LSTM Classification of sEMG Signals For Individual Finger Movements Using Low Cost Wearable Sensor

Abstract

The electrical activity of the muscles that control finger movements can be extracted during the performance of these movements and using machine learning techniques, the myoelectric signals can be decoded and classified according to the movement that generated the specific signal. The focus of this paper is to classify sEMG signal using easily accessible cheap hardware to capture the signal. Furthermore, to employ neural networks to classify the signal using established methodology i.e. feature extraction, with the highest possible accuracy. To classify these sEMG signals, an LSTM network has been developed and was able to classify 12 individual finger movements with accuracies reaching 90%.