Research On Motor Imagery EEG Analysis Method

EEG (Electroencephalogram) based on motor imagery generally reflects theelectrophysiological activity of nerve cells in the cerebral cortex, which contains a largenumber of physiological information related to different kinds of consciousness. In thisprocess, the subjects just need to imagine without any actual action.It means that we couldrecognize human activities by analysing the EEG signals.In this way, BCI (BrainComputer Interface) may create a new communication channel between the brain and anoutput device converting action or language instructions directly into signals, which candrive an external device instead of the the normal output pathways of nerves andmuscles.In this regard, it is of great interests to investigate how EEG changes inpathological or physiological states by external and internal stimulation. How to extractcharacteristic parameters reflecting the user’s consciousness from EEG is the key to BCI.In this paper, we take motor imagery EEG as the object of the research. Based on theanalysis of the event related desynchronization (ERD) and event related synchronization(ERS), mental activity and the movement intention of the brain can be extracted from theEEG. We will select the data sets of BCI Competions2003and2005provided by GrazUniversity of Technology. The feature extraction method we used is a combination ofwavelet transform and information entropy, which is called wavelet entropy (WE).Support vector machine (SVM) is chosen to identify the extracted feature as aclassification method. If we take a SVM based on RBF (Radial Basis Function) asclassifier, the parameters C and σ are determined by subjective experience in typicalapplications. By using the cross validation (CV) algorithm and genetic algorithm (GA),the classifier performance is developed by optimizing the parameters. It is also proved thatthe method is effective in multi-classes tasks. Besides the electrodes C3and C4, T3andT4can also be used to collect the motor imagery-based EEG..