Investigation of an sEMG Signal-Based Method for Online Hand Movement Classification

The human hand possesses remarkable dexterity and performs intricate tasks. Developing a reliable myoelectric prosthetic system for amputees is crucial. Although surface electromyographic (sEMG) signals have been utilized, their vulnerability to interference presents hurdles in hand movement identification. This work proposes a machine learning-based system for accurate recognition of hand movements. During the offline training process, artificial neural networks (ANN) have demonstrated superior performance compared to support vector machines and k-nearest neighbors, achieving a recognition rate of 99.63%. It underscores the exceptional ability of ANNs to learn complex nonlinear relationships within sEMG signals. Online experiments with a majority voting approach validated the feasibility of the system, reaching an average accuracy of 93.33%. This work provides technical support for fine control in myoelectric prosthesis, promising an improved quality of life for amputees.