With the effects of global climate change impacting the world, collective efforts are required to minimize greenhouse gas emissions. To reduce dependence on carbon-emitting plants, a method for accurate forecasting of solar energy is needed. While it has been a challenge developing a comprehensive database of solar panel data, one solution is to leverage machine learning capabilities to detect solar panel locations.
HyperionSolarNet utilizes deep learning methods and aerial imagery for automated detection of solar panel locations and total surface area. The framework, which consists of a two-branch model using an image classifier in tandem with a semantic segmentation model, provides an efficient and scalable method for detecting solar panels with reliable performance. Our work includes an application that provides users an interactive mapping tool that visualizes the predicted solar panel mask overlay from the classification and segmentation models.
For more information, please visit https://groups.ischool.berkeley.edu/HyperionSolarNet/