EEG Preprocessing Methods for BCI Imagined Speech Signals

Brain Computer Interface (BCI) is a computer-based system that acquires signals from the brain, preprocesses the collected signals and then translate the input into commands sent to the effectors. Acquiring signals from the surface of the brain is a difficult task due to the low amplitude electric processes that occur. The recorded waves are usually contaminated with other biosignals with higher amplitude. These artefacts affect the quality of the signals leading to difficulty in decoding the desired brain activities. In this paper we focus on eliminating the electrooculogram from the electroencephalogram signals collected during imagined speech. We compared two preprocessing methods: the Adaptive Filter and Principal Component Analysis decomposition and concluded that both methods improve the quality of the EEG signals. However, the developed PCA method turned out to offer better results regarding the correlation between signals from the same class.