SASG-DA: Sparse-Aware Semantic-Guided Diffusion Augmentation For Myoelectric Gesture Recognition

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Machine Interaction

Background:

• Surface electromyography (sEMG)-based gesture recognition is vital for human-machine interaction (HMI), especially in rehabilitation and prosthetics.
• Deep learning models for sEMG recognition struggle with limited training data, leading to overfitting and poor generalization.
• Existing data augmentation methods may generate redundant samples, limiting their effectiveness.

Purpose of the Study:

• To propose a novel diffusion-based data augmentation approach, Sparse-Aware Semantic-Guided Diffusion Augmentation (SASG-DA), to address data scarcity in sEMG gesture recognition.
• To enhance the faithfulness and diversity of augmented sEMG data for improved deep learning model performance.
• To mitigate overfitting and improve the generalization capabilities of sEMG recognition systems.

Main Methods:

• Developed SASG-DA, a diffusion-based augmentation technique.
• Introduced Semantic Representation Guidance (SRG) for enhanced generation faithfulness using fine-grained, task-aware semantic representations.
• Implemented Gaussian Modeling Semantic Sampling (GMSS) for flexible and diverse sample generation.
• Incorporated Sparse-Aware Semantic Sampling to target underrepresented data regions for improved utility.

Main Results:

• SASG-DA significantly outperformed existing data augmentation methods on benchmark sEMG datasets (Ninapro DB2, DB4, DB7).
• The proposed approach effectively generated both faithful and diverse sEMG samples.
• Experimental results demonstrated mitigation of overfitting and improved recognition performance and generalization.

Conclusions:

• SASG-DA offers an effective solution for data augmentation in sEMG gesture recognition.
• The method enhances deep learning model performance by providing high-quality, diverse training data.
• This approach holds significant potential for advancing HMI applications in rehabilitation and prosthetic control.