Photovoltaic Potential Estimation of Single Building Rooftop Based on High-Definition Map Image and Deep Learning

Aiming at problems such as inaccurate rooftop extraction and missing contour in the estimation of urban-scale solar PV utilization potential, this study proposed a method for estimating single-building PV potential. First of all, the rooftop area is identified according to the high-definition map image and semantic segmentation SegNeXt of the study area, and then the building contour is identified according to canny edge detection, the single building lines are then extracted based on geographic location, the number of pixels in the contour line is calculated, and finally, the photovoltaic power generation potential is calculated through the photovoltaic panel parameters. The method was validated for accuracy and precision through actual measurements and comparative experiments. The identification accuracy is more than 90%, and the problems of inaccurate extraction and missing contour of rooftops are solved well.