Optimized Signal Acquisition and Advanced AI for Robust 1D EMG Classification: A Comparative Study of Machine...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Electromyography (EMG) signal classification is vital for prosthetics, rehabilitation, and human-machine interfaces.
• Challenges include noise, non-stationarity, and inter-subject variability in EMG data.
• Optimizing signal acquisition and AI models is crucial for reliable EMG analysis.

Purpose of the Study:

• To compare machine learning (ML), deep learning (DL), and reinforcement learning (RL) for 1D EMG signal classification.
• To systematically evaluate the impact of signal acquisition parameters on EMG classification performance.
• To provide empirical guidance for selecting optimal acquisition settings and AI architectures for practical EMG systems.

Main Methods:

• Comparative analysis of ML, DL, and RL algorithms on synthetic and real-world EMG datasets.
• Systematic evaluation of signal acquisition parameters including quantization (8-10 bit) and sampling rate (2000 Hz).
• Performance assessment using metrics like classification accuracy.

Main Results:

• Optimal signal fidelity and data efficiency achieved with 8-10 bit quantization and 2000 Hz sampling rate.
• Ensemble methods (Gradient Boosting, Voting Ensemble) and DL architectures (LSTM, Transformer) showed high accuracy (up to 100% and 96.3%) on real EMG data.
• Reinforcement learning (Deep Q-Networks) achieved 100% accuracy on synthetic data, demonstrating potential for complex bio-signal representation.

Conclusions:

• Meticulous optimization of preprocessing and robust AI models significantly enhance EMG classification accuracy.
• The study provides empirical guidance for selecting optimal acquisition parameters and AI architectures for EMG analysis.
• Findings have direct implications for advancing prosthetic control and rehabilitation technologies.