sEMG-Based Inter-Session Hand Gesture Recognition via Domain Adaptation with Locality Preserving and Maximum Margin

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Surface electromyography (sEMG)-based gesture recognition shows high performance within a single session.
• Inter-session performance degrades significantly due to data distribution shifts, impacting user experience.
• Robust models are needed to mitigate these distribution differences for reliable gesture recognition.

Purpose of the Study:

• To develop a novel algorithm, locality preserving and maximum margin criterion (LPMM), for robust inter-session sEMG gesture recognition.
• To minimize data distribution discrepancies between different recording sessions.
• To enhance the overall accuracy and reliability of sEMG-based gesture recognition systems.

Main Methods:

• Proposed the locality preserving and maximum margin criterion (LPMM) algorithm for inter-session gesture recognition.
• Integrated domain alignment to preserve feature neighborhood structure and reduce class overlap.
• Employed pseudo-label selection and iteration result selection to prevent accuracy loss from mislabeled samples.

Main Results:

• Evaluated LPMM on two widely used EMG databases, achieving mean accuracies of 98.46% and 71.64%.
• Demonstrated superior performance compared to existing state-of-the-art domain adaptation methods.
• Successfully minimized data distribution shifts, enhancing inter-session recognition accuracy.

Conclusions:

• The proposed LPMM algorithm effectively addresses the challenge of data distribution shifts in inter-session sEMG gesture recognition.
• LPMM offers a robust solution for improving user experience by ensuring consistent performance across sessions.
• This method represents a significant advancement in domain adaptation for biomedical signal processing.