Support Us Subscribe to Our EMail

Subscribe
To Our Weekly Newsletter

Send Us a Tip
Astraea brings machine learning to images of Earth
Astraea Team, May 2017
Enlarge Image
Credit: Ryan M. Kelly, The Daily Progress
The founders of Astraea pose for a portrait at their offices in The Glass Building in downtown Charlottesville. Seated, left to right: Simeon Fitch, Matt Eldridge and Kimberly Scott. Standing, left to right: Daniel Bailey, Brendan Richardson.
1
Josh Mandell | Saturday, May 27, 2017 at 4:53 p.m.

Astraea, a software startup in Charlottesville, is ready to take on the world — specifically, the detailed images of it captured by satellites.

“We are building a tool so people without Ph.D.s in data science can interrogate one of the biggest datasets ever,” said CEO Brendan Richardson. “They will use it in ways we would never think of on our own.”

Founded last fall, Astraea is developing a machine learning engine optimized for the analysis of image data from Earth-observing satellites.

Machine learning enables computers to recognize patterns and search for relevant information without being programmed to perform specific tasks.

“There’s no way to meaningfully ingest all of this data without a machine learning platform that can handle it,” Richardson said. “People are doing interesting things with the data, but only in narrow ways.”

Many startups are using machine learning to extract intelligence from big data. However, Richardson said that studying image data from satellites remains daunting due to its ambiguity, distortion and sheer volume.

Some entrepreneurs and researchers have started to combine machine learning and satellite imagery in innovative projects.

Descartes Labs, a New Mexico startup, uses a machine learning algorithm to study satellite images of American farms, allowing it to predict corn yields with unprecedented accuracy.

At Stanford University, a team of computer scientists has made accurate and highly localized estimates of household consumption and assets in African nations by studying patterns in satellite images, including nighttime illumination, roofing materials and the density of roadways.

Daniel Bailey, Astraea’s chief technology officer, said insight from satellite imagery can support the discovery and sustainable use of natural resources.

“We believe that, through technology, there is a way to balance technological advances with care for our planet,” he said.

Richardson said Earth-observing satellites managed by the U.S. Geological Survey, the National Oceanic and Atmospheric Administration and NASA have collected about 25 petabytes of images — or 25 million gigabytes — since the 1970s.

These images can be obtained for free and have no restrictions on their use. More recently, private companies such as DigitalGlobe and Planet have launched hundreds of Earth-observing “nanosatellites” that take photos more frequently and at higher resolution than government satellites. These companies sell their images to governments and to other businesses.

Richardson said these constellations of privately owned satellites will create even greater demand for Astraea’s software.

“The size of data downloads is exploding, and it can take petabytes of data to answer a single question,” he said.

Bailey said Astraea will use funding from angel investors to hire more software engineers, data scientists and machine learning experts this year.

Bailey and other Astraea co-founders came to the startup from Charlottesville analytics firms that make use of machine learning, including Elder Research and Commonwealth Computer Research.

Kimberly Scott, Astraea’s vice president of data science, also has done astrophysics research for the National Radio Astronomy Observatory in Charlottesville.

“Charlottesville has a deep pool of data science talent,” Richardson said.

Richardson is a co-founder and former CEO of PsiKick, a sensor technology startup with offices in Charlottesville, California and Michigan. Before that, he worked for venture capital firms in the U.S. and Europe.

“I think it’s more fun to be involved in building a business than just investing in one,” he said.

Astraea is seeking pilot customers that can put its machine learning algorithms in the hands of other data scientists.

Simeon Fitch, Astraea’s vice president of research and development, said data scientists frequently devote more than 80 percent of their time to repackaging raw data into usable formats.

“If we can shrink that to 30 or 40 percent, that’s a huge gain in productivity,” Fitch said.

Richardson said that Astraea hopes to eventually serve a variety of businesses that “don’t even know how to begin answering questions with this data” — including hedge funds, insurance firms and agricultural companies. State and local governments are also potential clients, he said.

“Astraea will allow anyone studying our planet to see beyond what a human can see alone,” Bailey said. “They can go beyond those limitations.”

Charlottesville Tomorrow’s comment policy
comments powered by Disqus