Objective Typification of Building Exteriors Using Deep Learning-Focused on Public Office Buildings

Abstract

This study introduces an objective typification methodology that employs deep learning to analyze the exterior appearances of buildings. The conventional approach to typification was reliant on subjective analysis and was limited in terms of the number of structures that could be assessed. This study aimed to overcome these limitations by establishing an objective typification method using deep learning, focusing specifically on public office buildings. The research process involved a comprehensive survey of domestic public office buildings to compile an image dataset. Subsequently, a model was constructed utilizing Convolutional Neural Networks (CNN), a form of deep learning, to grasp the distinctive features of building images. These features, extracted from the CNN model, were then organized into groups through k-means clustering. The outcome of this clustering enabled the analysis of each clusters unique characteristics, facilitating the establishment of typification criteria such as building height, fa?ade pattern, materials, protrusions, and roof structures. This methodologys effectiveness was validated through a comparative analysis with prior research. The results of this study offer potential applications in fundamental investigations concerning the current state of public office buildings and in typification studies encompassing diverse architectural forms beyond public office buildings.