Better Living Through Big Data

Using AI to Discover the Moon’s Hidden Treasures

Moon craters

With the help of artificial intelligence, NASA’s Frontier Development Lab and Intel are mapping the moon’s craters to find hidden lunar resources.

Scientists believe the moon is rife with natural resources that could help space explorers settle the lunar landscape – much like early settlers did on earth.

But before they can access those resources, they need to find them.

“We have 50 years’ worth of NASA imagery from all sides of the moon,” said Shashi Jain, innovation manager at Intel’s Software and Services Group. “We’ve only recently begun to combine them and make one big, awesome map.”

Working with the NASA Frontier Development Lab (FDL), a team of Intel AI engineers and data scientists are tackling the challenge of building complex maps of the lunar poles.

Craters in the permanently shadowed polar regions of the moon are potentially filled with water, ice and other volatile resources that can be used to produce rocket fuel, an air supply for astronauts or other essential materials, according to Jain.

Shashi Jain
Intel’s Shashi Jain is working with NASA to create “one big awesome map” of the moon. Photo credit: Walden Kirsch.

The shadowy lunar surface creates artifacts in NASA images, which make it difficult to accurately map potential landing sites for lunar prospectors.

But that’s changing with the help of AI and new sensors that have been placed on the moon in the last 10 years, providing depth data and hyperspectral imagery.

Crews on long exploratory missions to outer space can’t carry all the resources they need, so finding things like water, hydrogen, carbon dioxide, nitrogen and methane may help NASA plan future missions to the moon or even to Mars.

Making Maps from Millions of Images

It turns out that making maps from lunar data is hard, said Jain. Planetary scientists get strips of imagery from orbiting satellites, which are at different lighting angles, scales and types. They have to manually line them up using craters and other features as landmarks. If any strips are out of alignment or too dark, the result is a poor quality map.

mapping the moon GIF
Rovers will soon return to the moon’s surface to investigate harvesting lunar water. Credit: NASA.

Deep learning, a branch of machine learning that uses neural network models to understand large amounts of data, could speed up the process of mapping the moon.

Whereas machine learning allows machines to act or think without being explicitly directed to perform specific functions, deep learning can accelerate processes like image recognition, quickly identifying and mapping craters and other obstacles on the moon.

“Space data is often massive, multidimensional and dynamic,” said James Parr, director of NASA FDL.

It’s critical for scientists to quickly process ever-evolving lunar data to help guide plans for future missions, according to Parr.

To get started, the team first needed to create a computer vision algorithm and train it to identify craters. NASA FDL and Intel built a crater image training set using 30,000 images. It took Jain six hours to manually find images containing a crater — but fully mapping the moon means looking at hundreds of millions of images.

Moon craters
Craters, formed by impactors plummeting into permanently shadowed regions at the south pole, may contain ice. Image credit: NASA/GSFC/Arizona State University.

In order to create detailed lunar maps, the team used two datasets from the NASA Lunar Reconnaissance Orbiter (LRO) mission — one set with optical images and the other with elevation measure data. Overlaying the two datasets created highly accurate maps, said Jain.

Since lunar craters acted as critical registration points to align the two datasets into one unified map, the team developed a computer vision algorithm to quickly and reliably identify craters.

The team automated lunar crater detection with 98.4 percent accuracy. By running their algorithm on the Intel Nervana Cloud, it took only one minute to classify 1,000 images, which is 100 times faster than human experts. The algorithm is also available in GitHub for use by other research teams.

Partners in Space Research

The NASA FDL space resource project was completed during a whirlwind eight-week program at the SETI Institute in Mountain View, California.

Moon craters in color
Current mapping quality is insufficient for rover mission planning. There is no GPS outside of Earth so rovers need to pre-plan safe traverses. Image credit: NASA/JPL.

The space resource team was just one of five teams that took part in challenges in the summer program. Other teams tackled planetary defense and space weather challenges, like long period comets, radar 3D shape modeling, solar-terrestrial interactions and solar storm prediction.

“It was the summer of exploration with artificial intelligence right here on Earth,” said Jain.

With AI-generated maps of the moon’s poles, soon NASA will have summers, winters and years of exploration on the moon and beyond.

 

Feature image credit: NASA/JPL/Cassini Imaging Team/University of Arizona.

iQ newsletter signup

Share This Article

Read Full Story