Mapping human perception of urban landscape from street-view images: A deep-learning approach

Abstract

Human perception of urban landscape, which signifies to what extent urban landscape is appreciated by local dwellers, informs human-oriented policies that reinforce public participation. Yet, conventional studies on human perception of urban landscape are largely dependent on individual experience, which may restrict the co-production of knowledge that can be operationalized across spatial scales and sectors. In this study, we mapped human perception of urban landscape in Shanghai by leveraging an advanced deep-learning approach and street-view images. Specifically, the ResNet50 model was employed to map four critical perceptions, i.e., security, depression, vitality, and aesthetic, at parcel level. We further explored the relationship between human perception and land-use types. Our results show that highly urbanized area (Puxi district encompassed by the Inner Ring Road) is perceived as more secure and vital, but more depressing. Besides, human perception varies substantially across different land-use types, among which administrative and service land is favored with regard to all the four perception types. This study advances our understanding of urban landscape through the lens of human perception, and provides nuanced insights into steering human settlement towards sustainability by strategically promoting mixed land use.

Keywords

• Human perception
• Deep learning
• Street-view image
• Urban landscape
• Land use