Mapping Solar Potential: Comprehensive PV System Detection with Aerial and Satellite Images Using SolarSAM

abstract

Automatic recognition of photovoltaic (PV) systems through remote sensing is critical for energy and infrastructure planning. This study explores the efficacy of deep learning in detecting PV systems using remote sensing. We introduce the adaptation of the Segment Anything Model (SAM) to this task, marking the first application of SAM for the detection and delineation of PV installations. We train the model on high-resolution images and then test its performance on both high-resolution aerial images and lower-resolution satellite images. Our results demonstrate high detection accuracy and precision especially for aerial imagery. We underscore the potential of advanced deep learning techniques in detecting and monitoring PV installations, facilitating more effective planning and deployment of renewable energy resources. Our findings suggest that the application of SAM in this domain could lead to significant advancements in energy infrastructure development, offering a promising tool for optimizing the integration and management of solar energy systems on a global scale.