University of Twente Student Theses


Comparison of common signal pre-processing techniques for long-term sEMG-based knee angle estimation

Ritmeester, J.C. (2022) Comparison of common signal pre-processing techniques for long-term sEMG-based knee angle estimation.

[img] PDF
Abstract:The rapid advancement of technology over the past decades have made it increasingly possible to help people through the use of advanced devices. One example of this are the microprocessor controlled knees, that allow people that have undergone a transfemoral amputation to regain the use of their leg. This prosthesis is driven by surface EMG (sEMG) signals that originate from the remainder of the limb after amputation. Unfortunately, sEMG signals are easily distorted by a number of sources, most of which are located inside the body. Numerous prior works have focused on improving the accuracy of these pros- theses by applying various filtering pre-processing steps, but no consensus has been reached on what combination of steps performs best. Furthermore, the performance in terms of accuracy is also depend- ent on the signal-to-noise ratio in the signals that are used to train the machine learning models that the prosthesis uses to control the knee. As it stands now, the performance deteriorates so quickly that the knee requires constant retraining, with the data acquisition process that comes along with, making it im- practical for widespread adoption. This thesis focuses on combining signal pre-processing techniques that are commonly found in scientific literature, and aims to find the best combination thereof with regards to multi-day model performance. The wavelet transform, empirical mode decomposition, and independent component analysis are compared, as well as the effect of differentiation of the signal, and two altern- ative ways to modify the signal by calculating the RMS in sub-windows and smoothing the envelope of the signals. Additionally, an LSTM-based autoencoder is proposed with the aim of reducing the dimen- sionality of the input signals, whilst simultaneously reducing the noise present in the signals. Through a process of gradual elimination, a number of best practices in terms of sEMG signal pre-processing were isolated.
Item Type:Essay (Master)
Roessingh, Enschede, Netherlands
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Interaction Technology MSc (60030)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page