Myoelectric Signals Pattern Recognition for Intelligent Functional Operation of Upper-Limb Prosthesis

This paper represents a comparative study of the classification accuracy of myoelectic signals using multi-layer perceptron with back-propagation algorithm and radial basis functions networks. The myoelectric signals considered are used to classify four upper-limb movements which are elbow bending, elbow extension, wrist pronation and wrist supination. The network structure for multi-layer perceptron is a fully connected one, while the structures used in radial basis functions network are both fully connected and partially connected. Two learning strategies are used for training radial basis networks, namely supervised selection of centres and fixed centres selected at random. The results suggest that radial-basis function network with fixed centres can generalise better than the others without enquiring extra computational effort.