🌟 Material Stream Identification System 🌟

Team Members: Menna Reda, Fatma Ibrahim, Sara Mohmed, Youssef Nasser, Mohammed Moustafa
Project: Automated Material Stream Identification using Machine Learning
Instructor: Hanaa Mobarez TA: Sara Ahmed Elnady

---

♻️ Project Overview

The Material Stream Identification (MSI) System is an end-to-end machine learning application designed to classify post-consumer waste into distinct material categories. This system emphasizes the complete ML pipeline: from data preprocessing and feature extraction to classifier training and real-time deployment.

The system currently classifies waste into seven classes:

• Glass
• Paper
• Cardboard
• Plastic
• Metal
• Trash
• Unknown (out-of-distribution or blurred inputs)

---

♻️ Features

• Data Preprocessing & Augmentation:

  • Resize, normalize, and clean images.
  • Apply augmentation (rotation, flipping, scaling, color jitter) to increase dataset size by ≥30%.

• Feature Extraction:

  • Convert raw images into fixed-length numerical feature vectors using a Convolutional Neural Network (CNN).
  • A pre-trained CNN is used to automatically extract high-level discriminative features directly from images.

• Machine Learning Models:

  • SVM Classifier: Trained on extracted features with hyperparameter tuning.
  • k-NN Classifier: Trained with different values of k and weighting schemes.
  • Best-performing model selected for real-time classification.

• Real-Time Deployment:

  • Processes live camera frames.
  • Displays the predicted class in real-time using OpenCV.

---

♻️ Project Structure

Material-Stream-Identification/
│
├── dataset/                 # Original dataset (ignored in Git)
├── dataset_augmented/       # Augmented dataset (ignored in Git)
├── features/                # Feature vectors and labels
│   ├── X_features.npy
│   └── y_labels.npy
├── models/                  # Trained models
│   ├── svm_best.pkl
│   └── knn_best.pkl
├── src/                     # Training and preprocessing scripts
│   ├── preprocess.py
│   ├── extract_features.py
│   ├── train_svm.py
│   └── train_knn.py
├── app/                     # Real-time application
│   ├── realtime_classifier.py
│   ├── model_loader.py
│   └── utils.py
├── notebooks/               # Experimentation notebooks
│   ├── feature_experiments.ipynb
│   ├── svm_testing.ipynb
│   └── knn_testing.ipynb
├── docs/                    # Project report
│   └── report.pdf
├── requirements.txt         # Python dependencies
└── main.py                  # Entry point for the project

---

Installation

1. Clone the repository:

   git clone https://github.com/1Menna/Material-Stream-Identification-System.git
   cd Material-Stream-Identification

2. Install dependencies:

   pip install -r requirements.txt

Note: dataset/ and dataset_augmented/ are not included due to size. Add your local dataset manually.

---

Dependencies

• numpy
• scikit-image
• scikit-learn
• OpenCV (opencv-python)
• tensorflow
• joblib

---

Contributing

• Fork the repository and create a new branch for your feature.
• Ensure your code follows the project structure and naming conventions.
• Submit a pull request for review before merging.

---

License

This project is for academic purposes. Do not use without permission.