2025A_DTIC_CNN_LSTM

Multi-Month Evaluation for EMG-Based Hand Gesture Recognition

This repository contains two evaluation scripts designed to assess the performance of deep learning models (CNN and CNN-LSTM) for hand gesture recognition (HGR) using electromyographic (EMG) signals over multiple months of data.

---

📁 Folder Structure

Expected dataset organization:

EMG-EPN-612 dataset/
├── Mes0/
│   └── trainingJSON/
│       └── userX.json
├── Mes1/
│   └── trainingJSON/
│       └── userY.json
...

---

🧠 Scripts Overview

1. multiFolderEvaluation.m – CNN Model Evaluation

This script evaluates a trained CNN model on all Mes*/trainingJSON folders. For each month and each user, it computes:

• Classification accuracy
• Recognition accuracy
• Overlapping factor
• Average processing time per frame

It generates:

• Per-user performance evolution plots (average of training and validation).
• Global plots across all users.
• Summary tables exported to Excel.

✅ Requirements

• Shared.m – constants and preprocessing functions
• evalRecognition.m – metric evaluation
• Trained CNN model (.mat file), e.g., Models/model_cnn_final.mat

---

2. multiFolderEvaluation_LSTM.m – CNN-LSTM Model Evaluation

This script evaluates a trained CNN-LSTM model following the same logic as the CNN version. It performs:

• Per-month evaluation (train/validation)
• Per-user performance plots
• Global performance plots
• Excel export of all aggregated metrics

✅ Requirements

• Shared.m
• evalRecognition.m
• Trained CNN-LSTM model (.mat file), e.g., ModelsLSTM/model_cnn-lstm_final.mat

---

📊 Metrics Computed

For each user and each month:

• Classification Accuracy: correct gesture type prediction
• Recognition Accuracy: correct gesture recognition (including timing)
• Overlapping Factor: alignment accuracy between prediction and ground truth
• Processing Time: average time per spectrogram prediction

---

📤 Outputs

• Console: average + standard deviation metrics per month

• Figures:

  • Per-user evolution of metrics across months
  • Global evolution of averages across users

• Excel file:

  • Resumen-Metricas-AVG-CNN.xlsx (for CNN)
  • Resultados_CNN-LSTM.xlsx (for CNN-LSTM)

Each file contains four sheets:

• Classification
• Recognition
• Overlap
• Processing_Time

---

⚙️ How to Run

1. Ensure all required files (Shared.m, evalRecognition.m, model .mat) are on your MATLAB path.

2. Set the baseDataDir and modelFile variables if needed.

3. Run the corresponding script:

   multiFolderEvaluation()        % For CNN model
   multiFolderEvaluation_LSTM()   % For CNN-LSTM model

---

👨‍💼 Authors

Developed as part of a research project at the Alan Turing Artificial Intelligence Laboratory – Escuela Politécnica Nacional.

---

📄 License

This project is for academic and research purposes only. Contact the authors for use in other contexts.