HrutikAdsare/waste-detection-yolov8

Waste Detection & Classification – YOLOv8

This model detects and classifies different types of waste to support smart waste segregation, recycling automation, and sustainability projects. It was trained using Ultralytics YOLOv8 on a custom waste dataset with ~250 images per class.

🔎 Detected Classes

• —Cardboard
• —E-waste
• —Glass
• —Medical waste
• —Metal
• —Organic waste
• —Paper
• —Plastic

🧠 Model Details

• —Architecture: YOLOv8
• —Framework: Ultralytics (PyTorch)
• —Task: Object Detection
• —Input: RGB images
• —Output: Bounding boxes + class labels
• —Training Data: ~250 images per class
• —Total Classes: 8

📊 Evaluation (Validation Set – Normalized Confusion Matrix Insights)

• —Organic: ~96% recall
• —Metal: ~81% recall
• —Paper: ~83% recall
• —Plastic: ~63% recall
• —Medical waste: ~54% recall
• —Cardboard: ~76% recall
• —E-waste: ~75% recall
• —Glass: ~60% recall (approx)

The model performs strongly on organic, metal, and paper categories. Performance on medical waste and plastic can be improved with more diverse training samples.

📈 Training Curves & Confusion Matrix

You can find:

• —Precision-Recall curves
• —F1 curve
• —Confusion matrix
• —Training visualizations

inside the repository files.

🚀 How to Use

Install Ultralytics

pip install ultralytics

Run Inference

from ultralytics import YOLO

model = YOLO("best.pt")  # path to the downloaded weights
results = model("test.jpg", conf=0.25)
results[0].show()

Batch Inference

results = model("path/to/images/", save=True)

🧪 Example Use Cases

• —Smart waste segregation systems
• —Recycling automation
• —Smart bins
• —Campus or city-level waste monitoring
• —Sustainability & eco-tech projects

⚠️ Limitations

• —Performance may drop on blurry or low-light images
• —Medical and plastic waste classes may need more data for higher accuracy
• —Not suitable for safety-critical decisions without human verification

📄 License

MIT License

👤 Author

Hrutik Adsare If you use this model, consider giving it a ⭐