Dr. Nils C. Prieur

Dr. Nils C. Prieur

ML Engineer / Earth & Planetary Scientist

Preparation of Lunar Data and First Training of a Boulder Detection Algorithm

2 minute read

Published:

Hey folks! While I’m still waiting for the fieldwork season to kick off next year, I’ve been keeping busy with some lunar data. My current focus is mapping boulders around young impact craters on the Moon—by “young,” I mean geologically speaking, anywhere from a few million to a couple of hundred million years old. Since the Moon lacks an atmosphere and active geology, erosion is quite slow, preserving impact craters for much longer periods than on Earth.

I’ve chosen the Censorinus impact crater as my first mapping site. Censorinus is a relatively small, 3.8-kilometer-wide crater on the lunar highlands. Its bright ejecta blanket and well-preserved rays make it an ideal candidate for studying ejecta boulder distributions. The crater was even featured in a 2016 study by Krishna and colleagues, who documented over 240,000 boulders around it. Fun fact: Censorinus formed from an oblique impact, meaning the asteroid hit at a low angle, which creates an asymmetric ejecta pattern—a perfect case for my research​.

I’m hoping to collaborate with the authors of that study, as it would be great to compare my mapping to theirs. In machine learning, it’s crucial to understand the variability in how different humans label data. This will help me refine my guidelines for annotating boulder outlines, ensuring consistent results across the project.

Speaking of annotation, there’s a lot to consider. Should I include shadows in the boulder outlines? How do I handle boulders near the resolution limit of the Lunar Reconnaissance Orbiter’s Narrow-Angle Camera (NAC) images? These high-res images are enormous, often over 70,000x10,000 pixels, and need to be broken down into smaller tiles for processing. I’ll need to establish best practices not just for myself but also for any students who might assist in the future.

Lastly, I’m beginning to train my boulder detection algorithm using Detectron2. It’s a cutting-edge platform for object detection, but adapting it to lunar data is a challenge. With so many rocks on the Moon, the algorithm needs a large and well-labeled dataset to perform accurately. Plenty of work ahead!

That’s all for now. Time to wrap up and get ready for a cozy Christmas back home in Norway. Take care, and happy holidays! 🎄

You May Also Enjoy

Transition to YOLOv8, always improving the algorithm, collecting more labels…

3 minute read

Published:

Bonjour, everyone! Time feels like it’s slipping away faster than ever these days. With paternity leave in full swing, finding work hours here and there has been a challenge (and honestly, it’s hard not to get distracted by Baby Finn’s cuteness). Despite the chaos, I’ve managed to make a major leap forward in the project: transitioning from Detectron2 Mask R-CNN to YOLOv8. Let me tell you—it’s been a game-changer.

First manuscript accepted! Moving back to Norway after two years in the US

2 minute read

Published:

Let’s go!!! 🎉 After nearly two years of hard work, my first article for the BOULDERING project has been accepted in the Journal of Geophysical Research: Planets! You can check it out here. This has been a long road, but also an incredibly rewarding one. Over 33,000 boulders were collected across more than 750 tiles/images from three planetary bodies: Earth, Mars, and the Moon.

Wifey at Stanford now! Baby Boss incoming in September

1 minute read

Published:

Big life updates, everyone! Siri is now at Stanford! Thanks to a fantastic program from the Norwegian Research Council (NRC), postdocs already involved in NRC projects can apply for a research stay abroad for up to a year. The NRC covers the salary, and with a high success rate for these grants, it’s a great way to build international collaborations. After some searching, Siri found a group at Stanford doing work closely related to her research at NMBU in Ås. A few weeks of visa wrangling later, she’s here!