Sequential Learning on sEMGs in Short- and Long-term Situations via Self-training Semi-supervised Support Vector...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Wearable sensing technology facilitates collecting unlabeled surface electromyogram (sEMG) data.
• Semi-supervised learning methods are crucial for utilizing unlabeled sEMG data with limited labeled data for supervised classification.
• High-performance motion control relies on machine learning-based supervised classifiers.

Purpose of the Study:

• To assess the impact of sequential learning of self-training support vector machine (ST-S3VM) on short- and long-term sEMG datasets.
• To evaluate the robustness of ST-S3VM under realistic conditions using public datasets.
• To compare ST-S3VM performance against other Support Vector Machine (SVM) classifiers.

Main Methods:

• Utilized two public datasets: one short-term and one long-term sEMG dataset.
• Implemented and compared ST-S3VM with four types of SVM classifiers, including ST-SVM, SVM, and S3VM.
• Evaluated classifier performance in both short-term and long-term scenarios.

Main Results:

• Sequential learning combined classifiers (ST-SVM and ST-S3VM) demonstrated superior performance compared to non-ST methods (SVM and S3VM) on both datasets.
• ST-S3VM achieved the highest performance in certain cases, while ST-SVM outperformed ST-S3VM in others.
• The study identified the potential for ST-S3VM to outperform ST-SVM, indicating room for improvement.

Conclusions:

• Sequential learning significantly enhances the performance of SVM classifiers for sEMG data analysis.
• ST-S3VM shows promise but requires further development to mitigate the influence of potentially detrimental unlabeled data.
• Future work will focus on refining ST-S3VM to better leverage unlabeled data and improve classification accuracy.