A Comparative Analysis on the Impact of Linear and Non-Linear Filtering Techniques on EMG Signal Quality of Transhumeral Amputees

Abstract

Myoelectric pattern recognition (MPR) based strategies have been well investigated and applied for the control of upper limb prostheses. Despite their wide adoption, EMG-PR based upper limb prosthesis is not yet available in the clinics and commercial stores for above-elbow amputees. On the one hand, above-elbow amputees do not have enough residual muscles to generate a rich set of signals needed to effectively control the device. On the other hand, the limited acquire signal from these category of amputee is often contaminated by several kinds of noises that make it difficult to decode their movement intent which serves as a control input. Hence, there is a need to improve the quality of signals generated by such amputees through an efficient approach, to enhance the decoding outcomes of their limb motions. Therefore, this study systematically investigated the capability of a linear (Wiener filtering (WF)) and non-linear (1 Dimensional median filtering: 1 D-Median) techniques in denoising EMG signals obtained from above-elbow amputees towards improving its overall quality. The performance of both filtering techniques was examined using high-density surface electromyogram (HD-sEMG) recordings obtained from four transhumeral amputees who performed five distinct classes of targeted limb motions across three time-domain features while a linear discriminant analysis (LDA) classifier was adopted. Experimental results showed that WF technique could better denoise the signals by achieving an increment of up to 5.0% across limb motions in decoding accuracy compared to the 1D-Median filtering technique. This result suggests that WF can help to improve the overall performance of the feature extraction, and then this may be one reason why it has found application across the domain.