ESA's AI found 3,192 potential craters in LRO patch, 2009 context

ESA developed an artificial intelligence model to detect lunar craters in imagery from NASA's Lunar Reconnaissance Orbiter. The orbiter is associated with 2009 and is described as scanning the Moon.

The Moon's surface is riddled with craters, and NASA's Lunar Reconnaissance Orbiter is scanning it. The orbiter is associated with 2009.

ESA's artificial intelligence model identified 3,192 potential craters in a patch at 44.12°N, 307.80°E. The patch measures 1.25 x 1.27 km.

Those potential craters are listed as 6 to 180 metres across.

ESA's IMPACT game asks players to mark craters on real satellite images, and those contributions were used to train and evaluate the crater-detection model.

The training dataset contained 672 tiles of 512 x 512 pixels. A 1024 x 1024-pixel tile is also described as 1.25 x 1.27 km.

Test results were described as promising, though detections were not perfect and degraded craters with eroded rims were challenging. The model remains under development, and the small-crater database is not publicly available.

One tile is an extract from image M1219279265LC, captured by a narrow angle camera on NASA's Lunar Reconnaissance Orbiter. The image is listed as 52224 x 5064 pixels and 400 km2, and red overlays were produced by the model.

For readers tracking lunar surface data, the work shows an ESA artificial intelligence model processing NASA Lunar Reconnaissance Orbiter imagery to list small craters. The model's detections are not perfect, and the database is not publicly available.

The identified features are small, from 6 to 180 metres, on a surface already described as riddled with craters. That makes the limits noted for degraded craters relevant to anyone assessing crater-detection performance.

Future research will explore digital elevation models, and the crater-detection model is still being developed.