Abstract
The Electromyogram (EMG) signals are used in exoskeleton robot control for the recognition of the electrical activity related to the muscle contractions. In this study, surface EMG signals are classified to recognize the different types of myoelectric signals. The performance of a classifier is affected by the variation of EMG signals due to the different categories of contraction. To avoid such variations, the Wavelet Packet Decomposition (WPD) is used for features extraction from surface EMG signals. Then, a set of features selection methods is employed to reduce the high dimensional features. After a feature selection, different ensemble tree classifiers like Random Forest, Rotation Forest and MultiBoost are used for classification. Results are compared by using total classification accuracy, F-measure and Area Under ROC Curve (AUC). An effective combination of WPD and Random Forest achieves the best performance, using k-fold cross validation, with a total classification accuracy of 92.1%. The proposed methods in this study have potential applications in exoskeleton robot control and rehabilitation.