Commissioned by the Swiss Federal Office of Energy, Adrian F. Meyer worked as Lead Engineer and Data Scientist to develop a workflow for automatically identifying photovoltaic and solar-thermal installations in aerial imagery. Processing combines georeferenced image preparation, reference annotation, deep learning and geometric analysis of detected installations. The completed project delivered a final report and a dataset of detected solar installations covering large parts of Switzerland. The findings support spatial inventories and the further development of solar-energy statistics.
Context and significance
A national inventory of solar installations needs spatially extensive information that can be updated as rooftops and energy systems change. Panels are directly visible in aerial imagery, yet their appearance, scale and mounting context vary. SolAI therefore connects image recognition with georeferencing and geometric post-processing.
The useful output is not just an image label but a geographically usable inventory. For operational updates, checking false detections is as important as finding installations missing from existing records.
01
The research question
How can photovoltaic and solar-thermal installations be automatically located and geometrically analysed?
02
Finding an answer
- Prepare georeferenced imagery and create reference annotations.
- Develop deep-learning methods for automatic detection.
- Analyse the geometry of detected installations.
03
Findings and applications
- A final report for the Swiss Federal Office of Energy.
- A dataset of detected solar installations covering large parts of Switzerland.
