KERNEL BASED HIDDEN MARKOV MODEL WITH APPLICATIONS TO EEG SIGNAL CLASSIFICATION

摘要

To enhance the performance of hidden Markov models for EEG signal classification, we present here a new model referred to as kernel based hidden Markov model (KHMM). Due to the embedded HMM structure, this model is capable of capturing well the temporal change of a time-series signal. Furthermore, KHMM has better discrimination and generalization capability inherited from kernel methods. We evaluate the kernel based hidden Markov model by applying it to EEG signal classification when motor imagery is performed, yielding positive experimental results.