A Novel Approach for Classification of Hand Movements using Surface EMG Signals

Surface Electromyography (EMG) signals are being used in various fields including myoelectric control for prosthetic devices, human machine-interaction and other clinical and biomedical applications. EMG signals acquired through surface electrodes include various sources of noise like electrode-skin interface and electromagnetic noise etc. Therefore, accurate processing and classification of EMG signals is a challenging step in order to achieve reliable results for myoelectric control and so on. This paper introduces a novel classifier using Stacked sparse autoencoders (SSAEs) for improved myoelectric control. Data of six surface EMG channels is collected from five able-bodied and two amputee subjects for two days using their right hand or the amputated one. For performance evaluation, offline classification error is used and performance of SSAEs is compared with state of the art LDA. For within day analysis, SSAEs (1.29 +/- 0.83%) outperformed LDA (4.09 +/- 2.15%) with p value of 0.0018 for both able-bodied and amputee subjects. In between days' analysis, SSAEs outperformed (P < 0.001) LDA for both able-bodied and amputee subjects. These findings suggest that deep features of Autoencoders have the potential to be used as control source for myoelectric control systems and it can significantly improve the performance as compared to classical machine learning algorithm.