Feature extraction of forearm EMG signals for prosthetics

This paper presents a new technique for feature extraction of forearm electromyographic (EMG) signals using a proposed mother wavelet matrix (MWM). A MWM including 45 potential mother wavelets is suggested to help the classification of surface and intramuscular EMG signals recorded from multiple locations on the upper forearm for ten hand motions. Also, a surface electrode matrix (SEM) and a needle electrode matrix (NEM) are suggested to select the proper sensors for each pair of motions. For this purpose, EMG signals were recorded from sixteen locations on the forearms of six subjects in ten hand motion classes. The main goal in classification is to define a proper feature vector able to generate acceptable differences among the classes. The MWM includes the mother wavelets which make the highest difference between two particular classes. Six statistical feature vectors were compared using the continuous form of wavelet packet transform. The mother wavelet functions are selected with the aim of optimum classification between two classes using one of the feature vectors. The locations where the satisfactory signals are captured are selected from several mounted electrodes. Finally, three ten-by-ten symmetric MWM, SEM, and NEM represent the proper mother wavelet function and the surface and intramuscular selection for recording the ten hand motions. Published by Elsevier Ltd.