🗑️ Waste Classification Using Transfer Learning & Fine-Tuning

An image classification system that automatically categorizes waste into Organic (O) and Recyclable (R) using Convolutional Neural Networks, built with VGG16 Transfer Learning and Fine-Tuning in TensorFlow/Keras.

---

📌 Project Overview

Improper waste segregation is a major environmental challenge. This project builds a deep learning model that can automatically classify waste images into organic and recyclable categories, supporting smarter and more efficient waste management systems.

---

🏗️ Model Architecture & Approach

The project follows a two-stage transfer learning strategy using the VGG16 pre-trained network:

1. Feature Extraction Model

   • VGG16 base (pre-trained on ImageNet) used as a frozen feature extractor.
   • Custom dense classification layers added on top.
   • Only the new top layers are trained.

2. Fine-Tuned Model

   • Top convolutional layers of VGG16 are unfrozen.
   • Model is retrained with a low learning rate to adapt pre-trained features to the waste classification task.
   • Achieves improved accuracy over the feature extraction model alone.

---

🧰 Tech Stack

• Language: Python
• Deep Learning Framework: TensorFlow, Keras
• Pre-trained Model: VGG16 (ImageNet weights)
• Data Handling: ImageDataGenerator (data augmentation & preprocessing)
• Evaluation & Visualization: Scikit-learn, Matplotlib
• Environment: Jupyter Notebook

---

📊 Workflow

1. Data Preprocessing — Image resizing, rescaling, and augmentation using ImageDataGenerator.
2. Train/Validation/Test Split — Organized into separate generators for training, validation, and testing.
3. Feature Extraction — Building and training a classifier on top of frozen VGG16 layers.
4. Fine-Tuning — Unfreezing top VGG16 layers and retraining with a low learning rate.
5. Evaluation — Accuracy/loss curves, classification report, and visual predictions on test images.

---

📈 Results

• Model performance evaluated using accuracy and loss curves for both the feature extraction and fine-tuned models.
• Fine-tuning improved classification accuracy compared to the feature extraction stage alone.
• Visual predictions on test images confirm the model's ability to correctly distinguish between Organic and Recyclable waste.

---

🚀 How to Run

1. Clone this repository.
2. Open Final Proj-Classify Waste Products Using TL FT.ipynb in Jupyter Notebook or Google Colab.
3. Install the required libraries:

   pip install tensorflow numpy scikit-learn matplotlib

4. Run all cells in order (Kernel → Restart & Run All).
5. View outputs including model summary, accuracy/loss curves, and test predictions.

---

🎓 Acknowledgment

This project was completed as part of the IBM AI Engineering Professional Certificate on Coursera, developed by IBM Skills Network.

---

👩‍💻 Author

Umaira Tahir