Towards Detecting Global Urban Poverty from Space

Abstract

Extreme urban poverty in the Global South persists, especially in slum areas, with slow progress despite initiatives like the SDGs. This research addresses the data gap by using remote sensing and AI to map slums on a large scale, supporting policymakers. Semantic segmentation and uncertainty estimation methods are applied to detect varying slum morphologies. The study extends to mapping slums in 55 cities using advanced models with uncertainty estimates, addressing the challenges posed by diverse slum structures across different regions.