Support vector machine-based classification scheme for myoelectric control applied to upper limb.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neuroscience

Background:

• Myoelectric control systems translate muscle electrical activity into device commands.
• Accurate classification of upper limb motions is crucial for effective prosthetic and assistive device control.
• Existing classifiers like LDA and MLP have limitations in performance and computational load.

Purpose of the Study:

• To propose and evaluate Support Vector Machine (SVM) for classifying upper limb motions using myoelectric signals.
• To determine the optimal configuration for SVM-based myoelectric control.
• To compare SVM performance against Linear Discriminant Analysis (LDA) and Multilayer Perceptron (MLP) neural networks.

Main Methods:

• Investigated data segmentation techniques, feature sets, SVM model selection, and postprocessing methods.
• Examined overlapped segmentation and majority voting for performance enhancement.
• Compared SVM with LDA and MLP for myoelectric signal classification.

Main Results:

• SVM demonstrated exceptional accuracy and robust performance in classifying upper limb motions.
• SVM exhibited a lower computational load compared to LDA and MLP.
• Classifier output entropy was identified as a potential online index for classification correctness.

Conclusions:

• SVM is a highly effective classifier for myoelectric control of upper limb movements.
• Optimized SVM configurations can significantly improve controller performance.
• The proposed methods, including entropy-based evaluation, support long-term, adaptive myoelectric control.