Optimizing RNNs for EMG Signal Classification: A Novel Strategy Using Grey Wolf Optimization.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Accurate electromyographic (EMG) signal classification is crucial for developing advanced biomedical applications and prosthetics.
• Recurrent neural networks (RNNs) have shown promise in processing sequential data, making them suitable for analyzing dynamic EMG signals.

Purpose of the Study:

• To evaluate and compare the performance of different RNN architectures (LSTM, GRU, Bidirectional RNN) against Support Vector Machines (SVMs) for classifying EMG signals.
• To assess both the accuracy and classification speed of these models for five distinct right upper extremity movements.

Main Methods:

• EMG signals were preprocessed using a Butterworth filter and segmented into 250 ms windows with a 190 ms overlap.
• Grey Wolf Optimization was employed to tune the parameters of GRU, LSTM, and Bidirectional RNN architectures.
• Performance was evaluated by comparing classification accuracy and response times against SVMs.

Main Results:

• All evaluated RNN architectures achieved 100% classification accuracy, surpassing the 93% accuracy of SVMs in the initial phase.
• Long Short-Term Memory (LSTM) demonstrated the fastest classification speed at 0.12 ms, followed closely by GRU and Bidirectional RNNs.
• In a second phase, RNNs maintained high accuracy (96.38%-98.46%), while SVM performance was not detailed.

Conclusions:

• Recurrent neural networks, particularly LSTM, offer superior accuracy and significantly faster classification speeds for EMG signal analysis compared to traditional methods like SVM.
• The findings underscore the potential of RNNs for real-time applications in areas such as prosthetic control and human-computer interaction.