An Investigation of Dimensionality Reduction Techniques for EMG-based Force Estimation

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Surface electromyogram (sEMG) signals provide insights into muscle activity.
• Accurate force estimation is crucial for prosthetic control and rehabilitation.
• High-density sEMG (HD-sEMG) offers richer spatial information compared to traditional sEMG.

Purpose of the Study:

• To estimate wrist-induced force using HD-sEMG signals from biceps brachii and brachioradialis muscles.
• To evaluate the impact of different artificial neural network (ANN) hidden layer sizes on force estimation accuracy.
• To assess the effectiveness of Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction in force estimation.

Main Methods:

• Acquisition of HD-sEMG signals during isometric elbow flexion.
• Extraction of time and frequency domain features from HD-sEMG signals.
• Application of ANNs with varying hidden layer sizes for force estimation.
• Implementation of PCA and t-SNE for feature set dimensionality reduction.

Main Results:

• ANN models successfully estimated wrist-induced force from HD-sEMG features.
• Feature extraction in time and frequency domains was effective for force estimation.
• Dimensionality reduction techniques (PCA and t-SNE) influenced force estimation accuracy.
• Optimizing ANN hidden layer size impacted model performance.

Conclusions:

• HD-sEMG signals combined with ANN are a viable method for wrist force estimation.
• Feature engineering and dimensionality reduction are important considerations for improving force estimation models.
• Further research can explore advanced feature extraction and machine learning techniques for enhanced accuracy.