EMG-based Gesture Recognition using Extreme Learning Machine

Electromyography (EMG) signals are crucial data to track muscle activities making them a key point for design of prosthetic devices. In order to classify the finger movements in upper limb prothesis, a recognition method based on an extreme learning machine (ELM) is studied in this paper. The target to be recognized is constructed by grouping two kinds of hand gestures, individual finger movements, and combined finger movements. The recognition feature matrix is established by decomposing the windowed signal using wavelet transform and extracting conventional time-domain features from that. These features are then fed to three different classifiers for recognition of binary class. Currently, the best accuracy achieved using ELM is above 95%, outperforming Support Vector Machine (SVM) and Fine Trees in this study