← Back to projects
data science · July 2025
Satellite Land-Use Segmentation
Placeholder — semantic segmentation of satellite imagery for land-use classification.
Python PyTorch Segmentation GeoSpatial
outcome
IoU 0.76
Problem
Placeholder — aerial imagery with five land-use classes; heavy class imbalance.
Approach
Placeholder — U-Net variant with focal loss; tile-based training pipeline.
Results
Placeholder — 12-point IoU lift over baseline, inference under 50ms per tile.