Improving the Performance Against Force Variation of EMG Controlled Multifunctional Upper-Limb Prostheses for Transradial Amputees

Abstract

We investigate the problem of achieving robust control of hand prostheses by the electromyogram (EMG) of transradial amputees in the presence of variable force levels, as these variations can have a substantial impact on the robustness of the control of the prostheses. We also propose a novel set of features that aim at reducing the impact of force level variations on the prosthesis controlled by amputees. These features characterize the EMG activity by means of the orientation between a set of spectral moments descriptors extracted from the EMG signal and a nonlinearly mapped version of it. At the same time, our feature extraction method processes the EMG signals directly from the time-domain to reduce computational cost. The performance of the proposed features is tested on EMG data collected from nine transradial amputees performing six classes of movements each with three force levels. Our results indicate that the proposed features can achieve significant reductions in classification error rates in comparison to other well-known feature extraction methods, achieving improvements of ≈ 6% to 8% in the average classification performance across all subjects and force levels, when training with all forces.

Myoelectric feature extraction using temporal-spatial descriptors for multifunction prosthetic hand control

“…Finally, it is important to mention here that TDPSD which we presented earlier in [12] is only focusing on the temporal part of the feature extraction and that it does not look at the relation between the EMG signals from the different channels. In comparison to TDPSD, and all other feature extraction methods, our new TSD method also looks at how the EMG power spectrum characteristics of the different channels change across the different movement and was therefore more capable of further reducing the classification error rates.…”

“…Simply speaking, f1, f2, and f3 are normalized versions of the power spectrum moments, f4 is a measure of sparsity, f5 is a measure of irregularity of the signal and f6 is a measure of the waveform length ratios. Further details about these features are in our original article [12]. In the remaining analyses, we denote the process of extracting the six features in Fig.…”

“…To evaluate the performance of the proposed feature set, denoted as Temporal-Spatial Descriptors (TSD), we compare the achieved classification performance against the classification performance of some well-known EMG feature extraction methods from the literature. These include: a combination of Autoregressive model parameters (5th order) with signal root mean square denoted as AR-RMS [16]; the combination of MAV, ZC, SSC, WL denoted as TD feature set [4]; the same combination with the WAMP features denoted as TD1 feature set; and the wavelet features extraction method, utilizing the mean of the squared wavelet coefficients at each level as features, with Symmlet family of wavelets and 5 decomposition levels, denoted as Wavelet; and our original TDPSD feature extraction method [12]. The dimensionality of all of the extracted feature sets was reduced using the Spectral Regression (SR) dimensionality reduction method proposed by Cai et al [17].…”

“…Such a gap is attributed to several factors including: lack of intuitive control, poor system reliability and the lack of robustness against practical problems like limb position change, electrodes shift, varying force levels, and signal nonstationarity. Recently, the extraction of features from the first and second derivative of EMG signal was shown to produce powerful features in EMG classification [12], with results suggesting that these feature can significantly outperform several well-known features in problems with changing limb position [13], varying contraction force levels and varying forearm orientations [14].…”

“…In this paper, we extend upon our recent work in this direction [12] by adopting an approach in which we, not only extract the features from the current analysis windows of each of the individual channels, but also utilize the differences between all channel pair combinations and their nonlinear mappings. In such a case, the extracted features not only look at the temporal evolution of the EMG patterns from each EMG channel but also at the spatial relation between the power spectrum activity of the different EMG channels.…”