Related Experiment Video
Updated: Dec 6, 2025
08:00
Decoding Natural Behavior from Neuroethological Embedding
Published on: October 3, 2025
405
Analysis and Usage: Subject-to-subject Linear Domain Adaptation in sEMG Classification
Summary
Linear domain adaptation for biosignal applications requires user-specific calibration. This study reveals that source-target data correlation significantly impacts classification accuracy, suggesting non-linear approaches for low-correlation scenarios.
Area of Science:
- Biomedical Engineering
- Machine Learning
- Signal Processing
Background:
- Biosignal applications necessitate lengthy calibration to adapt pre-trained classifiers to new user data.
- Linear domain adaptation (DA) transfer learning methods are explored to reduce calibration time by transferring pooled source data to target data.
- Previous applications of linear DA to surface electromyogram (sEMG) data assumed linearity, which contradicts the typically non-linear nature of sEMG.
Purpose of the Study:
- To investigate the impact of source-target data correlation on 8-class forearm movement classification using linear DA approaches.
- To determine the conditions under which linear DA may lead to negative transfer due to non-linear characteristics of sEMG data.
Main Methods:
- Applied linear domain adaptation (DA) transfer learning techniques to surface electromyogram (sEMG) datasets.
- Analyzed the correlation between source and target sEMG data across different forearm movement classes.
- Evaluated the classification accuracy of an 8-class forearm movement classifier based on varying source-target correlations.
Main Results:
- A significant positive correlation was observed between classification accuracy and the source-target data correlation.
- The degree of source-target correlation was found to be dependent on the specific motion class being analyzed.
- Linear DA approaches demonstrated reduced effectiveness when the source-target correlation was low, indicating potential negative transfer.
Conclusions:
- The correlation between source and target data is a critical factor influencing the performance of linear DA in sEMG classification.
- Forearm movement classification accuracy is positively associated with higher source-target data correlation.
- Non-linear DA approaches are recommended when source-target correlation is low, particularly across different subjects or motion classes, to avoid negative transfer.
