An Adaptive Feature Extraction Method for Motor-Imagery BCI Systems

Recently, the research on Brain-Computer Interface (BCI) technology has achieved great progress, and the BCI system based on Motor Imagery (MI) has been intensively studied in many labs. The essential part of signal processing in BCI is how to extract the MI features in electroencephalographic (EEG) and recognize the MI task accurately. One challenge lies in that EEG signals are non-stationary, whose features vary with time. The traditional methods often don't perform well in BCI, because it does not capture the change of EEG automatically. In this paper, an improved adaptive common spatial patterns (ACSP) method is proposed to adapt to the change of EEG. We test our method for adaptive feature extraction with data from BCI motor imagery experiment, and the efficacy is evaluated by the feature classification accuracy with a support vector machine (SVM) classifier. The results show the effectiveness of the improved adaptive algorithm.