Feature Extraction Based On Motor Imagery EEG Signal

Brain-computer interface(BCI)is a new artificial intelligence control system.This technology changes the interactive links between human and electronic equipment and creates a new man-machine interactive mode.BCI has vast application prospect in assisted rehabilitation area.When humans do some mind work,brain will generate various signals,these signals can be recorded in the form of electroencephalogram,magnetoencephalography,fMRI etc.Compared with others,electroencephalogram has lower manufacturing cost,higher temporal resolution,smaller size.This thesis does some researches on BCI system based on motor imagery tasks,especially feature extraction method based on EEG signals.Firstly,the brain's physiological structure,the generation mechanism and type of EEG signal and the research method of BCI system are introduced,focusing on common algorithm related to feature extraction and classification,and analyzes their advantages and disadvantages.Secondly,an experiment of recording EEG signal based on motor imagery tasks is designed.The experimental scheme,experimental equipment and attentions are introduced clearly.Then,Permutation Entropy is introduced to BCI system based on motor imagery as a new feature.This new feature's feasibility is proved and an improvement permutation entropy feature is proposed which uses EMD to decompose and rebuild EEG signal and then extracts permutation entropy feature.After extracting feature,LDA classifier is used to classify.The result is better.Finally,another feature extraction algorithm which is based on accumulated power is proposed.This new feature is an improvement to band power feature.Through accumulating band power,the classification accuracy and mutual information is enhanced.Compared with other algorithm,this feature has obvious advantage.