Wavelet Transforms for Eeg Signal Denoising and Decomposition

Abstract

EEG signal analysis is difficult because there are so many unwanted impulses from non-cerebral sources. Presently, methods for eliminating noise through selective frequency filtering are afflicted with a notable deprivation of EEG information. Therefore, even if the noise is decreased, the signal's uniqueness should be preserved, and decomposition of the signal should be more accurate for feature extraction in order to facilitate the classification of diseases. This step makes the diagnosis faster. In this study, three types of wavelet transforms were applied: Discrete Wavelet Transform (DWT), Wavelet Packet Transform (WPT), and Stationary Wavelet Transform (SWT), with three mother functions: Haar, Symlet2, and Coiflet2. Three parameters were used to evaluate the performance: Signal-to-Noise Ratio (SNR), Mean Square Error (MSE), and Peak Signal-to-Noise Ratio (PSNR). Most of the higher values of SNR and PSNR were 27.3189 and 40.019, respectively, and the lowest value of MSE was 5.0853 when using Symlet2-SWT level four. To decompose the signal, we relied on the best filter used in the denoising process and applied four methods: DWT, Maximal Overlap DWTs (MODWT), Empirical Mode Decomposition (EMD), and Variational Mode Decomposition (VMD). The comparison has been made between the four methods based on three metrics: energy, correlation coefficient, and distances between the Power Spectral Density (PSD), where the highest value of energy was 5.09E+08 and the lowest value of the PSD was -1.2596 when using EMD.