A Deep Learning Based Approach for Rooftop Solar Potential Estimation of a City: A Case Study of Indian Metropolis

We propose the implementation of deep learning-based architecture for extracting the built-up and application towards solar photovoltaic potential estimation. The U-Net model proposed for the extraction of building rooftops using medium resolution satellite data. Building footprints from OpenStreetMap is used to generate labels necessary for training the model. The analysis of study region establishes that about 1.89 Giga Watt could be generated from building roofs of the study region. The roof area estimation is validated using polygonal building outlines from 25 km
²
of the city locality. Analysis of this kind could help in reaching the targets set for implementation of visualized scale of programs by the governing bodies.