Extreme energy difference for feature extraction of EEG signals

Abstract

A new method called extreme energy difference (EED) is proposed for supervised feature extraction of electroencephalogram (EEG) signals. It is a linear feature extractor which aims at maximizing or minimizing the disparity of energy features between two classes of EEG signals. The final transform for feature extraction in EED is very concise, which is resolved by an eigenvalue decomposition problem. In the context of EEG signal classification for brain–computer interfaces, the performance of EED is evaluated with real EEG signals from different subjects. Experimental results on nine subjects show that the EED feature extractor is comparable to the state-of-the-art feature extraction method common spatial patterns (CSP). Furthermore on another benchmark data set, by combining features obtained by EED and CSP, we train a linear support vector machine classifier whose classification accuracy outperforms the best result reported. This shows EED can be a beneficial complement to CSP.