Methods Of Motor Imagery Signal Analysis Based On Bayesian Theory

A Brain Computer Interface (BCI) is a communication way that messages orcommands which an individual sends to the external world do not pass through the brain’snormal output pathways (peripheral nerves and muscles). Hence, Electroencephalogram（EEG）is of great significance for further study of the brain function and the mechanism ofbrain function. Due to the complexity of the structure and signal acquisition for humanbrain, the EEG data tend to have some peculiarities, such as high dimension and lowsignal-to-noise ratio, which increase the difficulty on our data classification and recognition.Because of the non-determinacy of EEG data, the method which is based on probabilityrepresentation usually can obtain better results.In this paper, a dynamical model based on the Bayesian theory will be established andused in parameter estimation and analysis of EEG data. First of all, we assume that the EEGdata conform to a kind of spatial-temporal stochastic model, so that a correspondingBayesian model of parameter estimation can be established based on the model hypothesis.Then, by using Maximum Likelihood (ML) and Maximum A Posteriori (MAP), theBayesian model can reveals its internal relationship in the hidden states. Because of thedissimilarity of state sequences from different observed sequences, we can clearly recognizeand classify the EEG data. In this paper，we will study several kind of dynamic stochasticmodels based on the Bayesian theory, such as Hidden Markov Model (HMM), SwitchingLinear Dynamical System (SLDS).Finally, the method based on Bayesian theory was used in Motor Imagery (MI) signalin the paper. Compared with the traditional approaches, the novel way greatly improved theaccuracy of MI data classification and recognition.