Hand gesture recognition using sEMG with LSTM

Abstract

Using electromyography (EMG) signals has spread in many fields. LSTM networks is one of the most suitable methods for processing EMG because of their structure. In this work, two LSTM models were build, one layer (1L-LSTM) and multi-layers ML-LSTM. They were trained using Ninapro-DB5 dataset after augmenting it by averaging. Different input sizes were tested. 1L-LSTM and ML-LSTM scored an accuracy of 98.5% and 99.7% respectively. Moreover, they needed low testing time in the range of [60,240] mcs. In addition, the signal length was did not have much effect when using multi layers. Keywords: LSTM, sEMG, hand gestures recognition, NinaPro-DB5, augmentation.