Using non-iterative methods and random weight networks to classify upper-limb movements through sEMG signals

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Signal Processing

Background:

• Surface electromyography (sEMG) is crucial for understanding and controlling upper-limb prosthetics.
• Accurate classification of sEMG signals is essential for intuitive prosthetic control.
• Existing methods like Support Vector Machines (SVM) have limitations in speed and accuracy for complex movements.

Purpose of the Study:

• To evaluate two non-iterative classification methods for 17 upper-limb movements using sEMG data.
• To compare the performance of these methods against a standard SVM classifier.
• To assess the impact of Principal Component Analysis (PCA) preprocessing on classification accuracy.

Main Methods:

• Utilized two non-iterative algorithms for sEMG signal classification.
• Compared performance against a Support Vector Machine (SVM) classifier across three distinct datasets.
• Implemented Principal Component Analysis (PCA) as a preprocessing step to enhance feature separation.

Main Results:

• Non-iterative methods demonstrated equivalent or superior classification accuracy compared to SVM.
• Regularized Extreme Learning Machines (RELM) achieved the highest accuracy without PCA (88.4% non-amputee, 79.4% amputee).
• PCA preprocessing significantly improved the accuracy of non-iterative methods, reaching 94% (non-amputee) and 85% (amputee) on average.

Conclusions:

• Non-iterative sEMG classification methods offer a promising alternative to traditional approaches.
• PCA preprocessing is highly effective in improving class separability and overall classification accuracy.
• These findings have implications for developing more responsive and accurate upper-limb prosthetic control systems.