TLDR: This research introduces a novel fuzzy logic-based classification system for controlling bionic upper limb prostheses. It addresses the challenge of signal contamination in electromyographic (EMG) biosignals by using a dual-ensemble approach: one-class classifiers detect signal contamination, and K-nearest neighbors classifiers recognize patient intent. The fuzzy model allows for ‘soft’ decisions, dynamically adjusting the influence of noisy signals to improve overall classification accuracy and robustness, especially in challenging, low signal-to-noise ratio environments. Experimental results confirm its superior performance over existing methods.
Controlling modern bionic hands, which are designed to mimic human upper limbs, often relies on electromyographic (EMG) biosignals. These signals, generated by muscle activity, are captured from the body surface and translated into commands for the prosthesis. However, achieving precise and reliable control is a significant challenge due to various factors that can degrade the quality of these biosignals.
One of the most critical issues is signal contamination. EMG signals are highly susceptible to noise from sources like electrical interference, muscle fatigue, or even slight shifts in electrode placement. This contamination can drastically reduce the accuracy of the system that recognizes a patient’s intended movements, leading to frustrating and incorrect actions by the prosthesis.
Researchers Pawel Trajdos and Marek Kurzynski from Wroclaw University of Science and Technology have proposed a novel solution to this problem. Their work, detailed in the paper A Compound Classification System Based on Fuzzy Relations Applied to the Noise-Tolerant Control of a Bionic Hand via EMG Signal Recognition, introduces a new recognition system designed to not only interpret EMG signals but also to detect and mitigate the adverse effects of these contaminations.
A Dual-Ensemble Approach with Fuzzy Logic
The core of their proposed system lies in its dual-ensemble structure and the innovative use of fuzzy logic. Instead of relying on a single, rigid decision-making process, the system employs two main components:
-
One-Class Classifiers (OCC): This ensemble of classifiers is dedicated to assessing the degree of contamination in individual EMG signal channels. Think of it as a purity checker for each signal input.
-
K-Nearest Neighbors (KNN) Classifier: This ensemble is responsible for recognizing the patient’s actual intent – what movement they are trying to make with the bionic hand.
What makes this system particularly unique is the integration of a coherent fuzzy model throughout the recognition process. Fuzzy logic, unlike traditional ‘yes’ or ‘no’ (crisp) logic, allows for ‘soft’ decisions, dealing with degrees of truth rather than absolute true or false. For example, instead of a signal being either ‘contaminated’ or ‘clean,’ fuzzy logic can determine that a signal is ‘partially contaminated’ to a certain degree.
In this system, the OCCs use fuzzy logic to assign a ‘membership function’ value to each signal, indicating its degree of purity. This purity information is then used to ‘correct’ the similarity measures within the KNN classifiers. Essentially, if a signal channel is deemed highly contaminated, its influence on the final decision for recognizing the patient’s intent is appropriately reduced. This dynamic adjustment helps the system make more reliable predictions even when some input signals are noisy.
Also Read:
- Advancing Brain-Computer Interfaces with Rapid Online Adaptation
- DACL: Enhancing Biosignal Analysis with Diffusion-Powered Contrastive Learning
Experimental Validation and Key Findings
The researchers rigorously tested their system using real biosignals from a public repository, involving data from nine different subjects with amputations. They simulated various types of noise, including power grid noise, signal attenuation, and Gaussian noise, across different signal-to-noise ratio (SNR) levels.
The experimental results were compelling:
-
The proposed fuzzy method significantly improved classification quality compared to a single, noise-insensitive classifier, especially when signal contamination levels were high (low SNR).
-
It outperformed other reference methods described in scientific literature, particularly those that combine classifiers at a basic prediction level or use simpler attribute weighting.
-
The system demonstrated greater robustness to noise, maintaining better performance even with heavily contaminated signals.
-
Interestingly, the specific mathematical form of the fuzzy membership function used (e.g., linear, spline) did not significantly impact the classification quality for the datasets studied, as long as it could distinguish between contaminated and uncontaminated samples.
In conclusion, this research presents a significant step forward in making bionic hand control more reliable and user-friendly. By intelligently detecting and accounting for signal contamination using fuzzy logic, the system can better interpret a patient’s intentions, leading to more accurate and consistent prosthetic movements.
