ICA-based spatial & temporal filtering for the classification of single-trial EEG

Abstract

The fundamental neurophysiological phenomenon used in a Brain-Computer-Interface is based on the changes of power in the particular local area and frequency bands. To extract feature from the information, spatial and temporal filtering have to be developed effectively. In an attempt to obtain good features, ICA-based feature extraction method is suggested. ICA-based filtering method is operated through two steps. In the first step, spatial area used to classify different imagination is selected. It``s called ICA-based spatial filtering. Then, the signal obtained from each spatial map are bandpass filtered. The selected bands are also obtained from ICA-based filtering. Through the ICA-based spatial and temporal filtering, new coefficients localized in the spatial and temporal aspect can be obtained. If ICA-based filtering is working well, they also represent the neurophysiological characteristic well. In my thesis, the points are denoted. The main advantage of ICA-based filtering is to extract the spatial & temporal component effectively in the even unclear situation. It``s because ICA is unsupervised method and decompose mixed signals into independent component effectively. Particularly, when it``s difficult to select proper frequency band in spectral analysis, using the most supervised temporal filtering is not guaranteed to make good performance. In these aspects, the features from ICA-based filtering coincides neurophysiological knowledge, and also makes good performance.