Reinforcement Learning for sEMG Denoising

Teaching an agent to clean noisy biological signals.

Surface electromyography (sEMG) signals are critical for prosthetic control systems, but they are often corrupted by motion artifacts and ECG interference.

This project explores a reinforcement learning approach to signal denoising by combining Deep Q-Networks with adaptive wavelet thresholding.

Instead of relying on fixed filtering rules, the agent learns how to optimize denoising strategies dynamically based on signal conditions. The result is a system capable of removing significant noise while preserving important muscle activation information.

• —85% Noise Suppression
• —Preserved muscle activation patterns
• —Generalized across multiple noise levels
• —Improved signal quality for downstream applications

Python · TensorFlow · Reinforcement Learning · DQN · Wavelet Transform · Signal Processing