CN104515905B - The EEG signals adaptive spectrum analysis method of subject based on CQT multiresolution - Google Patents

Abstract

The invention discloses a CQT-based multi-resolution adaptive spectrum analysis method of the subject's EEG signal, which sequentially preprocesses the original EEG signal and automatically finds the subject through the harmonic components and fine structure characteristics of the EEG signal. The difference, CQT-based multi-resolution analysis and calculation of the sampling bandwidth of each frequency band. The present invention can adaptively find out the different characteristics of the subjects according to the original brain wave signal, and can more accurately extract the spectrum-specific features of the EEG signal; based on the multi-resolution spectrum analysis method, the influence of the difference in the frequency band length of the EEG signal is considered , which improves the practicability; compared with the classic CQT spectrum analysis method, the number of calculations is reduced, and the efficiency is significantly improved; the variable resolution used in the present invention is much more flexible, and the accuracy and speed of feature extraction are improved.

Description

Adaptive Spectrum Analysis Method of Subject's EEG Signal Based on CQT Multi-resolution

technical field

The invention relates to the field of brain-computer interface and a method for analyzing brain wave signals, in particular to a CQT multi-resolution-based adaptive spectrum analysis method for brain signals of subjects.

Background technique

With the deepening of research on brain-computer interface technology, more and more methods for analyzing EEG signals have emerged. Although the brain-computer interface at the present stage has realized a part of the control function, this kind of control is still relatively primitive and primary control, and there are still many problems to be solved. The goal of the BCI system is to achieve a direct and natural mode of control through EEG. To achieve this goal, the key is to collect the characteristic EEG signal patterns that can reflect the thinking intention of the brain, and realize the decoding from thinking to control through the data processing module. However, due to the low signal-to-noise ratio and non-stationary random characteristics of the brain wave signal, even for the same subject, there may be some differences in the induced brain wave signal. Aiming at the difference between subjects, it is of great significance to find effective adaptive brain wave signal difference recognition technology.

At present, the processing methods of EEG signals are mainly based on time domain analysis and frequency domain analysis. Due to the sequential nature of EEG signals, most of the existing EEG signals use time-domain analysis methods, such as zero-crossing analysis, histogram analysis, variance analysis, peak detection and other methods. Time-domain analysis is mainly to directly extract waveform features, which is intuitive and has clear physical meaning. However, the results of point analysis in the time domain often depend on the distribution of sampling points, and the sampling of EEG signals is relatively scattered, so it is often difficult to obtain a good discrimination effect. Therefore, many scholars convert the time-domain EEG signal to the frequency domain, and extract its frequency-domain features for analysis and identification. For example, Barry and Chen found the difference between eyes closed (Eyes Closed, EC) and eyes open (Eyes Open, EO) using EEG spectrum analysis technology respectively. In engineering, frequency spectrum estimation is used to extract EEG spectral features, distinguish different sensory, motor or cognitive activities, and realize brain-computer interface (brain-computer interface, BCI). The famous Wadsworth BCI controls the movement of the cursor by calculating the mu and beta frequency band characteristics of the EEG of the sensorimotor cortex. Currently, autoregressive (AR) model spectrum estimation technology is widely used in EEG analysis. This method has high requirements on the linearity, stationarity and signal-to-noise ratio of the processed signal, so it is not suitable for the analysis and processing of long-term EEG data. In order to improve the performance of spectrum estimation, Bartlett and Welch respectively proposed a non-parametric estimation method of power spectrum based on Fourier transform. It divides the long-term data with a total length of N into M segments, and each segment has a length of L, and calculates the power spectral density of each segment and then averages them. Welch's method is improved on this basis, the data segments are allowed to overlap, and windowing is used for each segmented data to smooth the original power spectrum.

However, the existing frequency spectrum analysis is a method of obtaining a series of frequencies at equal intervals through uniform sampling at equal intervals. However, the frequency band intervals of EEG signals are not equal. For example, the frequency band division method we often use is defined as follows: δ (0.5~4Hz), θ (4~8Hz), α (8~13Hz), β (13~ 20Hz), and γ (30~50Hz). If we use the equal interval method to calculate the frequency band energy, the frequency range of the γ band is much larger than that of the δ and θ bands due to its large span. This method of dividing the frequency spectrum at equal intervals ignores the difference in frequency band length, which will inevitably cause certain errors. In addition, various cognitive signals represented by EEG signals have quite complex harmonic components and fine structures. Various background noises and different types of cognitive signals have different harmonic components and fine structures but have There is a certain regularity, and the same type of signal from different people also has its own characteristics. The current spectrum analysis methods do not take into account the differences between the subjects.

