Detecting solar rooftop photovoltaic panels in aerial images using neural networks: a transfer learning approach

Castello, RobertoRoquette, Simon2020-06-232020-06-232020-06-19https://infoscience.epfl.ch/handle/20.500.14299/169532This report summarizes the research results obtained by trying to improve an existing image segmentation deep learning model on photovoltaic panels (PV) [1], exploring mainly the impact of transfer learning. Many metrics and tools to assess performance in a more meaningful way have also been explored, and used to improve the model’s performance. Improvements of the dataset were also done, as well as an evaluation of the minimum amount of data needed to reach model’s best performance.Detecting solar rooftop photovoltaic panels in aerial images using neural networks: a transfer learning approachstudent work::semester or other student projects