MeghanaK25/sar-oil-slick-detection

SAR Oil Slick Detection Model

Model Description

This model is a ResNet-based deep learning model trained to detect oil slicks in Synthetic Aperture Radar (SAR) satellite imagery. It's part of an end-to-end AI pipeline for maritime monitoring and illegal discharge detection.

Pipeline Integration

This model is integrated into a comprehensive maritime monitoring system:

1. Data Ingestion: Airflow scheduler ingests ship AIS data every few minutes
2. Anomaly Detection: AIS anomaly model flags suspicious ship behavior
3. Satellite Data: For flagged zones, Sentinel-1 radar imagery is automatically fetched (±12 hours)
4. Preprocessing: Automated preprocessing using pyroSAR and snappy (no manual SNAP GUI required)
5. Oil Slick Detection: This SAR ML model outputs oil slick masks with confidence scores
6. Spatio-temporal Fusion: If oil slick and ship anomaly overlap within 10km and ±12h, ML Alert is issued
7. Real-time Dashboard: Alerts are pushed to Streamlit dashboard with live map and severity index

Model Architecture

• Base Model: ResNet (Convolutional Neural Network)
• Task: Binary classification/segmentation for oil slick detection
• Input: SAR satellite imagery (224x224 pixels)
• Output: Oil slick masks with confidence scores

Training Details

• Batch Size: 8
• Epochs: 15
• Learning Rate: 0.001
• Image Size: 224x224 pixels
• Training Samples: 1,574
• Validation Samples: 1,615

Performance

Based on training history:

• Final Training Loss: ~0.34
• Final Validation Loss: ~0.19
• Final Training Accuracy: ~85%
• Final Validation Accuracy: ~95%

Files Included

• model.pth: PyTorch model weights
• training_history.pkl: Complete training metrics and hyperparameters
• config.json: Model configuration
• inference.py: Sample inference script

Usage

import torch
import pickle
from PIL import Image
import torchvision.transforms as transforms

# Load model
model = torch.load('model.pth', map_location='cpu')
model.eval()

# Load preprocessing transforms (adjust based on your training setup)
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485], std=[0.229])  # Adjust for SAR imagery
])

# Inference
def predict_oil_slick(image_path):
    image = Image.open(image_path)
    input_tensor = transform(image).unsqueeze(0)
    
    with torch.no_grad():
        output = model(input_tensor)
        confidence = torch.sigmoid(output).item()
    
    return confidence

# Example usage
confidence_score = predict_oil_slick('sar_image.png')
print(f"Oil slick confidence: {confidence_score:.4f}")

Data Sources

• Sentinel-1 SAR Data: European Space Agency's Sentinel-1 satellite constellation
• Preprocessing: pyroSAR and snappy for automated SAR data processing

Applications

• Maritime pollution monitoring
• Illegal oil discharge detection
• Environmental compliance enforcement
• Ocean pollution assessment
• Real-time maritime surveillance

Citation

If you use this model in your research, please cite:

@misc{sar_oil_slick_model,
  title={SAR Oil Slick Detection Model for Maritime Monitoring},
  author={Meghana K},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/MeghanaK25/sar-oil-slick-detection}
}

License

This model is released under the Apache 2.0 License.

Contact

For questions or collaborations regarding this model or the maritime monitoring pipeline, please reach out through the Hugging Face model page or create an issue in the repository.

---

Note: This model is designed for research and environmental monitoring purposes. For production deployment in critical maritime surveillance applications, additional validation and testing may be required.