Detection of slums in Rio de Janeiro through satellite images

Abstract

According to UN-Habitat, more than one billion people live in informal settlements worldwide, of which 200 million living in Africa and another 100 million in Latin America, mainly in countries such as Brazil, Mexico, Colombia, Peru, and Argentina. Rio de Janeiro has 1,074 favelas, representing 22% of the city's total population, making it the Brazilian municipality with the highest percentage of people living in favelas. Ensuring human rights through access to basic services for the populations living in these settlements, through programs and public policies, depends on timely and reliable data. However, despite spending decades establishing their national statistical systems, usually based on data collection directly from individuals, in most countries, the data produced in traditional ways does not portray the dynamics of these populations promptly. As an alternative, we combined free satellite imagery with machine learning and deep learning to identify the area occupied by favelas in the city of Rio de Janeiro. We compared the results of eight distinct segmentation models using the IoU and F1 as metrics. Among the evaluated methods, two stood out for their performance: GradientBoost and XGBoost.