Discrete Wavelet transform and ANFIS classifier for Brain-Machine Interface based on EEG

In this paper, an on-line Brain-Machine Interface (BMI) based on Electroencephalography (EEG) that closes and opens a robotic hand when eyes are closed and open, respectively, is presented. This BMI is based on the measurement of the EEG bipolar connection: 01-P3. Moreover, since it is considered to be very important for some BMI biomedical applications a fast processing time and feature classification of the EEG signal, the authors propose a novel algorithm for on-line DWT processing of the EEG signal that, along with the feature classifier, have an average processing time (APT) of 37.9 ms. An Adaptive Neuro-Fuzzy Inference System (ANFIS) was used as the feature classifier obtaining an on-line average classification accuracy (ACA) of 96.0%; after an off-line ANFIS training. The average and the maximum value of the last two calculated level 4 detailed coefficients (cD
₄
), derived from Wavelet's decomposition, were used as the inputs of the ANFIS classifier. The detailed coefficient cD
₄
was selected due to the fact that this coefficient isolates the EEG alpha wave(7.19~14.4 Hz), which presents significant changes on the bipolar connection O1-P3 whenever a subject closes and opens its eyes. The output of the ANFIS classifier is the input voltage of a microcontroller, which generates a Pulse-Width Modulation (PWM) signal that controls the movement of the robotic hand.