Research On Classification Methods Of EEG Signal Evoked By Motor Imagery Tasks

EEG signal is a kind of bioelectricity produced spontaneously by the humanbrain. It can reflect the psychological state of human, which is the same as otherphysiological factors such as face temperature, pulse, and blood pressure. For a longtime, people wish to achieve the connection between the brain and the outside worldthrough the analysis on EEG signal. Now, it has been gradually becoming a realitywith the emergence of brain-computer interface (BCI) technology. Accurate andspeed for identifying different EEG signals are crucial goals that researchers try toachieve in the BCI application filed. It is an important way to solve the question toextract effective signal feature and select appropriate classification method. In orderto achieve this goal, in this paper, the specific research works are as follows:Firstly, since the EEG is highly non-stationary stochastic nonlinear signal andlow spatial resolution, this paper puts forward of the EEG signal classificationmethod of a combination of advantages electrode (Optimal Electrodes Selection,OES) and empirical mode decomposition (Empirical Mode Decomposition, EMD).Combined with previous research on the advantages of electrodes, filter outadvantages electrode according to the classification performance. On this basis,processing EEG single-channel signal uses the EMD method. Ultimately, thismethod can restructure feature, strengthen advantageous features, eliminateredundant features and improve the classification performance of the signal.Secondly, it puts forward an EEG signal classification method based onmuti-channel parallel processing principles. In the past, it always recognizes theconnection between the brain areas by observing the energy changes of brain regionson the time axis when the brain regions were activated. The method described in thispaper gives a new way which assumes the relation between corresponding brainregions through analysis the classification performance of the combination ofmulti-channel and the connection between the channels.Finally, it puts forward a classification strategy for EEG signal based oncontinuously imaginary. If the imaginary task in a period is uninterrupted, then theEEG signal collected is also continuous. At first, it supposes that transition pointexists between different tasks. The Euclidean distance between samples at the sidesof the transition point is deemed to be bigger than that between samples between the transition points. It would detect the transition point by setting a threshold. Besides,as is known to all, when imagine something continuously, it will lead to collect noisysignal due to factors such as inattentive or fatigue. The strategy proposes an idea ofsample purification. By setting the distance range between two samples adjacent toeach other, part of the whole samples is held and the category with most number isconsidered as the right category of the whole samples. In this paper, the strategyproposed was applied to the BCI2005dataset, and obtained the perfect recognitionperformance. The detail will be stated in the fourth chapter.