Multiclass motion identification using myoelectric signals and Support Vector machines

In this paper, different classifiers were trained to identify myoelectric registers, in order to recognize nine different motions related to four degrees of freedom of the forearm. Three main methods were compared, namely Linear Discriminant Analysis, Artificial Neural Networks and Support Vector Machines. The behavior of pattern recognition schemes was investigated using different amounts of data collected from 12 healthy subjects. The focus of this work is to identify the best classification scheme. Departure information was obtained using a preprocessing stage to extract either autoregressive or frequency domain features. Experiments show that the best performance is achieved employing frequency features and support vector machine classifier. This classification scheme demonstrates exceptional recognition accuracy of over the other methods.