This project was developed during my internship to classify waste images into 6 categories using image classification and transfer learning with EfficientNetV2B2.
The goal is to support smart waste management by automating waste type detection and encouraging eco-friendly disposal.
- Kaggle Trash Type Image Dataset
- Contains labeled images of:
cardboard, glass, metal, paper, plastic, trash
- Set up Google Colab + Kaggle API
- Downloaded and explored dataset
- Created data generators
- Built base model using EfficientNetV2B2
- Trained for 2 epochs (basic run)
- Re-trained the model for 10 epochs
- Applied data augmentation (flip, rotate, zoom, shift)
- Plotted training vs validation accuracy and loss
- Organized results in separate notebook
- Fine-tuned the model (unfrozen base layers)
- Trained on GPU for better accuracy
- Tested with real trash images
- ✅ Added unique prediction feature:
- Shows eco-friendly tip 🌱
- Suggests proper disposal 🗑️
| File | Description |
|---|---|
Week3_Garbage_Classifier.ipynb |
Final notebook with training + test |
graph.png |
Accuracy/loss plot |
prediction.png |
Model output with tip & disposal info |
🧠 Model predicted class
🌱 Sustainability tip
🗑️ Correct disposal advice
- Google Colab with T4 GPU
- TensorFlow / Keras
- EfficientNetV2B2 (transfer learning)
Built with 💻 + 💡 during Shell | Artificial Intelligence x Edunet Foundation Internship