Surface EMG based classification of basic hand movements using rotation forest

Abstract

This paper defines a man-machine interaction for the prosthetic hand control using a surface electromyogram (sEMG) signals. The surface EMG signals are used in hand movement recognition. Different types of muscle contraction can cause EMG signals to vary, affecting classification performance. In this study, MSPCA is used for denoising and WPD is used for feature extraction to evaluate their efficiency for classifying surface EMG signals, which were recorded during the grasping movements with various objects. The time-frequency domain features were extracted and used in the identification of intention from surface EMG signals. Furthermore, the performance of different classifiers is quantified in terms of the total classification accuracy. An effective combination of WPD and Rotation Forest classifier attains the finest performance with a maximum classification accuracy of 98.33% using k-fold cross validation. The proposed method has potential applications in the prosthetic hand control and exoskeleton robot control.