lunar-crater-shadowcam.jpeg
A ShadowCam full-resolution segment of the Faustini crater
The Moon's South Pole is a region of particular scientific interest and importance for future lunar exploration. Permanently shadowed craters the region hat have remained in darkness for billions of years, playing host to some of the most intriguing geological features in our Solar System. These deep craters have never been touched by the warmth of direct sunlight, are it is here that there may be significant deposits of water ice, which could be crucial for supporting and sustaining future human missions and potential lunar bases.
The Moon's south pole captured by the Clementine mission (Credit : NASA/JPL-Caltech)
Until recently, not a huge amount was known about the polar craters but enter the ShadowCam, a NASA-funded instrument on Korea’s Pathfinder Lunar Orbiter (KPLO.) Unlike previous imaging technologies, ShadowCam can capture images that are 200 times more sensitive, with a resolution of 1.7 meters per pixel. These high resolution images have facilitated the discovery of millions of previously unknown impact craters in these dark areas of the Moon. These newly identified craters provide crucial insights into lunar surface processes like impact events, volatile material distribution, and geological changes.
Artist impression of Korea Pathfinder Lunar Orbiter(Credit : Ministry of Science and ICT)
A research team led by P. Pokorny from the The Catholic University of America has developed specialised crater detection techniques to analyse the data. They employed advanced machine learning techniques identify craters in the images using the YOLOv8 object detection framework. YOLOv8 stands for ‘You Only Look Once version8’ and is a nod to the frameworks capability to locate multiple objects within an image in a single forward pass through the neural network, which makes it incredibly fast and efficient.
The neural network was developed with 25.9 million parameters specifically designed to identify craters across various image sizes. Their approach involves processing ShadowCam images by dividing them into multiple overlapping tiles at different resolutions which are then rescaled and analysed. To improve the accuracy, the team also used image augmentation techniques to eliminate duplicate detections. It took some work to train the model though using 5,240 impact craters from Lunar Reconnaissance Orbiter images.
Artist concept of NASA's Lunar Reconnaissance Orbiter (Credit : NASA)
They ran the model across 22,256 ShadowCam images, covering 5.3 million square kilometres of the Moon totalling 2.2 TB of data! It was an impressive feat though with the computational process requiring 3000 Graphics Processor Unit hours but it resulted in the identification of 1,013,440,231 impact craters larger than 16 meters in diameter! If you think this is impressive, it completed that task at a rate of 0.3 microseconds per crater and that’s with a mere1.8% false positive detections for craters between 16 meters and 4 kilometres in size.
With the success of their crater detection methodology, the team are now looking to apply their algorithm to future ShadowCam images. They’re not stopping here though as they want to improve the model by focussing on enhancing the detection capabilities for challenging crater types, including those in low-signal regions, degraded formations, and morphologically complex structures.
Source : Machine Learning Driven Detection of 1 Billion+ Lunar Impact Craters in Permanently Shadowed Regions Using Shadowcam Data
Mark Thompson
Science broadcaster and author. Mark is known for his tireless enthusiasm for making science accessible, through numerous tv, radio, podcast and theatre appearances, and books. He was a part of the aware-nominated BBC Stargazing LIVE TV Show in the UK and his Spectacular Science theatre show has received 5 star reviews across UK theatres. In 2025 he is launching his new pocast Cosmic Commerce and is working on a new book 101 Facts You Didn't Know About Deep Space In 2018, Mark received an Honorary Doctorate from the University of East Anglia.
You can email Mark here