Contents of the invention

The purpose of the present invention is to propose a method for adaptive spectrum analysis of EEG signals based on CQT multi-resolution based on the deficiency of the above-mentioned equal interval division spectrum analysis method and the large difference between the subjects, so as to overcome the problem of single resolution. Insufficient frequency analysis method can effectively extract the spectrum-specific features of EEG signals.

To achieve the above object, the present invention is implemented according to the following technical solutions:

A method for adaptive spectrum analysis of EEG signals based on CQT multi-resolution, comprising the steps of:

1) Preprocessing the original EEG signal:

a.Using unilateral mastoid as the reference electrode, carry out the reference conversion of bilateral mastoid as the reference electrode for the EEG signal, assuming that the left mastoid is the M1 electrode and the right mastoid is the M2 electrode, the conversion formula is as follows:

b. Segment each dimension of the EEG signal, the segment length is 2s, and there is no overlap between the segments. If the value of any point in each segment is greater than 80μv, the entire segment is discarded to exclude eye blink artifacts and myoelectric artifacts trace interference;

2) Automatically find the differences between the subjects through the harmonic components and fine structure characteristics of the EEG signal:

a. First assume that the EEG signal of the subject is dimension (m is the number of electrodes, n is the number of samples), from the signal Randomly take out one dimension, and take out a section of length L Calculate the average power spectrum P _l and frequency sequence f _l , the calculation formula is as follows:

where the frequency sequence

b. Use the spline interpolation method to improve the frequency resolution on the basis of f _l , and divide the EEG signal into 5 segments, δ(0.5～4Hz), θ(4～8Hz), α(8～13Hz), β( 13～20Hz), and γ(30～50Hz);

c. Find the position of the frequency when the maximum value of the frequency energy of each segment is used as the difference characteristic of the subjects;

3) Analyze and calculate the sampling bandwidth of each frequency band:

a. Estimate the center frequency based on CQT, the calculation formula is as follows:

In the formula, N _k is the corresponding window length when calculating the CQ transform of the kth frequency f _k , wN _k (n) is a window function with length N _k , Q is a constant factor in the CQ transform, and k is the sequence CQ spectrum The frequency subscript of ;

b. Calculate the interval (bandwidth) B _k of adjacent spectral lines, the calculation formula is B _k =f _k+1 -f _k ;

c. Calculate the sampling bandwidth N _k of each frequency band, and the calculation formula is N _k =Fs/B _k .

Compared with prior art, the beneficial effect of the present invention is:

1. It can adaptively find the different characteristics of the subjects according to the original brain wave signal, and can more accurately extract the spectrum-specific features of the EEG signal;

2. Based on the multi-resolution spectrum analysis method, the influence of the difference in the frequency band length of the EEG signal is considered, which improves the practicability. Traditional digital signal processing methods usually use single-resolution spectrum analysis methods such as Fourier transform. Due to the non-equal interval division of EEG signal frequency bands, some characteristics of the signal will be lost after processing, resulting in high error rates and misidentification rates;

3. Compared with the classic CQT spectrum analysis method, the number of calculations is reduced, and the efficiency is significantly improved. The classic CQT method follows the exponential distribution law, so the number of frequency points in the low-frequency region is much larger than the number of frequency points in the high-frequency region during the calculation process, which cannot represent the characteristics of the brain wave signal well. The variable resolution adopted in the present invention is much more flexible, which improves the accuracy and speed of feature extraction.

Description of drawings

Fig. 1 is a flowchart of the present invention;

Fig. 2 is a schematic diagram of the EEG signal after preprocessing in the present invention;

Fig. 3 is a schematic diagram of extracting the spectral features of the brain wave signal using a single-resolution equal interval spectrum analysis method;

Fig. 4 is the traditional CQT analysis method extracting the schematic diagram of spectrum feature of brain wave signal;

Fig. 5 is a schematic diagram of the spectral features of the electroencephalogram signal extracted by the analysis method of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. The schematic embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention.

