Satellite Surveillance Takes the Pulse of America’s Health
Can surveillance images taken by satellites 22,000 miles above provide deep insights into the people who reside in our planet’s urban areas?
Turns out these photographs can tell us more than we might expect.
A recent study1 confirmed our environment impacts our health in powerful ways. And the research team behind the study relied on a surprising technology to reach its conclusions: Artificial intelligence.
Rather than manually poring over 150,000 high-resolution satellite images to categorize environmental features, the team fed the data into a neural network and used AI to extract the insights.
Elaine Nsoesie,2 one of the study’s authors and a researcher at a university in Boston, spoke to us about how her team performed the study, and what AI could mean for the future of public health. (Responses have been lightly edited for clarity).
What was your hypothesis going into the study?
Quite a few studies have looked at different aspects of our environment and how that correlates with physical activity, obesity prevalence, and other health outcomes. Previous studies have looked at things like street structures, highways, greenery, spaces between houses, and so on. We were expecting we would see some associations between the built environment and obesity prevalence. We just weren’t sure how high those associations would be.
What role did artificial intelligence play in the research?
In this study, we could have gone out and physically counted everything in all the different communities that we were interested in studying. That would take a while… but if we could have AI do that work for us, it would ease the amount of manual work we had to do.
It took about 24 hours for the neural network to go through all the images. The amount of time it would take for someone to go through these images manually could be up to a month.
So is the AI just labeling all these features for you, doing the grunt work?
Yes and no. The AI looks at each image and assigns probabilities to what it sees. The AI represents everything in each image – whether it’s trees or a house or something else – as a number. It then combines all those numbers to use as a representation of that particular image, and then it combines images across a particular neighborhood to create a numerical picture of the environment.
We looked at cities that had high-obesity prevalence and cities that had low-obesity prevalence and found that neighborhood characteristics like parks and crosswalks have significant correlation with lower obesity levels. We found some surprising correlations, too, such as that areas near police stations are correlated with higher obesity levels. But overall, our correlations line up with obesity estimates published by the CDC.
Have you done any other studies along these lines?
Yes. Most of my interest is looking at new technology and how we can use that in public health research.
Another study that we did actually inspired this one in a way. A couple of years ago we looked at high-resolution satellite images of hospital parking lots in three Latin American countries. The idea was that we might be able to correlate the parking lot data with the prevalence of influenza-like illnesses in those countries, theorizing that more parked cars would mean higher levels of illness. Since they don’t often have access to statistical data, we found we could correlate that image data with the incidence of illness. But for that, we had to manually count all the cars.
How are you working with AI now?
We have a few ongoing studies. We have a similar study to our obesity research connecting environmental features with crime.
Going forward, how critical will AI be in this kind of research?
The main goal of this study was to see how we could use AI to analyze public health issues: Specifically, can we use those methods to improve how we collect data and how we process data?
These types of studies can definitely benefit public health, and these techniques can be one way of auditing different aspects of our environment.
Read More on the Data Powering Surveillance Systems:
- Deep Learning in Surveillance
- How Smart Video Surveillance is Changing Edge Architectures
- Durability and Surveillance in Edge Units
This content is produced by WIRED Brand Lab in collaboration with Western Digital Corporation.