Hybrid independent component analysis and twin support vector machine learning scheme for subtle gesture recognition

Area of Science:

• Biomedical Engineering
• Pattern Recognition
• Machine Learning

Background:

• Myoelectric signal classification faces challenges due to significant variations in surface electromyogram (sEMG) features.
• Existing pattern recognition methods struggle with classifying sEMG signals, especially when muscle activity and size differ, leading to dataset discrepancy.
• Multicategory classification often involves numerous one-versus-rest binary tasks, resulting in unbalanced datasets that hinder performance.

Purpose of the Study:

• To develop a robust methodology for myoelectric signal classification that accounts for large variations in data distributions.
• To address the issue of unbalanced datasets inherent in multicategory classification problems.
• To improve the accuracy and reliability of classifying surface electromyogram signals corresponding to different muscle activities.

Main Methods:

• Extraction of hybrid features using Independent Component Analysis (ICA).
• Application of Twin Support Vector Machine (TSVM) techniques for classification.
• Developing a learning methodology to handle unbalanced datasets and variations in pattern distributions.

Main Results:

• The proposed hybrid feature extraction and classification approach demonstrates improved performance in handling dataset discrepancy.
• The methodology effectively addresses challenges posed by unbalanced datasets in multicategory myoelectric signal classification.
• Enhanced accuracy in distinguishing between different muscle activities through improved pattern recognition.

Conclusions:

• The combination of Independent Component Analysis and Twin Support Vector Machines offers a promising solution for complex myoelectric signal classification.
• This approach provides a more effective way to manage variations and imbalances in sEMG datasets.
• The study contributes to advancing pattern recognition techniques for prosthetic control and other biomedical applications.