Feature Extraction of Mental Task in BCI Based on the Method of Approximate Entropy

Brain computer interface (BCI) is based on processing brain signals recorded from the scalp or the surface of the cortex in order to identify the different brain states and covert to corresponded control command. The key problems in BCI research are feature extraction and classification. In this paper, two experiments were performed, and the EEG data were recording during each experiment. One experiment contains five mental tasks, including ";baseline";, ";rotation";, ";multiplication";, ";counting"; and ";letter-composing";, the other contains two mental tasks which are left hand imagery movement and right hand imagery movement. EEG data recorded from both experiments are analyzed by approximate entropy (Apen), which is used to extract the characteristic feature of different mental tasks. A three-layer BP Neural Network classifier was structured for pattern classification. Different results were gained from the mental task experiment and imagery movement experiment. The results show that Apen is an effective method to extract the feature of different brain states.