Surface Electromyogram Neural Feature Based Gesture Classification Using Fastica

Myoelectric control using pattern recognition methods is a critical research area in the field of rehabilitation engineering. Myoelectric control includes the data acquisition, signal processing, classification of movements and control of assistive devices such as exoskeletons and prosthetic arms based on the myoelectric signals. Gesture classification is a complex problem and needs precision. In this work, gestures are performed in various limb positions and EMG (Electromyogram) signals are recorded for multiple trials in each position. Independent Component Analysis (ICA) which is a Blind Source Separation (BSS) technique is used to decompose the motor unit action potentials from the recorded EMG signal. Feature space which consists of the extracted neural information, in addition to Root Mean Square (RMS) and Waveform Length (WL) of the EMG signal is utilized for classification of gestures using Support Vector Machine (SVM) classifier model. The results indicate that the proposed scheme is able to classify eight different hand gestures performed at five different limb positions with 98% accuracy.