Myoelectrical signal classification for the hierarchical control of a human hand prosthesis

Most electrically powered upper limb prostheses are myoelectrically controlled. The myoelectric controllers use surface electromyographical signals as inputs. These signals, collected from the surface of the skin, have to be preprocessed before being used as inputs for the controller. In this paper we present a classifier for surface electromyographical signals based on an autoregressive (AR) model representation and a neural network, and the higher level of the hierarchical controller implemented using Finite State Machine. The results had shown that using a low order autoregressive model combined with feed forward neural networks achieves a rate of classification of 91% while keeping the computational cost low. Using the hierarchical controller, the necessary effort to control the prosthesis by the patient is reduced since the patient only have to initiate the movement which is finalized by the low level part of the controller. The inputs of the high level controller are obtained from the classifier. The outputs of the high level controller are applied as inputs to the low level controller.