Authors: Melanie Green*, University of Liverpool, Dani Arribas-Bel, University of Liverpool
Topics: Urban Geography, Remote Sensing, Spatial Analysis & Modeling
Keywords: socioeconomic characteristics, built environment, remote sensing, convolutional neural networks, machine learning
Session Type: Paper
Start / End Time: 12:40 PM / 2:20 PM
Room: Harding, Marriott, Mezzanine Level
Presentation File: No File Uploaded
Information about the characteristics and activities of humans becomes encoded in the landscape as it is continually shaped by the people who live there. These characteristics are traditionally measured using census data and represented using geodemographic classifications. Although not directly containing information on human characteristics, remotely sensed images at a high resolution contain a huge amount of detail on the features of the urban environment, including individual objects and their arrangement in space. This paper investigates the relationship between socioeconomic characteristics and the urban landscape using high-resolution aerial imagery and deep learning. A convolutional neural network is used to extract features from the images using transfer learning, and k-means clustering is applied to group together areas with similar features. When compared with socioeconomic groups from the Output Area Classification, different socioeconomic profiles are evident in each cluster, and these are spatially clustered in different areas of the city. This information can complement the information on the built environment that is traditionally used in geodemographic classifications and enhance understanding of the built and human characteristics of neighbourhoods.