In 2022, through our partnership with Cubeware GmbH, we led a pioneering research and development initiative exploring the practical application of Artificial Intelligence and Machine Learning within the public sector.
The project was commissioned by the Bavarian Government with the goal of creating a prototype platform capable of automatically classifying rural land and farmland using regularly updated satellite imagery. The vision was to provide government departments with a scalable way to understand land usage across vast geographical areas without relying on time-consuming manual analysis.
One of the project’s most significant challenges was that no off-the-shelf machine learning models existed that could accurately identify and classify the specific land types required for this use case. As a result, we designed and trained our own custom computer vision models from the ground up.
Success depended heavily on obtaining a high-quality training dataset. To overcome this challenge, we leveraged open-source geospatial datasets released by the U.S. Geological Survey (USGS), which had recently made available a comprehensive collection of labelled land-cover imagery as part of a research competition. These datasets provided the foundation for training our models to distinguish between a wide range of terrain types, including barren land, shrubland, forested areas, waterlogged regions, and agricultural fields.
Using these datasets, we developed a robust machine learning pipeline capable of analysing satellite imagery and assigning accurate classifications to individual land parcels. Throughout development, we set an ambitious target of achieving greater than 90% classification accuracy, ensuring the solution could demonstrate genuine operational value rather than remaining a purely academic exercise.
The resulting prototype successfully proved the viability of AI-driven land classification at scale and highlighted numerous real-world applications. Beyond identifying crop coverage and agricultural density, the technology demonstrated potential for environmental monitoring, land management, biodiversity analysis, infrastructure planning, and the detection of unusual land-use patterns. The project provided a compelling example of how modern AI techniques can transform vast volumes of satellite data into actionable insights for public sector organisations.
Key Outcomes
Custom machine learning models trained specifically for land-use classification.
Automated analysis of large-scale satellite imagery datasets.
Classification of multiple terrain types including agricultural land, forest, shrubland, barren earth, and waterlogged areas.
Targeted and achieved high levels of classification accuracy suitable for real-world deployment.
Demonstrated the potential for government agencies to make faster, more informed decisions using AI-powered geospatial intelligence.
This project remains one of our most innovative AI initiatives, showcasing how bespoke machine learning solutions can unlock valuable insights from complex environmental data at a national scale.
We’d love to have a chat to understand and discuss your project. Even if you’re still in the early idea phase. We’re happy to advise and guide you along the way.