Computer Vision — Satellite Imagery Analysis
A computer vision pipeline that processes high-resolution satellite imagery to extract structured geospatial features used in exposure management and event response.
Detect + segment
Tasks
Flood + building
Perils
PostGIS + tile API
Output
Problem & use case
Geospatial teams needed timely, country-scale building footprints and flood extent maps that commercial vendors could not deliver fast enough.
Solution & engineering approach
A YOLOv8 + U-Net hybrid pipeline running on GPU workers, with tiled inference, post-processing, and direct push into a PostGIS feature store.
Architecture & data flow
- 01
Tile and preprocess multi-band satellite scenes
- 02
YOLOv8 detects buildings + infrastructure objects
- 03
U-Net segments water bodies and flood extents
- 04
Vectorize masks → GeoJSON / PostGIS
- 05
Serve features via tile API to GIS dashboards
Technology stack
Implementation status
Implemented — detection and segmentation pipeline with PostGIS feature store and tile API