Detecting and Analyzing Solar Panels in Switzerland using Aerial Imagery

The renewable energy statistics estimate the production of solar power and solar heat. To validate these statistics, the existing installations are to be identified and quantified fully automatically using deep learning algorithms from aerial photographs. Thanks to these methods, the current and future potential for use in Switzerland can be determined more precisely. Above all, this will also allow the status of implementation of the Energy Strategy 2050 of Switzerland to be determined. To train the model various solar panels have been collected using crowd sourcing. A Flask-based Web-Service was created to collect data. To create the model a GPU Cluster using 4 NVidia V100 was used. It is shown how this can be accomplished.