The problem
The freely available ESA imagery is low resolution. Detecting individual containers and warehouse changes at that resolution, reliably enough to report on, is the entire difficulty of the problem.
Engagement detail
- CLIENT
- Market intelligence firm
- INDUSTRY
- Industrial & supply chain intelligence
- DISCIPLINE
- Computer Vision
PythonRESA satellite imageryFmaskTensorFlowGISQGIS
What we built
- Established a structured database of industrial buildings and warehouses across the target region as the spatial backbone for analysis.
- Built an object detection system that exceeded 80% accuracy despite the resolution limits of the source imagery.
- Designed a repeatable monthly geographic monitoring process to surface changes in warehouse and container locations as they happen.
- Applied Fmask and TensorFlow for pre-processing and enhancement, materially improving the usable signal in each scene.
- Used GIS and QGIS for spatial analysis and visual reporting, so findings arrived as maps rather than tables.
Next case study
Data & BIExecutive reporting platform for VZ Track
Read itAvailable for new engagements
Have a problem shaped like this one?
We will tell you what transfers from this engagement to yours, what does not, and what we would do differently now.