Feature extraction based on common spatial analysis for time domain parameters

In this paper, an approach of feature extraction by designing common spatial filters specifically for time domain parameters (TDP) is proposed. This approach is aiming at motor imagery detection in electroencephalogram (EEG). Particularly, this method calculates the derivatives of the original signals and then applies common spatial analysis (CSP) to each order of derivatives. Variances of the spatially filtered signal after taking logarithm are used as features. Quadratic discriminant analysis (QDA) is applied to the feature vectors and classifies the vectors into different categories. We evaluate our approach using data consisting of two classes: left-hand and right-hand movement imageries from three subjects, and comparison between the proposed method and applying CSP analysis to the whole set of EEG signal directly is presented. Our results show that the proposed method generates more discriminant features in this motor imagery classification issue.