Improved electromyographic signal nonlinear transformation feature extraction method

By correlating the time domain derivative and frequency domain signal of the electromyography signal through multi-order power spectrum and Fourier transform, a new feature vector is constructed, which solves the problem of inaccurate extraction of electromyography signal characteristics in the existing technology and improves the pattern recognition of electromyography signal accuracy and real-time.

CN118902479APending Publication Date: 2024-11-08HANGZHOU DIANZI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410939834.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Existing electromyography signal feature extraction methods usually regard electromyography signals as stationary signals, making it difficult to accurately describe muscle activity characteristics, especially in the recognition of actions under different forces or speeds, which affects the performance of action recognition.

Method used

Using multi-order power spectrum and Fourier transform, the signal time domain derivative is correlated with the frequency domain signal, the channel correlation coefficient is calculated through the sliding window, and discrete Fourier transform is carried out to build a new derivative of the time domain signal and the frequency domain signal. Eigenvectors, extract the amplitude changes of even-order power spectral density and p-order correlation coefficients, and are used for pattern classification.

Benefits of technology

It improves the sensitivity, specificity and accuracy of EMG signal feature extraction, reduces computing time and memory requirements, is suitable for embedded systems, and improves the accuracy and response speed of EMG signal pattern recognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure BDA0004943704320000031
    Figure BDA0004943704320000031
  • Figure BDA0004943704320000032
    Figure BDA0004943704320000032
  • Figure BDA0004943704320000042
    Figure BDA0004943704320000042
Patent Text Reader

Abstract

The invention relates to the technical field of electromyographic signal processing, in particular to an improved electromyographic signal nonlinear transformation feature extraction method which comprises the following steps: acquiring an electromyographic signal through a sensor, and converting the electromyographic signal into a digital signal to obtain an electromyographic signal time sequence; calculating an inter-channel correlation coefficient by adopting a sliding window to obtain a correlation coefficient vector; carrying out discrete Fourier transform on the obtained correlation coefficient vector to obtain a frequency domain discrete value to represent the square of a correlation function; associating the derivative of the time domain signal with the frequency domain signal; calculating the vector even order power spectrum density of each correlation coefficient; the amplitude change of each p-order correlation coefficient is calculated; and constructing a characteristic vector based on the even-order power spectrum density of each correlation coefficient vector and the amplitude change of the p-order correlation coefficient. The method disclosed by the invention is remarkably improved in the aspects of sensitivity, specificity, accuracy and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of electromyographic signal processing, in particular to an improved electromyographic signal nonlinear transformation feature extraction method. Background Art

[0002] Electromyographic signals are bioelectric signals of human action potentials superimposed on the surface of the skin. Electromyographic signals can be used to infer muscle activity and determine movement-related information of human body parts. Electromyographic signals are widely used in virtual reality and medical rehabilitation. Among them, electromyographic signal feature extraction is a key link in determining the movement state of body parts. Existing time domain or frequency domain feature extraction methods usually regard electromyographic signals as stationary signals or piecewise stationary signals. Even for the same posture action, the signal feature value may change under different forces or speeds, making it difficult to accurately describe the electromyographic signal characteristics, which will affect the subsequent action recognition performance. Summary of the invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and to propose an improved method for extracting nonlinear features of electromyographic signals, which has strong robustness and real-time performance.

[0004] In order to achieve the above object, the technical solution specifically adopted by the present invention is as follows:

[0005] An improved method for extracting nonlinear transformation features of electromyographic signals uses multi-order power spectra combined with Fourier transform to associate the signal time domain derivative with the frequency domain signal. Define an even r-order power spectrum and p-order amplitude transform to extract n=r+p features for pattern classification, which specifically includes the following steps:

[0006] Step S1: Collect electromyographic signals through sensors and convert them into digital signals, i.e., electromyographic signal time series;

[0007] Step S2: Using a sliding window, calculate the correlation coefficient between channels to obtain M-1 correlation coefficient vectors;

