Motor imagery on EEG signals are widely used in brain computer interface (BCI) system with many interesting applications. However, it is not easy to interpret motor imagery EEG signal due to non-stationary and noisy features of the signal. In this paper, we investigate three different techniques of energy calculation as a part of energy extraction methods including L2-norm, leverage score, and absolute Z-score. This BCI framework use CSP as motor imagery signal feature extraction method and extreme learning machine (ELM) to classify the features of motor imagery signal. In general, the investigated framework has proved that the energy extraction methods can improve the performance of CSP. Also, an effective EEG channel selection provides better performance in terms of classification accuracy. In general, the proposed energy extraction methods can offer up to 21% performance improvement in accuracy and 86% reduction number of channels as compared to the original CSP.
