Facebook has combined satellite images and deep learning to create a new population density map for the majority of Africa, which the company is releasing today.
Why it matters: Governments and nonprofit organizations need accurate population density maps to coordinate humanitarian aid, health care, and other resources to traditionally unmapped regions. Facebook says that its maps of other regions have already helped support rural electrification efforts in Tanzania and coordinate a measles vaccination campaign in Malawi, among other examples.
Global coverage: The work builds on the company’s previous release of similar maps for 22 countries, originally motivated by its initiative to bring more people online. The goal is to eventually map the entire world.
How they did it: First, a team at Facebook’s World.AI group had to train a neural network to recognize whether a patch of land within a satellite image contained a home. To do this, the researchers created a training data set by overlaying more than 100 million crowdsourced coordinates of homes from OpenStreetMap onto satellite images. They also used old-school computer vision tricks to verify that the images labeled without homes didn’t contain any telltale polygon-shaped objects. The resultant deep-learning model reached over 99% accuracy during testing, according to Facebook. Finally, they divvied up satellite images of the African continent into 100-foot-by-100-foot areas and used the neural network to create an accurate, high-resolution population density map.
Deep geography: Facebook isn’t the only company using deep learning to extract information from satellite imagery. Tech firms like Planet and Descartes Labs use similar techniques to classify crops, track renewable-energy adoption, and monitor the health of ecosystems. Last year, Microsoft also trained a deep-learning model to build a comprehensive data set of all the building footprints in the US.