A Study of Finger Movement Classification Based On 2-sEMG Channels

Electromyography signals are highly valuable bioelectric signals in diagnosing abnormal nerve and muscle problems. Besides, in recent decades, processing and classification of EMG signals has become a core issue in prosthetic control applications. The focus of this study is an investigation into individual and combined fingers movement recognition using surface EMG signals. The dataset was used belongs to ten different classes collected from ten subjects. There are several sequential steps obtained to analysis EMG signals in this paper (i.e. preprocessing, feature extraction, feature reduction, pattern recognition). At first, EMG signals have been segmented by the windowing process. After that, various feature sets were extracted from these segments. Feature vectors were then reduced by applying two different reduction methods: Principal Component Analysis and Bhattacharyya Distance. Finally, they were fed to two classifiers: Artificial Neural Network and Fuzzy Logic. Overall average classification accuracies of these two systems were 96.08(±0.9)% and 90.56(±3)% respectively.