Research On Recognition Algorithm Of EEG Related To Motor Imagery

As human control center, brain takes the responsibility for multiple physiological functions, so it will make great promotion in brain science, rehabilitation engineering and intelligence information processing to explore the mysteries of brain. Brain Computer Interface is one of the available technologies in brain science, which aims at creating direct information interaction between brain and external environment based on EEG. EEG related to motor imagery are widely used in BCI and its feasibility have been demonstrated. Considering the fact that abundant information about brain activity is involved in EEG, it is a feasible approach for the implementation of communication between brain and external environment using feature extraction and feature feature recognition.The first research emphasis of this paper is establishment of motor imagery recognition system, which contains mainly EEG preprocessing, feature extraction and feature recognition. Due to the high dimension and low signal to noise ratio of EEG, much research work on noisy data removing is done, and different combinations of several disconcordant preprocessing algorithm are carried out to figure out the optimum scheme. Analysis of the effect on effective component extraction causing by variety of different parameters in preprocessing is another research focus, and it leads to a fully exploration of EEG properties related to motor imagery. Experimental results reveal that preprocessing based on band-pass filtering and independent component analysis lead to the highest accuracy, which is 87.87%, and the corresponding optimum filter band is 12-16 Hz for EEG related to motor imagery. Meanwhile, the results also show that the number of independent component should be set to 10.Another research theme of this paper is EEG acquisition region selection algorithm based on independent component analysis and frequent itemsets, which can help to solve the high dimension problems of EEG. The region selection algorithm presented in this paper takes advantages of independent component analysis to find out regions affected by each component on scalp, and then frequent itemsets are mined, which is the basement for effective EEG acquisition regions. The algorithm help to achieve the following goals: reducing the redundancy data in EEG, increasing signal to noise ratio and improving data quality. Experimental results shows that the number of electrode on scalp reduces from 118 to 32 without negative effect on the accuracy of motor imagery recognition, meanwhile the effective acquisition regions for EEG of different band are not the same.