Pattern recognition of EMG signals for low level grip force classification.

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Signal Processing

Background:

• Upper limb amputation necessitates advanced prosthetic devices for restoring function.
• Myoelectric prosthetics utilize muscle signals for hand gesture and force level control.
• Accurate classification of force levels is crucial for intuitive prosthetic control.

Purpose of the Study:

• To develop and validate a pattern recognition algorithm for classifying different force levels using Electromyography (EMG) signals.
• To implement a two-step feature selection process combining ReliefF and Neighborhood Component Analysis (NCA).
• To optimize the number of muscles required for effective force level classification.

Main Methods:

• Acquisition of Electromyography (EMG) and fingertip force signals during varying force contractions.
• Application of a two-step feature selection: ReliefF for general ranking and NCA for personalized selection.
• Classification using Support Vector Machines (SVM) and Random Forest (RF) algorithms.
• Optimization of muscle set size for classification accuracy.

Main Results:

• A maximum classification accuracy of 99% was achieved using SVM with two muscles.
• Optimal features included Auto Regressive coefficients, Willison Amplitude, and Slope Sign Change.
• Mean classification accuracies were 94.5% for SVM and 91.7% for RF across subjects.

Conclusions:

• The proposed two-step feature selection effectively identifies key features for myoelectric control.
• The algorithm enables high-accuracy classification of hand grasp force levels, enhancing prosthetic usability.
• Optimizing muscle selection and feature extraction is vital for robust prosthetic control systems.