Facebook uses satellite imagery machine learning and AI to prepare maps for locating unconnected communities across the world.
How Facebook uses satellite imagery machine learning and AI
Maps tell us so much more than how to get from A to B, or where C is in relation to D. They can be tools of power and snapshots of history; they can give urban planners the information to plan infrastructure. After a disaster, population density and crisis maps help to direct aid and aid workers. Throughout time, different cultures and industries have produced radically different images of the world. Today there are more than 7 billion humans sprawling across Earth. Out of this, 10% of the world’s population lives in areas where Internet connectivity is simply not available. They live in dense urban centers, in small towns linked by farms, and alone on the outskirts of jungles. But no one is sure where, exactly, many of them living. In an effort to get more and more people onto the Internet platform, Facebook recently released a series of maps to better understand the challenges facing unconnected communities. Since information from some of these areas was either unavailable or too coarse, Facebook resorted to imagery from DigitalGlobe and sorted them out using machine learning and artificial intelligence. The project is part of Facebook’s Connectivity Lab, the technical arm of its Internet.org initiative. The data, which can be found on the website of CIESIN at Columbia University, is a joint effort between Facebook, the university’s Center for International Earth Science Information Network, and the World Bank.
DigitalGlobe and its Big Data
The existing maps of populations in many parts of the world are too coarse, outdated, and inaccurate. To solve this problem, information from high-resolution satellites proves invaluable since it provides a consistent global information dataset for mapping population locations.
The Facebook Connectivity Lab is leveraging DigitalGlobe’s Geospatial Big Data initiative to determine population densities. It will be completing an accurate mosaic of the globe at 50 cm resolution and will be replenishing this basemap of the world on a frequent basis. Further, DigitalGlobe’s Geospatial Big Data platform, GBDX, makes this rich content along with its 15-year digital library of over 90 petabytes of high resolution satellite image data available to anyone, for processing in the Cloud. Available for use alongside the data is some of the best computer vision algorithms designed to convert pixels into meaningful data.
The High Resolution Settlement Layer provides estimates of human population distribution at a resolution of 1 arc-second (approximately 30 meter) for 2015. The estimates are based on the latest census data of each country and satellite imagery from DigitalGlobe. They are created based on the World Bank statistical model that incorporates population demographics, infrastructure development and existing internet penetration to identify the connectivity gaps and how to fix them.
AI-aided dataset analysis
With this effort, Facebook has joined a growing cast of Silicon Valley companies scrambling to perfect ‘machine learning’ — a newly popular form of artificial intelligence — that could unlock value in petabytes of satellite imagery. “There’s a lot of location data out there, but there hasn’t been a good way to use it to answer questions,” says Kevin Lausten, Director — Geospatial Big Data, DigitalGlobe. “If you can start to correlate all this information, you can uncover business opportunities.”
Facebook serves one-fifth of the world’s population and more than half of the roughly 3.2 billion people estimated to be Internet users at the end of last year. So it is but natural that it will need intelligence beyond human capabilities to collate and analyze this data. Today, it has one of the most amazing AI platforms that run as its backbone. Just as Facebook set out to rebuild the hardware industry half a decade ago with the Open Compute project, it has more recently created an internal platform to harness artificial intelligence so it can deliver exactly the content you want to see. And it wants to build this ‘machine learning’ platform to scale.
“We are trying to build more than 1.5 billion AI agents — one for every person who uses Facebook or any of its products,” Joaquin Candela, the head of the newly created Applied Machine Learning group, told Fortune last year.
To generate the maps, Connectivity Lab worked with Facebook’s data science division, infrastructure unit, and machine learning and artificial intelligence groups. The coalition analyzed satellite imagery for 20 countries (Algeria, Burkina Faso, Cameroon, Egypt, Ethiopia, Ghana, India, Ivory Coast, Kenya, Madagascar, Mexico, Mozambique, Nigeria, South Africa, Sri Lanka, Tanzania, Turkey, Uganda, Ukraine, and Uzbekistan), covering 21.6 million sq km, which amounted to 350TB of data. Using a mixture of computer vision techniques, including the image-recognition engine Facebook uses to identify people’s faces in photos, the team was able to identity human-built structures. The company sifted through more than 14 billion geospatial images captured by DigitalGlobe.
Machine learning and AI succeeded in identifying outlines of buildings and highlighted those for which it had high confidence while suppressing areas not likely to contain man-made structures. Once the structures were laid out, Facebook was able to use them as proxies for where and how people live. Using census data, the team redistributed the population datasets evenly across each location, under the assumption that the method was the least error-prone way of determining how many people lived in each building.
Using this building information, Columbia University used census data to generate population estimates, validate the result against other, more coarse-grained datasets and then the World Bank Living Standards Measurement Study (LSMS) program validated the final dataset against anonymized ‘ground-truth’ household surveys.
“The leap from ‘estimate’ to ‘accurate’ carries big bottom line implications for industries worth hundreds of billions of dollars. And geospatial data analytics companies are only beginning to wrap their heads around the possibilities,” says Luke Barrington, Senior Director — Platform Products, DigitalGlobe.
How many likes!
The population datasets are released publicly since Facebook believes this data could be useful to NGOs, humanitarian agencies or even companies in many ways, including for things like aid planning, disaster response, and infrastructure planning. The resulting maps revealed locations of over 2 billion disconnected people spread across 20 countries, many of them developing nations where even basic mapping data is scarce.
To begin with, population dataset maps for Malawi, South Africa, Ghana, Haiti, and Sri Lanka have been released. The social media giant hopes to keep adding more countries over the coming months and plans to increase connectivity around the world based on this data. These new maps can inform service providers where connectivity infrastructure should be deployed, whether they are fibre networks or Wi-Fi hotspots, or communication networks with high-altitude balloons or UAVs. The most efficient network technology depends on the ensuring communications networks designed on proximity to population.
“Developing communication technologies to reach the rest of the world is not a challenge we will solve alone. We want this data to be a new source of information to empower other organizations to achieve their missions. For example, The Red Cross is already planning to use the population maps to help combat malaria in Malawi, and to target humanitarian aid after natural disasters. By sharing our data and the problems we are working on publicly we hope to stimulate an open approach to solving local and global challenges,” Facebook said in a statement.
The Facebook map of the world is just a beginning. The industry is at a critical junction where the convergence of computing power, satellite imagery, machine learning and AI are coming together to answer more questions than before. What’s not to like about that?