A novel sEMG data augmentation based on WGAN-GP.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Surface electromyography (sEMG) signal classification is crucial for advanced mechanical prostheses.
• Generalist databases are needed to improve the accuracy and robustness of sEMG-based movement classification.
• Synthetic data generation can augment existing datasets, enhancing their diversity and utility.

Purpose of the Study:

• To propose a generative adversarial network (GAN) variant for synthesizing sEMG biosignals.
• To evaluate the effectiveness of GAN-generated sEMG data in improving movement classification accuracy.
• To compare the proposed method with traditional techniques like Magnitude Warping and Scaling.

Main Methods:

• Utilized a variant of generative adversarial networks (GANs) to create synthetic sEMG biosignals.
• Employed a convolutional neural network (CNN) for classifying movements based on sEMG data.
• Integrated 200 synthetic samples per movement into the training dataset for classification.

Main Results:

• The inclusion of synthetic sEMG samples led to a notable 4.07% increase in movement classification accuracy.
• The GAN-based synthetic data generation approach demonstrated superior performance compared to Magnitude Warping and Scaling.
• The proposed method effectively enhanced the generalizability of the sEMG database.

Conclusions:

• Generative adversarial networks offer a powerful tool for creating synthetic sEMG data to improve prosthetic control.
• Synthetic data augmentation using GANs enhances the performance and robustness of machine learning models for sEMG classification.
• This approach provides a viable strategy for developing more sophisticated and accurate mechanical prostheses.