Force-Invariant Improved Feature Extraction Method for Upper-Limb Prostheses of Transradial Amputees

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rehabilitation Technology

Background:

• Extracting force-invariant features from electromyogram (EMG) signals is challenging due to muscle physiology.
• Existing methods struggle to isolate unique information across varying force levels.

Purpose of the Study:

• To propose an improved force-invariant feature extraction method for enhanced EMG-based gesture recognition.
• To enhance the robustness and performance of EMG signal analysis for prosthetic control.

Main Methods:

• Developed a novel method using nonlinear transformation of power spectral moments, amplitude changes, and spatial correlation coefficients.
• Incorporated nonlinear transformation to balance forces and improve gesture discrimination.
• Utilized an EMG dataset from nine transradial amputees for validation.

Main Results:

• The proposed method significantly improved pattern recognition performance.
• Demonstrated substantial gains in accuracy, sensitivity, specificity, precision, and F1 score.
• Achieved higher performance compared to six existing feature extraction methods across three classifiers.

Conclusions:

• The enhanced feature extraction method provides superior robustness and accuracy for EMG-based gesture recognition.
• The method is computationally efficient, requiring less time and memory.
• Offers a more robust solution for prosthetic limb control and human-computer interaction.