Classification of EMG signals with CNN features and voting ensemble classifier.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neuroscience

Background:

• Electromyography (EMG) signals are crucial for controlling prosthetic hands.
• Efficiently classifying hand gestures from EMG signals remains a significant challenge.
• Improving classification accuracy requires effective feature extraction and classification strategies.

Purpose of the Study:

• To develop a robust framework for accurate hand gesture classification using EMG signals.
• To reduce redundancy in feature extraction by employing Convolutional Neural Networks (CNNs).
• To enhance classification performance by integrating CNN features with a k-nearest neighbor (KNN) ensemble classifier.

Main Methods:

• EMG signal features were extracted using a CNN to minimize time and frequency domain redundancies.
• Extracted CNN features were fed into a KNN classifier with varying numbers of neighbors.
• An ensemble of KNN classifiers was created and combined using a hard voting mechanism.

Main Results:

• The proposed framework achieved high classification accuracy on benchmark datasets.
• Achieved $$ classification accuracy on the CapgMyo database.
• Achieved $$ classification accuracy on the Ninapro DB4 database.

Conclusions:

• The combination of CNN feature extraction and a KNN ensemble classifier effectively improves EMG-based hand gesture classification.
• This approach addresses challenges in EMG signal processing for prosthetic hand control.
• The framework demonstrates significant potential for enhancing the functionality of prosthetic devices.