Complex signal de-noising method with empirical mode decomposition (EMD) and dictionary learning combined

Abstract

The invention discloses a complex signal de-noising method with empirical mode decomposition (EMD) and dictionary learning combined. Signals with noise are subjected to empirical mode decomposition (EMD) to obtain a group of intrinsic mode function signals with orders from low to high; a wavelet soft threshold de-noising method is adopted to de-noise the intrinsic mode function with noise; the intrinsic mode function after de-noising and other intrinsic mode functions are added to obtain de-noising signals, and a noise component is obtained through residual addition; the de-noising signals are segmented to a group of training samples, and a KSVD algorithm is adopted to train a group of signal dictionaries; the noise component is segmented to a group of training samples, and the KSVD algorithm is adopted to train a group of noise dictionaries; the signal dictionaries and the noise dictionaries are combined to obtain a mixed dictionary; signals which need de-noising are subjected to sparse decomposition in the mixed dictionary, a group of sparse vectors is obtained, and coefficients corresponding to noise atoms in the sparse vectors are set to be zero; and the mixed dictionary and the processed sparse vectors are multiplied to obtain final de-noising signals. Thus, multiple noise components can be effectively removed.

Description

technical field [0001] The invention relates to a complex signal denoising method. In particular, it involves a complex signal denoising method based on empirical mode decomposition combined with dictionary learning. Background technique [0002] In many fields involving data processing, such as mechanical vibration analysis, image and voice recognition, and meteorological data interpretation, we all face a common problem: different types of noise signals are inevitably mixed into the data during the collection and transmission process, which will affect the later stage. Data processing brings great difficulties, and the validity and accuracy of the processing results are affected. Therefore, how to eliminate or suppress noise and restore useful signal components from polluted signals is a very valuable research topic, and many researchers are devoted to the research of this topic. [0003] After decades of unremitting efforts by scholars in various fields, signal denoisin...

AI Technical Summary

Problems solved by technology

The algorithm generally needs to know multiple signals, but the actual signal often has only one noisy signal, which leads to the limitation of the denoising method

[0004] Due to the nonlinear and non-stationary characteristics of many complex signals, it is difficult to use regular basis functions to represent them, and the type of noise contained in them cannot be predicted in advance, which makes it difficult for existing denoising algorithms to obtain satisfactory denoising results

Images