Urban data-driven perceptual assessment framework of neighborhood growth and decline based on deep learning image processing

Abstract

Changes in urban neighborhood can lead to growth and development, or they can lead to decline and present a number of challenges. The growth of an urban area is a dynamic process that affects the region at different spatial and temporal scales and is related to factors that drive change, such as the region's environment, politics, geography, and other considerations. The decline of urban areas leads to deterioration of facilities and unbalanced economic and social development. In general, the process of decline of urban areas over time is an important urban issue. Changes such as the growth and decline of these neighborhoods lead people to recognize urban change. Previous studies on perceiving and recognizing urban change has mainly considered using image data to measure people's perceived mood and atmosphere. People's perceptions of urban areas can be a factor in the potential for urban area change. These previous studies shed little light on the extent to which the visual components of urban landscape images influence the perception of urban area growth and decline. Therefore, it is necessary to investigate the perception of the horizontal components of urban scenes that determine the perception of urban growth and decline. This paper sets the research question "How can we measure the factors that affecting the perception of growth and decline in urban neighborhood area?" and aims to investigate the factors that lead to the perception of growth and decline in urban areas using multi-dimensional urban spatial image processing and deep learning-based analysis. The detailed research questions for the research question are as follows. 1) How can we measure perceptual assessments of urban growth and decline using 2D urban imagery? 2) How can we measure perceptual assessments of urban growth and decline using 3D urban point cloud data? 3) How can we create a deep learning model to assess perceptions of urban growth and decline? To answer each of these questions, we collected two-dimensional images of urban areas and three-dimensional point cloud data to analyze the components of urban areas. We then converted the images into virtual reality and conducted a survey to gauge people's perceptions. The questionnaire consists of physical, environmental, social, and economic elements that make up a street. Next, we processed the two-dimensional image data and three-dimensional point cloud data to classify the components of the urban area, and analyzed the relationship between the survey results and each component through multivariate linear regression analysis to generate a score for the growth and decline of the urban area. Finally, based on the calculated scores, an evaluation model using deep learning was created to derive important factors for the growth and decline of urban areas. This dissertation analyzed the factors affecting the growth and decline of urban areas using multidimensional image data processing of urban space, and found that it can contribute to analyzing the development and decline of urban areas and identifying factors with a different measurement range than existing studies. In addition, this dissertation can provide knowledge such as setting factors for policy making and development of urban regeneration areas.