A Preprocessing Method of EEG Based on EMD-ICA in BCI

In order to remove artifacts automatically and effectively from the Electroencephalography (EEG) in Brain Computer Interfaces (BCIs), a new preprocessing algorithm called EMD-ICA (Empirical Mode Decomposition, Independent Component Analysis) is explored. The EMD-ICA method includes the following steps: Firstly, EEG signals from single or multiple channels are decomposed into a series of intrinsic mode functions (IMFs) using EMD. Each IMF can be approximately used as an input channel of the ICA, and these IMFs constitute the input matrix of the ICA. Then, the input matrix is separated into a set of statistics independent components (ICs) by ICA. Furthermore, each of statistics ICs is analyzed by using the method of sample entropy to automatically determine whether it is artifact signal. Finally, the ICs determined as artifacts are eliminated and the remaining ICs are reconstructed. The reconstructed EEG is used in the following feature extraction and classification. To evaluate the effect of the proposed method, common spatial patterns (CSP) and support vector machine (SVM) algorithm are used to extract and classify the EEG data from two datasets. The experimental results show that the proposed method can remove various kinds of artifacts effectively, and improve the recognition accuracy greatly.