Research On The Extraction And Classification Of EEG In Brain-Computer Interface

Over the past decade, there has been great interest and a rapid development in the research of brain-computer interface technology. BCI can provide a new communication option that does not depend on peripheral nerves and muscles for those with neuromuscular impairments, and might also give those without disabilities a supplementary control channel or a control channel useful in special circumstances. Signal analysis and processing is a key issue for implementation of a BCI system as well as improving its reliability and performance, and which is the main contents in the paper.First, a summary of the relating concepts of EEG and its statistics nature was given in chapter 2, which are the fundamentals for farther discussion. Then, the extraction and classification methods discussed at emphasis. The main research work of this paper focuses on the application of wavelet transform technique in EEG analysis, which could be summarized as the following three aspects. (a) In chapter 3 a novel threshold is proposed for wavelet threshold de-noising method. First,hard threshold, soft threshold and the proposed threshold are used for de-noising and decomposing of the EEG data. And then the wavelet coefficient is used as extracted feature set and is fed to a probabilistic neural network classifier to organize the EEG signals into different activities. It is shown that the novel threshold is better than the last two for EEG signal de-noising based on wavelet transform. (b) Through research the 60 channels EP amplitude variation rule in different times of left-right hand imagine movement tasks, the feasibility of using the brain hot space remove track as the feature of the imagine movement EEG is discussed. At the same time, a new method--barycentric method is given in the paper. It proves that there are different centroids in different electroencephalogram, and this method can be used in assistant classification. (c) The basic method is given in chapter 4, and the superposed mean value algorithm is used in EP preprocessing, which can improve the signal-to- noise ratio (SNR) to the extent, and improve the classified result. Then, based on the result of the encephalogram analysis experiment, the 4 channels EEG are accessed to the extraction and classification, which are C3,C4,P3,P4, and wavelet transform and probabilistic neural network are used for testing. It can be approved that processing on 4 channels EEG is better than 2 channels EEG. So increasing the channels accessing in appropriate processing order of complexity is acceptable.