HYBRID CLASSIFICATION STRATEGY OF EMG SIGNALS FOR ROBOTIC HAND CONTROL

Abstract

Among the different bio-signals modalities, Electromyographic signal (EMG) has been one of the frequently used signals in the bio-robotics applications field. This is due to the fact that the EMG reflects directly the muscle activity of the user following the human motion intention. Consequently, the decoding of this intention is an essential task for controlling devices such as prosthetic hands and exoskeletons, based on EMG signals. This paper deals with the processing of EMG signals of the forearm muscles, in order to control two degrees of freedom (2 DoFs) robotic hand. The main contribution of this paper is the proposal of a hybrid approach that combines a pattern and a non-pattern recognition-based strategy. The proposed approach aims to take advantage of both strategies and overcome their shortcomings leading to a better analysis of the user movement intention. The EMG recorded signals are processed for feature extraction based on a Wavelet Packet Decomposition (WPD) method and classification using an Artificial Neural Network (ANN). Furthermore, we investigate the effect of the various parameters such as the applied force level, the number of the EMG channels and the window length of the EMG signal. The proposed approach is validated experimentally under realistic conditions. Very interesting results have been obtained for user intention decoding.