Exploration of Fuzzy Logic As a Means to Handle Imprecise EMG Signals In Pattern Recognition Classifiers
DOI:
https://doi.org/10.57922/mec.2481Abstract
Myoelectric pattern recognition systems have the potential to offer intuitive selection and nearly seamless switching between different prosthetic hand grip patterns. This is made possible by using surface electromyogram (sEMG) signals to decode the user's intent each moment in time instead of the user sequentially switching between pre-programmed patterns via a unique motion. However, despite many advances in machine learning algorithms, myoelectric hands face numerous clinical barriers which prevent widespread user acceptance and adoption [2]. These clinical barriers include accuracy declines from sEMG signal shifts, imprecise control, operation lag times, and daily retraining time burdens [3,2]. Fuzzy Logic is a powerful tool which can transform ranges of numerical values into linguistic variables for performing mathematical approximations much like how we use language to describe subsets of populations without having exact numbers [4,5]. Therefore, since sEMG signals are notoriously noisy and have imprecise ranges, Fuzzy Logic may offer a way to account for this inherent signal property yet still be able to decipher the overall control signal command. This quality has the potential to address some of the clinical challenges of being able to reliably differentiate between active contraction and rest states, even if the sEMG signal has shifted due to fatigue or untrained arm positions which other machine learning algorithms seem to struggle with handling [3,7]. Based on promising results from Ajiboye & Weir, we seek to re-explore Fuzzy Logic as a rule-based pattern recognition system [1]. Our preliminary data shows that a Fuzzy C-Means (FCM) system is able to maintain higher accuracies across multiple bin sizes with averages ranging from 76-82% for the resting & momentary “OFF” data compared to a Linear Discriminant Analysis (LDA) system with averages ranging from 53-73%. Therefore, progress from a control perspective seems to have been made as it is easier to reliably return to a resting state before making a desired posture again instead of waiting for the control system to determine if the desired state is actually “OFF”. While this is intriguing, more optimization still needs to be done to have this FCM system obtain higher “ON” postural contraction accuracies closer to the clinical standard-of-care LDA system.
Downloads
Published
2024-08-15
How to Cite
[1]
S. Lorelli and R. Weir, “Exploration of Fuzzy Logic As a Means to Handle Imprecise EMG Signals In Pattern Recognition Classifiers”, MEC Symposium, Aug. 2024.
Conference Proceedings Volume
Section
Myoelectric Control Algorithms
