Controllo di una protesi mioelettrica = Control of a myoelectric prosthesis

Starting from the electromyography, usually recorded on the surface of the skin, a prosthesis can be controlled following a classical step-flow. By adding a real-time pre-processing to a single-channel surface electromyography, a new signal is obtained: the deconvolution, which provides an estimation of the firing rates of the muscle units involved in the motion. In this work, after giving an overview of the classical pattern control strategies, we resolved a classification problem with both the electromyography and the associated deconvolution. The problem is: 10 motion classes, 2 recording channels on 10 healthy subjects. Classical time-domain features are extracted from the signals and reduced by Mutual Component Analysis, and the classification is done by both a Support Vector Machine and a k-Nearest Neighbours. The overall results of this work are that the classification results are better and more robust when using the estimation of the firing rates than the classical signal. Even if deeper insight in this new technic is required to state a clear advantage over the electromyography, this preliminary work gives promising outcomes for a further use of the deconvolution as prosthesis control technic.