BUILDING ROOFTOPS EXTRACTION FROM SATELLITE IMAGES USING MACHINE LEARNING ALGORITHMS FOR SOLAR PHOTOVOLTAIC POTENTIAL ESTIMATION

Abstract

This research involves the extraction of buildings from satellite images to be used for solar PV potential estimation. Pre-processing algorithms were used to enhance the image of a city in Cairo, Egypt called Madinaty. Then, two machine learning techniques were used to extract the buildings’ rooftops. SVM exceeded Naïve Bayes in terms of the detection accuracy with an F1 score of 94.7%. The detected gross areas of the rooftops were used in the second phase of this thesis, which is the PV potential estimation of PV panels mounted over the detected rooftops. Three methods were used for the solar modeling of the study area which are PVWatts calculator, PVGIS, and ArcGIS. The estimated PV potentials were calculated to be 21, 24.9, and 22.3 GWh/year for the three methods respectively. CO2 was reduced by an average of 62% after using solar panels instead of depending on traditional energy sources.