[0008] Step S3: performing discrete Fourier transform on the obtained M-1 correlation coefficient vectors to obtain frequency domain discrete values ​​to represent the square of the correlation function;

[0009] Step S4: Connect the derivative of the time domain signal with the frequency domain signal to form a new feature vector:

[0010] Step S5: Calculate the even-order power spectral density P0,…,P of each correlation coefficient vector r-1 ;

[0011] Step S6: Calculate the amplitude variation of each p-order correlation coefficient D0,…,D p-1 ;

[0012] Step S6: construct a feature vector V based on the even-order power spectrum density of each correlation coefficient vector and the p-order correlation coefficient amplitude change:

[0013] V=[P0,…,P r-1 ,D0,…,D p-1 ]

[0014] In the formula, P0 and P r-1 is the power spectral density. D0 and D p-1 is the correlation coefficient

[0015] Furthermore, in step S1, the electromyographic signal time series S j for:

[0016] S j =s[i],0≤i≤N-1,0≤j≤M-1 (1)

[0017] Where N is the number of sampling points, j is the signal channel index, there are M signal channels in total, and s[i] is the signal component of the EMG signal time series.

[0018] Furthermore, in step S2, the inter-channel correlation coefficient adopts the Pearson correlation coefficient or the Spearman correlation coefficient, where:

[0019] C t,q =c[k],0≤k≤N-1 (2)

[0020] In the formula, C t,q is the correlation coefficient between channels t and q, c[k] represents the correlation coefficient component between channels; N represents the number of sampling points.

[0021] Furthermore, in step S3, the frequency domain discrete value is used to represent the square of the correlation function:

[0022]

[0023] In the formula, f k is the frequency discrete value, T * [f k ] is T[f k ] conjugate, is the signal power.

[0024] Furthermore, in step S4, the relationship between the derivative of the time domain signal and the frequency domain signal is obtained as follows:

[0025] F[Δ l c[k]]=f k l T[f k ] (4)

[0026] Where F is the discrete Fourier transform operator and l is the order.

[0027] Furthermore, in step S5, the even-order power spectrum density of each correlation coefficient vector is calculated by the following formula:

[0028]

[0029] Where P is the power spectral density.

[0030] Furthermore, in step S6, the amplitude change of each p-order correlation coefficient is calculated by the following formula:

[0031]

[0032] Compared with the existing feature extraction methods, the present invention has the following characteristics and beneficial effects:

[0033] 1) The method of the present invention has significant improvements in sensitivity, specificity, accuracy, etc.

[0034] 2) The method of the present invention requires relatively less computing time and memory, is suitable for embedded system development, and has good real-time performance.

[0035] 3) The nonlinear transformation of the power spectrum moment, amplitude change rate and amplitude mean in the method of the present invention helps to improve the classification ability, and the evaluation of the inter-channel correlation coefficient helps to improve the accuracy of electromyographic signal pattern recognition.

[0036] 4) The method of the present invention helps to improve the response speed of the electromyographic acquisition electronic system. DETAILED DESCRIPTION

[0037] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0038] An improved method for extracting nonlinear transformation features of electromyographic signals according to an embodiment of the present invention comprises the following steps:

[0039] Step S1: Collect electromyographic signals through sensors and convert them into digital signals, which are electromyographic signal time series:

[0040] S j =s[i],0≤i≤N-1,0≤j≤M-1 (1)

[0041] Where N is the number of sampling points, and j is the signal channel index (a total of M signal channels).

[0042] Step S2: Use a sliding window (window width is W) to calculate the correlation coefficient between channels and obtain M-1 correlation coefficient vectors with a length of R;

[0043] C t,q =c[k],0≤k≤R-1 (2)

[0044] Among them, C t,q is the correlation coefficient between channels t and q, which can be Pearson correlation coefficient or Spearman correlation coefficient; the selection of window width W depends on the specific situation. R is the number of dynamic correlation matrices.

