Fuzzy discriminant analysis based feature projection in myoelectric control

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Machine Learning

Background:

• Myoelectric signals (MES) are crucial inputs for prosthetic hand controllers.
• High-dimensional feature vectors from multi-channel MES increase computational cost.
• Dimensionality reduction is essential for efficient and accurate pattern recognition in prosthetic control.

Purpose of the Study:

• To propose a novel feature projection technique for reducing MES dimensionality.
• To enhance the performance of pattern recognition classifiers for prosthetic hand control.
• To improve the generalization capability and reduce computational cost in MES-based systems.

Main Methods:

• Developed a Fuzzy Logic Fisher's Linear Discriminant Analysis (FLDA) method.
• Incorporated fuzzy membership degrees to mitigate overlapping data points.
• Utilized Mutual Information (MI) to weight features based on their discriminatory contribution.

Main Results:

• The proposed FLDA method demonstrated superior performance compared to other techniques.
• FLDA effectively reduced feature vector size while preserving essential discriminatory information.
• Tested on a seven-class MES dataset, FLDA proved its efficacy in pattern recognition.

Conclusions:

• FLDA offers an effective approach for dimensionality reduction in myoelectric control.
• The method enhances classifier generalization and reduces computational load.
• FLDA shows significant promise for improving multifunction prosthetic hand control systems.