Classification of single-trial EEG signals
H. Zhou · 2005
Most of the existing electroencephalography (EEG) analysis methods have been developed on the basis of averaging over multiple trials in order to improve their performance against noise. These techniques usually work well in terms of their classification capability. However, they have shown deficiency in the presence of noise or artefact. In this paper, we explore an efficient artefact-removal technique based on a well-established independent component analysis (ICA) method (FastICA). This method is then used to decompose the original EEG signals, where artefacts or significant noise can be removed from the training single trial data. Our second contribution is to develop a modified SVM classification technique, based on the statistical estimation of the elements of the kernel matrix.