A Research On The Feature Extraction And Analysis Of EEG Signals Based On Harmonic Wavelet Transform

A wealth of brain information is provided in EEG signals. Some characteristic parameters can be extracted for representing different cerebral function states by careful analyses and processing. Electroencephalogram (EEG) has been widely used in neurophysiology researches and clinical disease diagnosis, because it is safe, painless, economical and convenient. EEG plays an important role in some aspects, for example, estimating on the human mental statements, diagnosing and treating brain and neurological diseases, monitoring the depth of anesthesia, analyzing sleep quality and stages, and so on.In order to provide some new assistant information for automatic identification and diagnosis of clinical EEG signals, and offer valuable theoretic reference for EEG diagnosis and monitoring instrument with excellent performance. The feature extraction about time-frequency distribution and basic rhythms changes is studied in this paper, introducing a modern signal processing method, harmonic wavelet transform. The main research work of this dissertation has three parts as follows:First, the basic knowledge on EEG signals is summarized, that is, generation mechanism, data collection and contents in frequency band are stated, the development history stages is reviewed, the significance of EEG is summed up, and in this research field the development status and trend in recent years at home and abroad are reviewed. Second, the conventional Fourier transform fails to display local information of EEGsignals both in time and frequency domain, which are typically time-varying and non-stationary. The EEG signals or spectrum characteristic at different time reflect different cerebral function states directly. Time-frequency analysis is just effective for representing the characteristics of Electroencephalogram (EEG) records. Harmonic wavelet introduced in this paper is a powerful tool for time-frequency analysis, and is especially suited for analyzing and characterizing EEG, because of its excellent capability of filtering perfectly and locating frequency bands accurately. 3D time-frequency plots and the corresponding contour plots of clinical data from different patients are obtained by discrete harmonic wavelet transform. The simulated results indicate that time-frequency representations based on harmonic wavelet transform improve the limitation of traditional spectrum analysis and are helpful for doctors to analyze EEG more comprehensively, objectively, and with less dependence on their individual opinions; and that they are of great clinical application value.