On classifiability of wavelet features for EEG-based brain-computer interfaces
IJCNN'09 Proceedings of the 2009 international joint conference on Neural Networks
Hi-index | 0.00 |
A novel feature extraction method of spontaneous electroencephalogram (EEG) signals for brain-computer interfaces (BCIs) is explored. The method takes the wavelet packet transform (WPT) as an analysis tool and utilizes two kinds of information. Firstly, EEG signals are transformed into wavelet packet coefficients by the WPT. And then average coefficient values and average power values of certain subbands are computed, which form initial features. Finally, part of average coefficient values and part of average power values with larger Fisher indexes are combined to form the feature vector. Compared with previous feature extraction methods, the proposed approach can lead to higher classification accuracy.