A Deep Learning Approach for Urban Block: Automated Extraction Tool for Urban Forms

Abstract

Increasing access to geographic data and mapping technologies has pushed urban morphology research toward more quantitative and data-driven approaches. At the same time, the unprecedented rapid change in the urban form has prompted a growing number of research to capture, analyze, and understand the phenomenon in recent years. However, a thorough, systematic approach to evaluating and comparing urban forms in this setting is yet to be developed. The aim of this study is to build a comprehensive approach to defining urban form indicators by developing a simplified yet representative classification of the urban form. Notably, urban block as a constitutional feature of urban form is evaluated in relation to numerical indices. The applied methodology comprises the detection and classification of urban form using a deep convolutional neural network. The study attempts to use automated methods to address the gap in urban form classification and characterization. The methodological process encompasses a non-local classification of urban form, followed by an examination of the identified features of the urban block. The preliminary outcome of this study consists of an in-depth analysis of urban block indicators in the comparative literature. This will be one of the inputs of the deep learning model to classify urban blocks.