As shown in Figure 1 and Figure 2, the EEG signal adaptive spectrum analysis method based on CQT multi-resolution, comprises the following steps:

1) Preprocessing the original EEG signal:

a.Using unilateral mastoid as the reference electrode, carry out the reference conversion of bilateral mastoid as the reference electrode for the EEG signal, assuming that the left mastoid is the M1 electrode and the right mastoid is the M2 electrode, the conversion formula is as follows:

b. Segment each dimension of the EEG signal, the segment length is 2s, and there is no overlap between the segments. If the value of any point in each segment is greater than 80μv, the entire segment is discarded to exclude eye blink artifacts and myoelectric artifacts trace interference;

2) Automatically find the differences between the subjects through the harmonic components and fine structure characteristics of the EEG signal:

a. First assume that the EEG signal of the subject is dimension (m is the number of electrodes, n is the number of samples), from the signal Randomly take out one dimension, and take out a section of length L Calculate the average power spectrum P _l and frequency sequence f _l , the calculation formula is as follows:

where the frequency sequence

b. Use the spline interpolation method to improve the frequency resolution on the basis of f _l , and divide the EEG signal into 5 segments, δ(0.5～4Hz), θ(4～8Hz), α(8～13Hz), β( 13～20Hz), and γ(30～50Hz);

c. Find the position of the frequency when the maximum value of the frequency energy of each segment is used as the difference characteristic of the subjects;

3) Analyze and calculate the sampling bandwidth of each frequency band:

a. Estimate the center frequency based on CQT, the calculation formula is as follows:

In the formula, N _k is the corresponding window length when calculating the CQ transform of the kth frequency f _k , wN _k (n) is a window function with length N _k , Q is a constant factor in the CQ transform, and k is the sequence CQ spectrum The frequency subscript of ;

b. Calculate the interval (bandwidth) B _k of adjacent spectral lines, the calculation formula is B _k =f _k+1 -f _k ;

c. Calculate the sampling bandwidth N _k of each frequency band, and the calculation formula is N _k =Fs/B _k .

Combining Fig. 3, Fig. 4 and Fig. 5, it can be seen that the analytical method of the present invention is more effective than the FFT single-resolution analytical method to obtain spectral line amplitudes close to equal amplitude, and the side lobe amplitude is smaller. The technical scheme of the present invention can find the difference characteristics of the EEG signals of the subjects in the original EEG signals, while reducing the number of calculations, it still maintains the characteristics of CQT multi-resolution, and the obtained spectral line amplitude is close to constant amplitude. The technical proposal of the present invention well overcomes the deficiency of extracting spectrum features with a single resolution, reduces the number of calculations, and can also improve the features of brain wave signals.

The technical solution of the present invention is not limited to the limitations of the above-mentioned specific embodiments, and any technical deformation made according to the technical solution of the present invention falls within the protection scope of the present invention.

Claims (1)

1. a kind of EEG signals adaptive spectrum analysis method of subject based on CQT multiresolution, it is characterised in that: including under State step:

1) EEG signal is pre-processed:

A. bilateral mastoid process is carried out as the reference of reference electrode to EEG signal and is converted as reference electrode using unilateral mastoid process, it is false If left mastoid process is M1 electrode, conversion formula is as follows:

Wherein X_n ^(m)It is EEG signal, m is number of poles, and n is number of samples；

B. it takes and is segmented per one-dimensional EEG signal, segment length 2s, without overlapping between each section, if occurred in certain section any one A numerical value is greater than the point of 80 μ v, then gives up whole section, excludes the interference of blink artefact and Muscle artifacts；

2) subject difference characteristic is automatically found by the harmonic components of EEG signal and fine structure characteristic:

A. from EEG signal X '^(m) _nIn take one section of certain one-dimensional, a length of L at random, calculate average power spectra P_lWith frequency sequence f_l, meter It is as follows to calculate formula:

Wherein, centre frequency corresponding to the l articles spectral line isL=0,1..., N-1；

B. using spline method in frequency sequence f_lOn the basis of improve frequency resolution, and above-mentioned EEG signal is divided into 5 Section: δ (0.5~4Hz), θ (4~8Hz), α (8~13Hz), β (13~20Hz) and γ (30~50Hz)；

The position of frequency is as subject difference characteristic when c. finding out each band frequency Energy maximum value；

3) analyze and calculate each frequency band samples bandwidth；

A. estimate that centre frequency, calculation formula are as follows based on CQT:

N in formula_kIt is to calculate kth frequency f_kCQT when corresponding window length,Be length be N_kWindow function, Q is Invariant in CQT, k are the frequency index of sequence C Q spectrum；

B. the interval B of adjacent spectral line is calculated_k, calculation formula B_k=f_k+1-f_k；

C. each frequency band samples bandwidth N is calculated_k, calculation formula N_k=F_s/B_k。