Dimensionality reduction for classification of object weight from electromyography.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Human-Computer Interaction

Background:

• Electromyography (EMG) measures muscle activity but generates complex, high-dimensional signals.
• EMG is crucial for human-machine interfaces, especially for analyzing hand movements.
• Predicting object properties from EMG alone presents a significant challenge.

Purpose of the Study:

• To develop an automated pipeline for predicting object weight using only EMG data during reach-grasp-lift tasks.
• To shift from manual feature engineering to automated feature extraction for EMG signal analysis.
• To evaluate the effectiveness of dimensionality reduction techniques and classification algorithms on pre-processed EMG signals.

Main Methods:

• An automated pipeline was developed to process pre-processed EMG signals.
• Various dimensionality reduction methods were applied to extract intrinsic EMG features.
• Classification algorithms, including k-Nearest Neighbors, were tested on the reduced-dimension data.
• A running-window analysis was employed to assess temporal performance.

Main Results:

• The Laplacian Eigenmap algorithm demonstrated superior performance among dimensionality reduction methods.
• Optimal classification accuracy of 88% F1 score (3-way classification) was achieved using Laplacian Eigenmaps and k-Nearest Neighbors.
• The developed method showed comparable results to studies utilizing both EMG and EEG.
• The running-window analysis indicated rapid information capture and stability in the EMG signal during object manipulation.

Conclusions:

• Automated feature extraction from EMG signals can effectively predict object weight in reach-grasp-lift tasks.
• Laplacian Eigenmaps combined with k-Nearest Neighbors provide a robust approach for EMG-based prediction.
• This EMG-only method offers a promising, efficient alternative for human-machine interface applications.