Deep mapping gentrification in a large Canadian city using deep learning and Google Street View

Area of Science:

• Urban Studies
• Computer Science
• Geographic Information Systems

Background:

• Gentrification is a complex urban process characterized by visible neighborhood changes.
• Advances in computer vision and deep learning offer new tools for environmental attribute mapping.
• Automated analysis of visual data can support the study of urban dynamics.

Purpose of the Study:

• To develop and apply a deep learning model for detecting visual property improvements associated with gentrification.
• To create detailed spatial maps of gentrification-related changes over time.
• To validate the model's findings against traditional urban development data.

Main Methods:

• Development of a Siamese Convolutional Neural Network (SCNN) for detecting visual changes in image sequences.
• Application of the SCNN to Google Street View (GSV) images of 86,110 properties over nine years (2007-2016) in Ottawa, Canada.
• Utilizing Kernel Density Estimation (KDE) to map the spatial concentration of detected property improvements.

Main Results:

• The SCNN achieved 95.6% test accuracy in identifying gentrification-like visual changes.
• Mapped improvements showed strong concordance with building permit data in Ottawa (2011-2016).
• The study confirmed known gentrifying areas and revealed previously unrecognized areas undergoing gentrification.

Conclusions:

• Automated visual analysis using deep learning provides a novel and effective method for mapping gentrification at the property level.
• The SCNN-KDE approach offers a granular understanding of urban change dynamics.
• This methodology enhances the identification and study of gentrification processes in urban environments.