Remote Sensing Data Preparation for Recognition and Classification of Building Roofs

Buildings are among the most significant urban infrastructure that directly affects citizens’ livelihood. Knowledge about their rooftops is essential not only for implementing different Levels of Detail (LoD) in 3D city models but also performing analyses related to usage potential (solar, green, social), construction assessment, maintenance, etc. At same time, more detailed information we have environment, adequate digital twins can create. This paper proposes an approach dataset preparation using orthophoto with a resolution 10 cm. The goal obtain roof images into separate GeoTIFFs categorised by type (flat, pitched, complex) way suitable feeding rooftop classification models. Although initially elaborated classification, it be applied developing other deep-learning recognition, segmentation, and estimation. consists 3617 roofs covering Lozenets district Sofia, Bulgaria. During its preparation, local-specific context considered.