Analysis of pressure perception of internal roads in existing settlements based on random forest/deep learning

Abstract

The quality of road space has an influential role in human psychological health. Quantifying residents' perceptions of road spatial quality and exploring the relationship between perceptions and constituent elements have become topics of interest to many parties. In this study, based on the high-resolution remote sensing images acquired by Baidu Maps, the road images of built-up settlements in Beijing in three different periods (1949-1977, 1978-1997, and 1998-2022) were semantically segmented. The large-scale rating data and subjective evaluations of residents' spatial perceptions of roads were analyzed using random forest and multiple linear regression methods. The results show that stress, a negative perception, is positively correlated with buildings and cars, and negatively correlated with vegetation, sky, and sidewalks. The composition of elements affecting residents' stress perceptions in communities of different build-up years was consistent, but the degree of influence of each element was ranked differently. This study attempts to summarize the road characteristics and residential advantages of different built-up neighborhoods, and proposes optimization strategies for upgrading road space in old neighborhoods and countermeasure suggestions for future improvement of the human living environment.