Enhanced Finger Movement Detection Using sEMG by Data Augmentation

Abstract

Surface Electromyography (sEMG) is a technique to capture electrical activity in muscles during contraction. Individual finger movement has not received much attention as a significant proportion of the current sEMG research targeting the hand has focused on gestures. Accurate classification of individual finger movements is essential for several applications including robotic prostheses and computer security applications. A problem we face in classifying individual finger movements is data sparsity resulting from device availability, acquisition time, and patient privacy laws. To alleviate the problem of limited training data, we propose to synthetically augment the training data. Although some sEMG data augmentation methods have been studied in the literature, their contribution in improving prediction performance is still limited. Pattern mixing has shown promising performance in general time series augmentation, but has not been studied for sEMG. However, one major limitation of pattern mixing is its expensive computational cost. Therefore, in this paper, we propose a random combination method which helps to diversify our training data, as well as to reduce the time required for building the synthetic data. Our empirical study on several subjects using sEMG demonstrates both the effectiveness and efficiency of our proposed method. %Additionally, we believe our study to be the first to evaluate pattern mixing techniques on sEMG data.