High energy spectrogram with integrated prior knowledge for EMG-based locomotion classification.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rehabilitation Technology

Background:

• Electromyogram (EMG) signal representation is vital for classifying human locomotion and transitions.
• Accurate EMG signal classification is essential for developing advanced assistive devices.

Purpose of the Study:

• To develop and evaluate a spectrogram-based approach for classifying EMG signals corresponding to different locomotion modes.
• To assess the impact of feature selection and prior knowledge on classification accuracy and computational efficiency.

Main Methods:

• EMG signals from seven leg muscles during walking, stair ascent, and transitions were analyzed.
• Spectrograms were computed, summed into histograms, and classified using if-else rules with a matching score.
• Leave-one-out cross-validation was employed to evaluate classification error, with and without prior locomotion knowledge.

Main Results:

• Initial classification error was below 20%, reduced to below 11% with prior knowledge of locomotion type.
• Removing three muscles decreased classification accuracy but reduced computation time by 42.8%.
• The spectrogram-based method demonstrated robust classification across different locomotion modes.

Conclusions:

• A spectrogram and histogram-based method effectively classifies EMG signals for locomotion modes.
• Prior knowledge significantly improves classification accuracy, while feature reduction impacts accuracy and computation time.
• This technique shows promise for real-time control of multi-mode assistive devices.