Knowledge Transfer to Improve sEMG Simultaneous Proportional Movement Detection: A Data Transformation Approach

Simultaneous proportional detection (SPD), based on surface electromyography (sEMG) signals, is a promising approach that achieves higher performance when trained on complex movements. However, collecting such training data is time-consuming for the user. To enable real-world applications, it is crucial to balance precision and ease of use. This study proposes a data transformation approach to enhance detection performance by transferring knowledge from a pre-existing dataset of complex movements to a simpler movement dataset. So, instead of asking a user to perform a variety of movements each session, a pre-recorded set of complex wrist movements data, including star-shaped trajectories, is used to add richness to a simpler set of wrist movement data requested from the user to perform movements only in X and Y. A linear transformation model is proposed to transfer movement knowledge between datasets, improving accuracy while maintaining the advantages of transfer learning. Experimental results confirm that the proposed knowledge transformation model improves R² performances by 5-10%. Additionally, the approach reduces data acquisition time, enhancing practicality for real-world applications in human-machine interaction, while being effective for both inter-session and inter-subject scenarios.