Crater counting has been used since the 1960s to determine ages for planetary surfaces. The more craters, the older the surface. But its always been a laborious manual effort.
We developed an advanced machine learning algorithm to automate this process, and validated it against the manually counted datasets. With this algorithm, we analysed a 3.9TB mosaicked (5m/pixel) image of Mars. The previous (definitive) manually counted dataset was 380,000 craters, which stopped at 1km diameter craters. Our algorithm returned 94 million craters down to 50m in diameter in 24 hours. This is now the world’s largest database of Martian craters. It allows us to see age variation across Mars at unprecedented resolution.
This is the first time a machine learning surface age tool has been applied to derive new knowledge of our solar system. We have used it to address multiple science questions, including pinpointing locations on Mars that are the sources of meteorites we have here on Earth, effectively the first Mars Sample Return.
We are now able to count down to the smallest sizes that we can see – as small as 10m across – and we can do it in a robust, quantitative, reproducible way. We can measure all of the craters across the entire surface of a planet or a moon, and derive a complete age map for that body at ultimate resolution. Our latest focus is the Moon, where we now have a quantitative dataset for >240 million craters. This is a principal area of focus SSTC collaboration with NASA in Artemis.