Dimensionality Reduction for the Control of Powered Upper Limb Prostheses

Myoelectric prosthesis control is difficult to achieve, because the optimal selection of features from a set, extracted from a myoelectric signal, is known as being a very challenging problem in prosthetic control. Dimensionality reduction means to retain MES information, that is important for class discrimination and to discard irrelevant data. Dimensionality reduction strategies can be categorized into feature selection and feature projection methods according to their objective functions. This work brings forward a statistical cluster analysis technique, which we call the "Guilin Hills selection method". The goal of our efforts is to control electrically powered upper limb prostheses with a minimum number of sensors and a low-power processor. The paper introduces the method and brings forward exemplar results for the selection of two time-domain features. With the proposed method, we can clearly distinguish four hand-positions from myoelectric signals of two sensors. The method also suggests feature combinations that lead to less robust classifications but require fewer sensors