Effects of Training and Calibration Data on Surface Electromyogram-Based Recognition for Upper Limb Amputees

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Surface electromyogram (sEMG)-based gesture recognition is crucial for intelligent prostheses.
• Temporal variations in sEMG data reduce the efficiency of current recognition models.
• Understanding the impact of calibration and training data is vital for improving performance.

Purpose of the Study:

• To evaluate the effect of varying calibration data on sEMG gesture recognition accuracy.
• To assess the impact of the amount of training data on benchmark performance for amputees.
• To propose strategies for enhancing cross-session gesture recognition models.

Main Methods:

• Collected seven sessions of sEMG data from two upper limb amputees.
• Utilized the publicly available Ninapro DB6 dataset from ten healthy subjects.
• Analyzed the impact of calibration data and varying amounts of training data on recognition accuracy.

Main Results:

• Calibration data improved average accuracy by 3.03% and 6.16% for amputee subjects and 9.73% for the Ninapro DB6 dataset.
• Increasing the number of training sessions proved more effective than increasing training trials for accuracy improvement.
• Identified key factors for enhancing cross-session models.

Conclusions:

• Calibration and sufficient cross-session training data are critical for commercializing intelligent prostheses.
• Strategies to maximize dataset utilization are essential for improving model performance.
• Findings provide a foundation for more robust and reliable prosthetic control systems.