: Calculate the global distribution of cross-frequency coupling values ​​between different frequency bands within all time segments, and set the baseline value based on the global distribution;

[0033] Step 5: Compare the cross-frequency coupling values ​​between different frequency bands within each time segment with the baseline values. The cross-frequency coupling values ​​that exceed the baseline values ​​are taken as the final frequency domain features. The final frequency domain features are the extracted coupling features.

[0034] In step 2, the multiple intrinsic mode functions (IMFs) for a given time segment are obtained as follows: the number of channels of the non-stationary signal is equal to the number of projection directions. The signal is projected along each projection direction, and the local extrema of each projected signal are calculated. Subsequently, the local extrema of each projected signal are mapped back to the original multi-channel space, and multi-dimensional envelopes are generated along each projection direction using a multi-dimensional interpolation method based on radial basis functions. The mean of the multi-dimensional envelopes along all projection directions is calculated, and the mean of the multi-dimensional envelopes along all projection directions is subtracted from the original multi-channel signal of the current time segment to obtain the first IMF. The first IMF is subtracted from the original multi-channel signal of the current time segment to obtain the new multi-channel signal of the current time segment. The second, third, ..., Nth IMFs can be obtained sequentially in the same manner, where N is the total number of IMFs.

[0035] like Figure 2 As shown, in this embodiment, N=4, that is, the multi-channel non-stationary signal of a certain time segment is decomposed into 4 intrinsic mode functions, namely the first intrinsic mode function IMF1, the second intrinsic mode function IMF2, the third intrinsic mode function IMF3 and the fourth intrinsic mode function IMF4.

[0036] By decomposing the multi-channel signal as a whole, the modal correspondence and consistency of the decomposition results across channels are ensured. This method employs a specific multidimensional interpolation strategy, which, compared to standard EMD or standard MEMD, effectively reduces modal aliasing and improves the modal consistency between channels, thus providing a cleaner signal foundation for subsequent cross-frequency coupling analysis.

[0037] The specific method for obtaining the cross-frequency coupling value in step 3 above is as follows:

[0038] A Hilbert transform is performed on each intrinsic mode function (IMF) to construct an analytic signal. The instantaneous amplitude and phase of each IMF are then calculated using the analytic signal. The instantaneous phase of a low-frequency IMF is selected as the phase signal, and the instantaneous amplitude of a high-frequency IMF is selected as the amplitude signal. The coupling strength between the phase and amplitude is quantized using modulation exponentiation to obtain the cross-frequency coupling values. In this embodiment, step 3 is calculated separately for each time segment, thus obtaining the cross-frequency coupling numerical sequence for the entire recording time history.

[0039] The specific processing procedure in step 4 is as follows: Statistical analysis is performed on the cross-frequency coupling values ​​between different frequency bands within all time segments, calculating the mean and standard deviation. The baseline value is the mean plus k times the standard deviation, where k is a preset positive number. The value of k is determined based on experiments or experience; in this embodiment, k=2. This baseline value can adaptively reflect the background coupling level of the signal itself.

[0040] In this embodiment, the non-stationary signal is an electroencephalogram (EEG) signal. This EEG signal can be from a human or an animal.

[0041] The method for acquiring human EEG signals is as follows: In a quiet environment, multi-channel EEG signals are recorded from the subject in a resting state with eyes closed for a set duration (10 minutes in this embodiment). In this embodiment, the time segments corresponding to the cross-frequency coupling values ​​exceeding the baseline may correspond to micro-mental state changes that are not perceived by the subject in a resting state, such as fluctuations in attention or changes in drowsiness.

[0042] The method for acquiring EEG signals in animals is as follows: In vivo EEG recording electrodes are implanted in the brain of the experimental animal, followed by postoperative recovery; the animal is placed in a free-movement environment, and its multi-channel EEG signals are continuously recorded for a predetermined duration. In this embodiment, the duration is set at 24 hours. In this embodiment, the time segments corresponding to cross-frequency coupling values ​​exceeding the baseline may be related to the animal's behavioral state (such as sleep-wake cycles).

[0043] This embodiment also relates to a computer-readable storage medium storing a computer program that, when executed by a processor, performs the coupling feature extraction method for multi-channel non-stationary signals as described above.

[0044] 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 technical principles 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 method for extracting coupling features from multi-channel non-stationary signals, characterized in that... Includes the following steps: Step 1: Divide the multi-channel non-stationary signal to be tested into several time segments according to the preset duration to obtain the multi-channel non-stationary signal of each time segment; Step 2: Decompose the multi-channel non-stationary signal of each time segment into multiple intrinsic mode functions to obtain multiple intrinsic mode functions for each time segment; Step 3: Analyze the phase-amplitude coupling between different frequency bands using multiple intrinsic mode functions of each time segment to obtain the cross-frequency coupling values ​​between different frequency bands within each time segment; Step 4: Calculate the global distribution of cross-frequency coupling values ​​between different frequency bands within all time segments, and set the baseline value based on the global distribution; Step 5: Compare the cross-frequency coupling values ​​between different frequency bands within each time segment with the baseline values. The cross-frequency coupling values ​​that exceed the baseline values ​​are taken as the final frequency domain features. The final frequency domain features are the extracted coupling features.

2. The coupling feature extraction method according to claim 1, characterized in that: The method for obtaining multiple intrinsic mode functions for a certain time segment in step 2 is as follows: Using the number of channels of the non-stationary signal as the number of projection directions, project the signal along each projection direction and calculate the local extremum point of each projected signal; Subsequently, the local extrema of each projected signal are mapped back to the original multi-channel space to generate multidimensional envelopes in each projection direction; Calculate the mean of the multidimensional envelopes in all projection directions, and subtract the mean of the multidimensional envelopes in all projection directions from the original multi-channel signal of the current time segment to obtain the first intrinsic mode function. Subtract the first intrinsic mode function from the original multi-channel signal of the current time segment to obtain the new multi-channel signal of the current time segment. The second, third, ..., Nth intrinsic mode functions can be obtained in the same way as above, where N is the total number of intrinsic mode functions.

3. The coupling feature extraction method according to claim 2, characterized in that: Multidimensional envelopes for each projection direction are generated using a multidimensional interpolation method based on radial basis functions.

4. The coupling feature extraction method according to claim 1, characterized in that: The specific method for obtaining the cross-frequency coupling value in step 3 is as follows: The instantaneous amplitude and instantaneous phase of each intrinsic mode function are obtained. The instantaneous phase corresponding to one low-frequency intrinsic mode function is selected as the phase signal, and the instantaneous amplitude corresponding to one high-frequency intrinsic mode function is selected as the amplitude signal. The coupling strength between the phase and amplitude is quantized by modulation exponent to obtain the cross-frequency coupling value.

5. The coupling feature extraction method according to claim 4, characterized in that: Perform a Hilbert transform on each intrinsic mode function to construct an analytic signal; use the analytic signal to solve for the instantaneous amplitude and instantaneous phase of each intrinsic mode function.

6. The coupling feature extraction method according to claim 1, characterized in that: The specific processing procedure in step 4 is as follows: statistically analyze the cross-frequency coupling values ​​between different frequency bands in all time segments, calculate the mean and standard deviation, and the baseline value is the mean plus k times the standard deviation, where k is a preset positive number.

7. The coupling feature extraction method according to any one of claims 1 to 6, characterized in that: The non-stationary signal is an electroencephalogram (EEG) signal.

Images