README

Most of the code is under NDA, however a preview of the code used can be found in this repository

Overview

This document provides instructions for using our research code, specifically focused on training and evaluating models for surface electromyography (sEMG) signal classification. The codebase includes scripts for data collection, processing, model training, and evaluation.

Table of Contents

• Data Collection
• Data Loading and Processing
• Model Training and Evaluation
• Model Explanations
• Grasp Modes

Data Collection

Data is collected using an Arduino setup, which captures sEMG signals. The dataCollection.py script is used to read and store this data.

Data Loading and Processing

datasetLoading.py

• Load and preprocess sEMG data using the sEMGDataset class.
• ComputeBasicFeatures transform is applied for initial feature extraction.

Model Training and Evaluation

Prerequisites

• Python 3.x
• Libraries: PyTorch, NumPy, Pandas (complete list in requirements.txt)
• sEMG dataset in the specified format

Setup

1. Install Python 3.x and all required libraries.
2. Clone the repository and navigate to the project directory.
3. Prepare your dataset according to the format used in dataCollection.py.

Training

1. Data Loading: Utilize the sEMGDataset class from datasetLoading.py to load your dataset.
2. Preprocessing: Apply the ComputeBasicFeatures for initial data processing.
3. Model Initialization: Initialize the HybridModel from modeling/hybridModel.py.
4. Training Configuration: Set up the training environment using the Adam optimizer with a learning rate of 0.001.
5. Execute Training: Run the training loop for the desired number of epochs (default is 10).
6. Monitoring: Monitor training progress and performance on a validation set.

Evaluation

• After training, evaluate the model's performance on a separate test dataset.
• Document the model's accuracy, precision, recall, and F1-score.

Saving and Utilizing the Model

• Instructions for saving the trained model to a file.
• Guidelines for loading the model for future inference tasks.

Model Explanations

LSTM Model

• The LSTM model is designed to handle sequential data, capturing temporal dependencies in sEMG signals.

Hybrid Model (CNN-LSTM)

• This model combines Convolutional Neural Networks (CNNs) for feature extraction with LSTM layers for sequence modeling, suitable for complex sEMG patterns.

Grasp Modes

The system classifies sEMG signals into five grasp modes:

1. Cylindrical Grasp
2. Tip Grasp
3. Hook Grasp
4. Palmar Grasp
5. Lateral Grasp

Each mode has distinct applications in prosthetic control and robotics.

---

For further inquiries or support, please contact Sebastian Levy at sebastianlevy@cmu.edu

---