Skip to content

Latest commit

 

History

History
95 lines (70 loc) · 2.74 KB

README.md

File metadata and controls

95 lines (70 loc) · 2.74 KB

🚀 Waste Classification using CNN with Transfer Learning

📌 Overview.

This project implements a CNN-based waste classification model using Transfer Learning with the VGG16 architecture. The model classifies waste into two categories: Organic and Recyclable, leveraging a two-phase training strategy (freeze-unfreeze).

🌟 Key Features.

  • ✅ Two-phase training strategy: Freeze VGG16 layers initially, followed by fine-tuning the last few layers.
  • 📈 High Performance: Achieves up to 98% accuracy on test data.
  • 🎨 Extensive Data Augmentation: Various augmentation techniques applied to improve model robustness.
  • 🔮 Future Potential: Plans to expand functionality, improve predictions, and cover more waste categories.

📁 Dataset.

The Waste Classification dataset was sourced from Kaggle. Key details include:

  • 🗂 Classes: Organic and Recyclable waste.
  • 📦 Structure: Pre-defined training and testing splits.
  • 🖼 Format: RGB images with dimensions of 224x224 pixels.

🛠️ Requirements.

Ensure the following libraries and frameworks are installed:

tensorflow keras numpy pandas opencv-python matplotlib tqdm kagglehub

🚀 Usage.

1️⃣ Clone the Repository

git clone https://github.com/Raviteja5469/CNN-model-for-Waste-Classification.git

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Run the Training Script

python train.py

📊 Model Performance.

Metric Accuracy
Training Accuracy ~95%
Validation Accuracy ~92%
Test Accuracy ~90%

🔍 Training Strategy.

  1. Initial Training: Freeze VGG16 layers.
  2. Fine-tuning: Unfreeze and train the last 5 VGG16 layers.
  3. Learning Rate Adjustment: Reduce on plateau.
  4. Data Augmentation: Apply transformations for robustness.

🔮 Future Improvements.

  • 🔄 Implement additional data augmentation techniques.
  • 📚 Experiment with other pre-trained models.
  • 🖥 Add real-time prediction capabilities.
  • 🏷 Expand to more waste categories.

👥 Contributing.

Contributions are welcome! Feel free to fork the repository, make changes, and submit a Pull Request.

📞 Contact

Platform Link
Author Seguri Raviteja
E-mail ravitejaseguri@gmail.com
GitHub Raviteja5469
LinkedIn Seguri Raviteja

🙌 Acknowledgments

  • 🏗 VGG16 Pre-trained Model
  • 📊 Waste Classification Dataset Creators
  • 🛠 TensorFlow and Keras Teams