Deep Neural Network Approach for Hand, Wrist, Grasping and Functional Movements Classification using Low-cost sEMG Sensors

This paper presents a Deep Neural Network approach for the classification of 41 hand, wrist, grasping and functional movements based on a public dataset of low-cost sEMG sensors called Ninapro DB5. The recent advancements in sensor technology, mechatronics, signal processing techniques and edge computing hardware equipped with GPU make dexterous prosthetic hands with non-invasive sEMG sensors and control capabilities of machine learning possible. However, its high cost means the technology is not accessible to most people. Therefore, the objective of this paper is to investigate the control and capabilities of a low-cost setup of sEMG sensors for a prosthetic hand. The acquisition setup includes two Thalmic Myo armbands for the total of 16 channels with the sampling rate of 200 Hz. Our approach achieved an overall accuracy of 91% with a macro recall of 77% for the classification of 41 movements, outperformed other algorithms such as SVM, Random forest, and XGBoost. These results suggest that a development of practical prosthetic hand could be possible with low-cost sEMG sensors.