Study of Feature Extraction Methods for BCI Applications

Abstract

In this work, different types of feature extraction methods for the detection of alpha waves and sensorimotor rhythms were analyzed. These signals were acquired through EEG. For the detection of alpha waves, the discrete wavelet transform and a neural network were used as feature extraction and classification methods respectively, resulting in an average detection accuracy of 89,1%. A BCI for the recognition of alpha waves was implemented through this method. Additionally, three feature extraction techniques for the identification of sensorimotor rhythms were proposed and studied. Discrete wavelet transform, autoregressive components and spatial filtering were the analyzed techniques; the classification method used was a neural network. An average accuracy of 60,81%, 61,78% and 64,59% was obtained for each method, respectively.