Development of sEMG based Movement Recognition System for Wrist Disarticulation Amputees

A myoelectric control system has been developed for the amputated subject in this study. Surface myoelectric signals from the residue forearm of the amputated subject were collected with a portable eight-channeled myoelectric signal sensing device. Seventeen features and the raw signals have been applied for the classification of six basic finger movements. The performances of these features, and the signal preprocessing techniques were evaluated with LDA and the Gaussian classifier. The results showed that the classification accuracy of the Gaussian classifier was much higher than that of the LDA for 16 out of 18 features, while their computational cost was comparable. Normalization did not help to increase classification accuracy for the Gaussian classifier, and its improvement for LDA was also limited. With the increase of training samples, the classification accuracies of both the classifiers increased, and the increase speed tended to slow down when training samples were more than 3. Meanwhile, the standard deviation of the results from the Gaussian classifier tended to narrow down. However, no reduction was observed from the results of LDA. Therefore, the Gaussian classifier was suggested for myoelectric control in this study.