Research On Feature Extration Methords In The Brain-Computer System

The BCI(Brain-computer interfaces, BCI) system is such system that can realize the recognition of mental task which is to express the patient's ideas according to the sampling EEG(Electroencephalogram, EEG) signals. Many neuromuscular disorders can disrupt the channels through which the brain communicates with and controls its external environment, which usually deprive patients of their basic movement function and normal communication. Seriously, those information expressing their ideas may be completely locked into their bodies. BCI is a direct information communication and control channel established between human and outside world, and it is a wholly new communication system that does not depend on the brain's normal output pathways of peripheral nerves and muscles. This paper focus on some topics described as following, which is based on the summarizer of present domestic and overseas research actuality.(1) Feature extraction of the EEG based on frequency-space filterAfter analyzing the origin, the mechanism and the framework of the EEG, we preprocess it by the method of frequency-space filter and use power spectral density, AR model parameter to extract the frequency band power as the features of the EEG recorded during imagination of hand movements. Moreover, we compared the performance of these features with others.(2) Feature distribution of EEG in time domainBP (Bereitschaftspotentia, BP) is one of the main components of MRCP (movement-related cortical potentials, MRCPs), which slowly decreases before voluntary limb movement. Moreover, the amplitude difference of some different electrodes is obvious. Contralateral dominance is another significant character of BP. Therefore, we study the feature distribution in time domain and get the better result compared with other feature extraction methods.(3) BP neural network algorithmThe key characteristic of neural network in pattern recognition is to solving complex nonlinear problems. Facing the recognition of mental task based on complex EEG, we get the best result by adjusting parameters and choosing algorithms of BP neural network. Certainly, the result is only based on the Berlin data set for BCI competition 2003.(4) Neural network assembling algorithmGenerally, there are a lot of characters with different properties to express or discribe the patterns. It is difficult to contain all characters for a single classifier because each character vector performs differently. We try to assemble different sub-nerual networks which correspond to their character subspace.