Using free Sentinel-1 synthetic aperture radar data and machine learning to identify structural changes in bridges and roads — before failure occurs.
Simulated output from a Sentinel-1 analysis pipeline. Each dot represents a detected anomaly in SAR backscatter across two satellite passes — a potential indicator of structural change or ground movement.
The pipeline runs end-to-end on free Sentinel-1 data. No commercial satellite access required. Each processing step follows peer-reviewed methodology from ESA and published infrastructure monitoring research.
Sentinel-1 IW GRD scenes pulled automatically via Copernicus Data Space API on each 12-day satellite pass. Two co-registered scenes per analysis window.
ESA SNAP toolbox pipeline: orbit file application → thermal noise removal → radiometric calibration → terrain correction using SRTM DEM.
Pixel-wise backscatter differencing across temporal stack. Log-ratio change maps computed. Coherence loss used as secondary indicator for structural deformation.
Physics-guided U-Net classifies change pixels into severity tiers. Trained on publicly available labeled SAR datasets. Auxiliary inputs: bridge geometry, soil type, slope angle.
Anomalies above threshold trigger structured alerts with coordinates, severity score, and change magnitude. Output served as GeoJSON and visualised on this page automatically.
This project documents my learning journey through satellite remote sensing and applied machine learning — from first principles through to a working, automated infrastructure monitoring pipeline.
Every tool used is free and open. Every methodology step references published research. The goal is to show what's possible when satellite data, AI, and engineering curiosity are combined without a commercial budget.
I'm interested in collaborating with engineers, researchers, and infrastructure operators who want to explore what SAR-based monitoring could do for their work.