Remote Sensing based Renewable Energy System Recognition using Deep Learning

The transition from fossil fuels to renewable energies is a key requirement for a sustainable energy future. Achieving this transformation requires not only technological innovation but also reliable, up-to-date, and spatially explicit data on existing infrastructures. For decentralized energy systems, precise information on the spatial distribution and technical characteristics of renewable energy systems is crucial. At the same time, the growing availability of satellite and aerial imagery provides unprecedented spatial and temporal resolutions, but extracting actionable insights at scale requires advanced deep-learning methods, particularly image-based recognition. This cumulative dissertation is guided by three central research questions: (1) How can training data for deep-learning models be generated efficiently and reproducibly? (2) How robust are segmentation approaches across different spatial resolutions and geographic contexts? (3) How can the positional accuracy of existing datasets of renewable energy systems be improved? The results show that registry and building data can be combined to automatically generate reproducible training datasets, reducing reliance on manual labeling. Multi-resolution segmentation improves robustness and transferability, enabling reliable recognition of photovoltaic systems across diverse regions and scales. Object-detection models systematically correct positional errors in wind-turbine datasets, enhancing their accuracy and reliability. The dissertation makes several key contributions. Methodologically, it advances automated training-data generation, multi-resolution segmentation, and object detection-driven coordinate correction. Applied contributions include open, validated datasets on photovoltaic systems and wind turbines, strengthening the empirical foundation for modelling, planning, and scenario development. Together, these contributions demonstrate how remote sensing and deep learning can be combined to improve data quality and support evidence-based energy-transition strategies.