A research team from the Department of Computer Science at the Faculty of Electrical Engineering and Computer Science, VSB – Technical University of Ostrava, is developing methods to improve the accuracy of solar potential estimates in the Moravian-Silesian Region. The project builds on a previous study conducted in cooperation with the ENET Centre, which assessed the technical potential of photovoltaic and wind power plants using a statistical approach applied to a limited sample of buildings. However, this approach did not allow for sufficiently accurate spatial evaluation.The current research uses machine learning methods—specifically neural networks and computer vision techniques to automate spatial analysis based on orthophotos. “The tool we are developing is designed to identify building rooftops from aerial images, classify their type (e.g. flat or sloped), and determine their potential suitability for photovoltaic panel installation,” explains David Seidl from the Department of Computer Science.
Correctly identifying the type of roof is crucial for calculating the installed capacity, as it affects not only the orientation and tilt of the panels but also the technical feasibility of the installation itself. A key element of the entire system is the creation of a sufficiently large and well-annotated training dataset, which allows AI models to learn how to recognize relevant patterns. “Our students use an internal web-based tool to manually label rooftops on randomly selected satellite images. They define both the shape and exact location. These manually created annotations serve as input data for training the model and gradually improving its accuracy,” Seidl adds. The goal of the research is to develop a tool capable of independently analyzing areas at the regional level and eventually across the entire Czech Republic. The results can significantly contribute to more accurate estimations of photovoltaic potential and serve as a practical foundation for energy planning at both regional and national levels.
