Useful signal identification method based on NA-MEMD and GMM clustering

Abstract

The invention discloses a useful signal identification method based on NA-MEMD and GMM clustering. According to the existing empirical mode decomposition method, how to accurately identify IMF components containing useful information at each scale after a multi-variable signal is decomposed into intrinsic mode function components at a plurality of scales remains dependent on the prior knowledge, and the identification rate is relatively low. According to the method of the invention, first, a multi-variable signal is decomposed by use of an NA-MEMD algorithm to get IMF components at different scales; then, the IMF components at all scales are mapped to a low-dimensional subspace by use of a spectral regression dimension reduction algorithm to extract corresponding low-dimensional feature vectors; next, a clustering analysis is made of the low-dimensional feature vectors at all scales by use of a GMM clustering algorithm; and finally, IMF components containing useful information are identified according to the clustering result. The method has a broad application prospect in electroencephalogram signal processing and in nerve data analysis.

Description

technical field [0001] The invention belongs to the field of electroencephalogram signal processing, and relates to a method for identifying useful components of electroencephalogram signals, in particular to a method for identifying useful signal components based on noise-assisted multivariate empirical mode decomposition and Gaussian mixture model clustering. Background technique [0002] Electroencephalogram (Electroencephalogram, EEG) is a comprehensive reflection of the activity of nerve cells in the brain in the cerebral cortex, and contains information related to brain conditions and thinking processes. Because non-implantable EEG is relatively simple, fast, harmless to people, and has high time resolution, it has become one of the most important signal acquisition methods in EEG signal processing and application. However, EEG is acquired through scalp electrodes, and the signal is very weak and the background noise is strong. Therefore, it is of great significance to...

AI Technical Summary

Problems solved by technology

However, both EEMD and EMD can only handle one-dimensional signals and are not suitable for analyzing multi-channel data

In 2010, Rehman et al. improved the classic EMD algorithm and proposed a multivariate empirical mode decomposition (Multivariate EMD, MEMD) algorithm, which can decompose multi-channel data at the same time, and can avoid the intrinsic mode function (Intrinsic Mode Functions, IMFs) number, frequency mismatch problem, but there is still the problem of mode aliasing

[0005] In summary, after using EMD and its improved method to decompose IMF components on several scales from multi-channel EEG signals, how to accurately identify IMF components containing useful information on each scale has not been effectively solved. solve

Images