ANALYSIS AND IMPLEMENTATION OF BRAIN WAVES FEATURE EXTRACTION AND CLASSIFICATION TO CONTROL ROBOTIC HAND

Abstract

In this paper, feature extraction methods such as time domain, frequency domain and spatial domainwere investigated. Where Mean Absolute Value (MAV), Integrated Absolute Value (IAV), Zero Crossing (ZC), RootMean Square (RMS), Waveform Length (WL) and Slope Sign Change (SSC) are the used time domain features.Autoregressive Feature (AR) is the frequency domain feature and the spatial domain feature is the CommonSpatial Patterns (CSP). Channel selection algorithm was proposed for dimensionality reduction using Matlab code.Results of the above algorithm were compared with Matlab library of Principle Component Analysis (PCA). Theextracted feature vectors were fed into Support Vector Machine with Radial Basis Function kernel (SVM-RBF) totrain the classifier. The pair of algorithms (feature extraction plus dimensionality reduction) that owned the lowestclassification error rate were used to control a Humanoid Robotic Hand (HRH) in offline mode. EEG dataset of twoclasses and three bipolar channels was used. Results showed that CSP features achieved the lowest classificationerror rate for both dimensionality reduction techniques with 2.14%. Results recommends to use (CSP plus channelselection algorithm) over (PCA plus PCA) since the former owned lowest classification processing time of 8.2s over8.5s for the later.