HyperionSolarNet: Solar panel detection from aerial images

Summary: Researchers from UC Berkeley set out to create a comprehensive database for locating solar panels given the importance of being able to assist analysts and policymakers on defining strategies for further expansion of solar energy.

Challenge: With the effects of global climate change impacting the world, collective efforts are needed to reduce greenhouse gas emissions. The energy sector continues to be the single largest contributor to climate change and many efforts are focused on reducing dependence on carbon-emitting power plants and moving to renewable energy sources, such as solar power.

Their work focused on creating a world map of solar panels, identifying locations and total surface area of solar panels within a given geographic area. The researchers used deep learning methods for automated detection of solar panel locations and their surface area using aerial imagery. The framework, which consisted of a two-branch model using an image classifier in tandem with a semantic segmentation model, was trained on a created dataset of satellite images.

Findings: Their work provided an efficient and scalable method for detecting solar panels, achieving an accuracy of 0.96 for classification and an IoU score of 0.82 for segmentation performance.

How Labelbox was used: The research team annotated 836 images containing solar panels using the Labelbox platform, and produced corresponding segmentation masks, resizing all images to a size of 512x512 pixels for training and testing. The researchers manually annotated these images using Labelbox and created mask labels for them, afterwards, evaluating the HyperionSolarNet segmentation model against these test set images and finding  an IoU score of 0.82.

Read the full PDF here.