Mapping solar photovoltaic arrays using unmanned aerial vehicles and deep learning

Abstract: Solar home systems (SHS), a cost-effective solution for rural communities far from the grid in developing countries, aresmall solar panels and associated equipment that provides power to a single household. A crucial resource for targetingfurther investment of public and private resources as well as tracking the progress of universal electrification goals isshared access to high-quality information such as location, size, and capacity of individual SHS installations. Thoughrecent studies utilizing satellite imagery and machine learning to detect solar panels have flourished, they struggle toaccurately locate many SHSs due to limited resolution (some small solar panels only occupy several pixels in satelliteimagery). In this work, we explore the viability and cost-performance tradeoff of using drone imagery as an alternativefor automatic SHS detection in satellite imagery. To examine this question we collected, and publicly-release, a largedataset of high-resolution drone imagery encompassing SHSs imaged under a variety of real-world conditions. We alsomanually annotated the locations and sizes of all SHSs in the imagery to enable the training and performance validationof deep learning recognition models. We utilize this dataset – the first of its kind – to evaluate the capabilities of deeplearning models to recognize SHSs, including those that are too small to be reliably recognized in satellite imagery. Wealso conduct an analysis of the cost-effectiveness of drone versus satellite-based mapping of solar arrays.

Recommended citation: Ren, S., Hu, W., …, Fetter, R., Bradbury, K., Malof, J. (2021). Mapping solar photovoltaic arrays using unmanned aerial vehicles and deep learning