Research On Meg-Based Brain Computer Interface

Brain-computer interfaces (BCI) is a new communication channel, which do not depend on the brain's normal output pathways of peripheral nerves and muscles. It can be translated the predefined physiological parameters of brain into BCI commands. The Magnetoencephalography (MEG) does not require the reference point and does not need contact with the skin in the measurement process. The MEG can directly reflect the state of brain activity field source, therefore research which is based on MEG brain-computer interface system has very important significance.The research object is the key technology in brain computer interface in this paper. In order to improve the signal to noise ratio,the Unscented Kalman Filter based on Auto Regressive Model is proposed,selecting the appropriate model order and noise parameters, the experimental results are compared with the wavelet denoising.The results show that the average rate of this algorithm is relatively high, it reached to 54.35%, which is better than the results of Wavelet Denoising,53.0%, it is also better than the recognition rate 46.9% in the BCI competitionⅣ. This shows that the time-domain filtering can be a good pretreatment for MEG signal.Sencondly,feature extraction algorithm is the most important factor in the BCI, Two algorithms are proposed. One is an of the empirical mode decomposition (EMD)-based AR model.The other is the principal component analysis (PCA) and linear discriminant analysis (LDA) method. these two algorithms have achieved a good recognition results, providing the main feature.Thirdly, machine learning algorithm is also another important aspect in BCI. In view of the simi-supervised clustering combined the advantages of training data prior knowledge, semi-supervies fuzzy clustering algorithm based on training center is put forward, the experiment results show that this algorithm has a higher recognition rate.Finally, from the above algorithms, the frequency compoents may differ for the different subject. Supposed that the MEG signal is composed by non-Gaussian noise and harmonic signal, the frequency components have been estimated.It found that the frequency of each channel composition is very similar, the stable regional of frequency is differences for the experimenter.