This project implements Oil Spill Detection from satellite imagery using semantic segmentation with two state-of-the-art deep learning models: U-Net and DeepLabV3.
The goal is to identify and segment oil spills in the ocean from remote sensing data, enabling early environmental hazard monitoring and prevention.
- Two segmentation architectures: U-Net and DeepLabV3 (ResNet backbone)
- Binary segmentation: Oil Spill (1) vs Background (0)
- Data preprocessing: Image and mask resizing, normalization
- Training and evaluation with:
- Pixel accuracy
- IoU (Intersection over Union)
- Dice/F1 score
- Prediction visualization for qualitative results
- Confusion Matrix to analyze model performance
The project assumes you have satellite images and binary masks: