SeNic: An Open Source Dataset for sEMG-Based Gesture Recognition in Non-Ideal Conditions

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Signal Processing

Background:

• Surface electromyogram (sEMG) signals are crucial for intent recognition in human-machine interaction.
• Existing sEMG datasets often lack the variability of real-world conditions, limiting system robustness.
• Bridging the gap between lab settings and daily use is essential for practical sEMG applications.

Purpose of the Study:

• To introduce the SeNic benchmark dataset for sEMG intent recognition under non-ideal conditions.
• To provide a common platform for researchers to develop more robust sEMG-based systems.
• To quantify the impact of various non-ideal factors on sEMG recognition accuracy.

Main Methods:

• Collected 8-channel sEMG signals from 36 subjects performing 7 gestures.
• Introduced non-ideal factors including electrode shifts, individual differences, muscle fatigue, inter-day variations, and arm postures.
• Utilized a 3D-printed annular ruler for controlled electrode placement and shifts.
• Validated sEMG signal quality in temporal and frequency domains.

Main Results:

• Gesture recognition accuracy in ideal conditions confirmed the high quality of the SeNic dataset.
• Demonstrated significant adverse impacts of non-ideal factors on sEMG signal amplitudes and recognition accuracies.
• Quantified the degradation caused by electrode shifts, fatigue, and other real-world variables.

Conclusions:

• The SeNic dataset serves as a valuable benchmark for evaluating sEMG-based intent recognition systems in realistic scenarios.
• It highlights the critical need to address non-ideal factors for robust human-machine interaction.
• SeNic is freely available, fostering collaborative research and development in the sEMG community.