Study On Feature Extraction Of Motor Imagery Based Brain Computer Interface

Brain computer interface (BCI) is a kind of communication system that cancombine brain with external environment, without the brain’s normal output pathwaysconsisted of peripheral nerves and muscles. With the development of science andtechnology, BCI becomes a research hotspot in the world, because of its broadapplication prospects and great theoretical research value. The BCI based on theevent-related desynchronization or synchronization (ERD/ERS) of motor imageryelectroencephalogram (EEG) is an important research direction, and also the focus ofthis paper. It is the core link of BCI to extract features from nonlinear andnon-stationary EEG signals, and Hilbert-Huang (HHT) is a new analysis method, whichis suitable for processing non-linear and non-stationary signal, so the feature extractionbased on HHT for motor imagery EEG is the main content of this paper.In this paper, the EEG acquisition program is designed to acquire EEG of left handand right hand motor imageries. Firstly this paper introduces the structure of the brainand the principle of EEG signals. Then based on the characteristics of the motorimagery EEG signals and the EEG acquisition system, we designed the experimentalparadigms for motor imagery EEG and acquired the EEG data of five subjects. Thesedata were used in the following research.In this paper, we also study on the energy feature extraction method based on theprinciple, algorithm and the time-frequency characteristics of Hilbert-Huang transform.This method can extract energy features considered time and frequency factor becauseof the good time-frequency properties of HHT. In order to verify the validity andpracticability of the method, we extracted features from the data of the BCI competition2003and Independent acquisition by the three methods AR model, Band power (BP)and HBE, and these features were classified by the Fisher linear discriminant analysis.The experimental results show that HBE feature extraction method gets higherclassification accuracy than BP and AR feature extraction methods.At last, we study on the synchronous features of motor imagery EEG. This paperintroduces the method of phase synchronization feature extraction named phase lockingvalue (PLV), and then on the basis of PLV, a novel synchronization feature extractionmethod based on Ensemble Empirical Mode Decomposition (EEMD) named frequency locking value (FLV) is proposed. FLV feature extraction method can measure thechange of instantaneous frequency between different electrodes or an electrode itself. Atlast, the data of BCI competition2008and independent data were used to verify theeffectiveness of the proposed feature extraction method, the experimental results showthat the FLV feature extraction method based on EEMD has higher classificationaccuracy than PLV feature extraction methods, and the synchronization characteristicscombined with the characteristics of energy feature could improve the classificationaccuracies of left and right-hand motor imagery obviously.This paper introduces the Hilbert-Huang transform into motor imagery EEG, andnovel energy and synchronization feature extraction methods are proposed. Theseproposed feature extraction methods obtain good classification results. The research ofthis paper provides new methods and thoughts to the study of BCI.