For the past For 15 years, NASA’s Mars Reconnaissance Orbiter has been touring the Red Planet studying its climate and geology. Every day, the orbiter sends back a treasure trove of images and other sensor data that NASA scientists have used to search for safe landing places for rovers and to understand the distribution of water ice on the planet. Of particular interest to scientists are photographs of the orbiter’s crater, which can provide a window into the deep history of the planet. NASA engineers are still working on a mission to return evidence from Mars; without the stones that will help them calibrate the satellite data remotely with surface conditions, they have to make a lot of educated assumptions when it comes to determining the age and composition of each crater.
For now, they need other ways to delete this information. A tried and true method is to extrapolate the age of the oldest craters from the characteristics of the newest craters on the planet. Because scientists can know the age of recent impact sites in a few years – or even weeks – they can use them as a basis for determining the age and composition of much older craters. The problem is finding them. Combining the value of the images of a planet in search of telltale signs of fresh impact is tedious work, but it is exactly the kind of problem an AI has made to solve.
Late last year, NASA researchers used a machine learning algorithm to discover fresh Martian craters for the first time. AI discovered dozens of them hiding in the image data on the Mars Reconnaissance Orbiter and revealed a promising new way to study the planets in our solar system. “From a scientific perspective, it’s interesting because it increases our knowledge about these features,” says Kiri Wagstaff, a computer scientist at NASA’s Jet Propulsion Laboratory and one of the leaders of the research team. “The data was there all the time, only we hadn’t seen it ourselves.”
Mars Reconnaissance Orbiter carries three cameras, but Wagstaff and her colleagues trained their AI using images from Context and HiRISE images. The backdrop is a relatively low-resolution grayscale camera, while HiRISE uses the largest reflecting telescope ever sent into deep space to produce images with resolutions about three times higher than the images used on Google Maps.
First, AI was fed nearly 7,000 orbital photos of Mars – some with previously discovered craters and some without any – to learn the algorithm how to detect a new impact. After the classifier was able to accurately detect the craters in the training set, Wagstaff and her team loaded the algorithm onto a supercomputer at Jet Propulsion Laboratory and used it to comb through a database of more than 112,000 images. on the orbiter.
“There’s nothing new with the technology behind machine learning,” says Wagstaff. “We used a fairly standard convolutional network to analyze image data, but being able to scale it is still a challenge. That was one of the things we had to deal with here. ”
The latest craters on Mars are small and could be only a few meters, which means that they appear as dark spots pixelated on contextual images. If the algorithm compares the image of the candidate crater with a previous photo from the same area and finds that the dark patch is missing, there is a good chance that it will find a new crater. The date of the previous image also helps to set the timeline for when the impact occurred.
Once AI identified some promising candidates, NASA researchers were able to make a few subsequent observations with the orbiter’s high-resolution camera to confirm the existence of craters. In August last year, the team received the first confirmation when the blinder photographed a group of craters that had been identified by the algorithm. It was the first time an AI had discovered a crater on another planet. “There is no guarantee that there will be new things,” says Wagstaff. “But there have been many of them and one of our big questions is: what makes them harder to find?”