Electromyography Signal Analysis and Classification using Time-Frequency Representations and Deep Learning.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electromyography (EMG) signal analysis is vital for rehabilitation and motor control.
• Accurate classification of EMG signals aids in developing advanced prosthetics and human-computer interfaces.
• Exploring novel representations and deep learning architectures can improve EMG classification performance.

Purpose of the Study:

• To investigate time-frequency representations of EMG signals for improved classification.
• To develop and compare conventional and deep learning models for EMG signal classification.
• To evaluate the impact of data augmentation on classification accuracy.

Main Methods:

• Recorded single-channel surface EMG signals for forearm flexion and extension differentiation.
• Utilized various time-frequency EMG representations to train conventional and deep learning models.
• Compared pre-trained convolutional neural network (CNN) models (GoogLeNet, SqueezeNet, AlexNet) and traditional classifiers.
• Applied data augmentation techniques to raw EMG signals and their time-frequency representations.

Main Results:

• GoogLeNet, SqueezeNet, and AlexNet achieved classification accuracies of 92.71%, 90.63%, and 87.5%, respectively.
• Data augmentation improved GoogLeNet's accuracy to 96.88%.
• The proposed approach showed superior performance on a 10-class EMG dataset and with traditional classifiers.

Conclusions:

• Time-frequency representations combined with deep learning models significantly enhance EMG signal classification.
• Data augmentation is an effective strategy for improving the robustness and accuracy of EMG classification models.
• The findings support the application of these advanced methods in clinical rehabilitation and motor control systems.