The first spacecraft to ever soft-land on the Moon wasn’t American, it was the Soviet Luna 9. Luna touched down in February 1966, three years before Apollo 11, it sent back photos and everything. One problem… nobody could find it. For 60 years the exact landing spot was uncertain, so while the NASA’s Lunar Reconnaissance Orbiter can photograph the Moon down to 25 cm per pixel, they didn’t know where to look. You can see Apollo descent stages, rover tracks, individual pieces of equipment, but no Luna.
A team from Birkbeck College, the SETI Institute, and JAXA built something called YOLO-ETA (You-Only-Look-Once – ExtraTerrestrial Artefact), a lightweight convolutional neural network adapted from TinyYOLO v2. They trained it on 125 hand-labeled images of Apollo landing sites, taught it what spacecraft hardware looks like from orbit, then turned it loose on the area where Luna 9 was supposed to be.
Annnnd… It found something! A cluster of objects near 7.03° N, –64.33° E, about 5 km from the old reported coordinates. The model flagged the same features across nine different LROC images taken under different lighting conditions, with confidence scores up to 77%. Nearby dark patches could be impact craters from the lander’s jettisoned side modules. The team also matched the local terrain against Luna 9’s original surface panoramas… and it lines up!
There is some stuff to check out GitHub if you want to try it yourself, ya know… for when you want to find things on the moon.
An AI trained on American moon landing sites found a Soviet spacecraft that’s been chilling on the Moon since 1966. Come back next time to catch my new show on the History Channel “Cold War Archaeology with Neural Networks” sponsored by Burger King.
Sabine Hossenfelder has a great video on this too, of course, I should have posted this sooner! There is a lot more to this story, so might do some more later….
from Adafruit Industries – Makers, hackers, artists, designers and engineers! https://ift.tt/PqHo6Xe
via IFTTT
