A discriminant bispectrum feature for surface electromyogram signal classification.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rehabilitation Engineering

Background:

• Surface electromyogram (sEMG) signals are crucial for prosthetic limb control.
• Accurate classification of sEMG signals is essential for intuitive prosthetic function.
• Existing feature extraction methods have limitations in discriminating complex sEMG patterns.

Purpose of the Study:

• To propose and evaluate a novel Discriminant Bispectrum (DBS) feature extraction technique for sEMG signal classification.
• To compare the performance of DBS against conventional sEMG feature extraction methods.
• To assess the effectiveness of DBS for prosthetic control applications.

Main Methods:

• Feature extraction using Discriminant Bispectrum (DBS), involving bispectrum matrix integration and Fisher linear discriminant (FLD) projection.
• Comparison with conventional features: autoregressive coefficients, root mean square, power spectral distribution, and time domain statistics.
• Evaluation of feature separability using FLD subspace visualization and Davies-Bouldin clustering index.
• Classification accuracy (CA) assessment using four linear and non-linear classifiers.

Main Results:

• Discriminant Bispectrum (DBS) demonstrated superior feature separability compared to conventional methods.
• DBS achieved higher classification accuracy (CA) across various classifiers.
• The optimal CA achieved with DBS reached 99.4% for identifying sEMG motion patterns.

Conclusions:

• The proposed Discriminant Bispectrum (DBS) approach offers a significant advancement in sEMG feature extraction for prosthetic control.
• DBS effectively enhances the discriminative power of sEMG signals, leading to improved classification accuracy.
• This method holds strong potential for developing more responsive and intuitive prosthetic devices.