[0045] Step S3: Perform discrete Fourier transform on the M-1 correlation coefficient vectors to obtain frequency domain discrete values, which further represent the square of the correlation function:

[0046]

[0047] Among them, f k is the frequency discrete value, T * [f k ] is T[f k ] conjugation;

[0048] Step S4: Relate the derivative of the time domain signal to the frequency domain signal:

[0049] F[Δ l c[k]]=f k l T[f k ] (4)

[0050] Where F is the discrete Fourier transform operator and l is the order.

[0051] Step S5: Obtain the even-order power spectrum density of each correlation coefficient vector according to equations (2) and (3), where:

[0052]

[0053] Where P is the power spectral density.

[0054] Step S6: Calculate the amplitude change of the p-order correlation coefficient according to formula (5):

[0055]

[0056] Step S7: construct a feature vector V based on the even-order power spectrum density of each correlation coefficient vector and the p-order correlation coefficient amplitude change:

[0057] V=[P0,…,P r-1 ,D0,…,D p-1 ] (7)

[0058] Afterwards, the obtained feature dimension representation vector V can be input into the classifier for pattern recognition.

[0059] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An improved method for extracting nonlinear features of electromyographic signals, characterized in that: The steps include: Step S1: Collect electromyographic signals through sensors and convert them into digital signals, i.e., electromyographic signal time series; Step S2: Using a sliding window, calculate the correlation coefficient between channels to obtain M-1 correlation coefficient vectors; Step S3: performing discrete Fourier transform on the obtained M-1 correlation coefficient vectors to obtain frequency domain discrete values ​​to represent the square of the correlation function; Step S4: Connect the derivative of the time domain signal with the frequency domain signal to construct a new vector: Step S5: Calculate the even-order power spectral density P0,…,P of each correlation coefficient vector r-1 ; Step S6: Calculate the amplitude variation of each p-order correlation coefficient D0,…,D p-1 ; Step S6: construct a feature vector V based on the even-order power spectrum density of each correlation coefficient vector and the p-order correlation coefficient amplitude change: V=[P0,…,P r-1 ,D0,…,D p-1 ] In the formula, P0 and P r-1 is the power spectral density; D0 and D p-1 is the correlation coefficient.

2. The improved method for extracting nonlinear features of electromyographic signals according to claim 1, characterized in that: In step S1, the electromyographic signal time series S j for: S j =s[i],0≤i≤N-1,0≤j≤M-1 (1) Where N is the number of sampling points, j is the signal channel index, there are M signal channels in total, and d[i] is the signal component of the electromyographic signal time series.

3. The improved method for extracting nonlinear features of electromyographic signals according to claim 1, characterized in that: In step S2, the inter-channel correlation coefficient adopts the Pearson correlation coefficient or the Spearman correlation coefficient, where: C t,q =c[k],0≤k≤R-1 (2) In the formula, C t,q is the correlation coefficient between channels t and q, c[k] represents the channel correlation coefficient; R is the number of dynamic correlation matrices.

4. The improved method for extracting nonlinear features of electromyographic signals according to claim 1, characterized in that: In step S3, the frequency domain discrete value is used to represent the square of the correlation function: In the formula, f k is the frequency discrete value, T * [f k ] is T[f k ] conjugate, is the signal power.

5. The improved method for extracting nonlinear features of electromyographic signals according to claim 1, characterized in that: In step S4, the relationship between the derivative of the time domain signal and the frequency domain signal is obtained as follows: F[Δ l c[k]]=f k l T[f k ] (4) Where F is the discrete Fourier transform operator and l is the order.

6. The improved method for extracting nonlinear features of electromyographic signals according to claim 1, characterized in that: In step S5, the even-order power spectrum density of each correlation coefficient vector is calculated by the following formula: In the formula, r-1 is an even number (0, 2, 4, 6, 8, ...).

7. The improved method for extracting nonlinear features of electromyographic signals according to claim 1, characterized in that: In step S6, the amplitude change of each p-order correlation coefficient is calculated by the following formula: