The Effects of the Number of Channels and Gyroscopic Data on the Classification Performance in EMG Data Acquired by Myo Armband

Abstract

Processing and classification of Electromyography (EMG) signals is a common practice in prosthetic arm design for hand amputees. This study investigates how reliable classification results may be obtained from less muscle data, as is the case for the muscle loss in hand amputees. In order to increase classification performance, features from gyroscopic data, not previously used in studies in the literature, were investigated. Data was acquired from 10 normal subjects using the Myo armband for 7 hand gestures: fist, fingers spread, wave-in, wave-out, pronation, supination, and rest. Subjects repeated each gesture 30 times. EMG signals were preprocessed to extract features. Twenty (20) features were used in the feature matrix; 14 time domain and 6 frequency domain. Features were selected to determine the highest accuracy using the Support Vector Machine (SVM) and k-Nearest Neighbor (KNN) as classification algorithm. The Classification Learner application in Matlab? was used for classification. The highest accuracy using all EMG channels was 98.38 %. When the number of channels was reduced to 3, the accuracy was over 90 %. It is observed that gyroscopic features increase the performance when a small number of EMG channels is used.