The detection of residential developments in urban areas: Exploring the potentials of deep-learning algorithms

Abstract

A rich volume of research has detected urban growth by quantifying the land use/land cover (LU/LC) changes based on remote sensing technologies. However, the research has limitations in identifying various formats of urban growth, particularly small-scale urban growth, such as infill development or redevelopment in urban areas, prompted by smart growth and sustainable urban development. This paper aims to design a framework for the accurate detection of residential infill development in the City of Los Angeles by employing a deep-learning method that has been increasingly applied to analyze phenomena in cities. In order to do so, this paper develops six models that reflect the variations of image classification methods, deep-learning algorithms, and estimation types. The results from the models emphasize the potential of the deep-learning models for the detection of micro-urban growth at a city scale. However, there is room for the improvement of estimation accuracy in the cases that detect some new developments and replacements as not developed parcels. The findings suggest that the performance of the models depends primarily on the articulations of the training dataset rather than the types of deep-learning algorithms. Findings from the models provide the city with insights into land use and transportation planning decision-making based on a better understanding of the spatial distribution patterns of urban growth and development.