Using AI to analyze photos taken from above, discovered a series of mysterious giant drawings in the middle of the desert in Peru
- Tram Ho
The Nazca Desert, located on the coast of southern Peru, is famous for its huge population of strange drawings known as the “Nazca Lines”. The drawings include people, fish, birds, bugs, lizards, dogs, cats, monkeys, trees and flowers. Some of the images are more supernatural, including a two-faced humanoid monster and a giant two-headed man-eating snake. They are believed to be the product of the ancient Nazca people from the 1st to 6th centuries AD.
The remarkable thing about the “Nazca lines” is that their size is so large (the largest drawing is up to 275m long) that it can only be seen clearly from above using satellite imagery, aerial photography, scanners aerial photography and drone photography. Some of the drawings have become very blurred due to the effects of time, requiring researchers to use artificial intelligence to help in the search process.
A strange shape in the Nazca desert was captured by an aircraft at an altitude of nearly 4km.
Using ‘deep learning’ technology to find mysterious drawings
Previously, the team at Yamagata University spent about five years analyzing high-resolution aerial photographs of the entire area. During that time they identified several new figures. But when they got their hands on even higher resolution aerial photographs in 2016, they turned to an artificial intelligence method called “deep learning” to examine them.
This is a method made in collaboration with IBM Japan and the IBM Thomas J. Watson Research Center in the US, which conducts advanced research on artificial intelligence. Deep learning systems and the methods they use to process data are inspired by how the human brain processes information. This method exhibits high performance in tasks such as image recognition, speech processing, and natural language processing. Through deep learning, the machine automatically identifies specific objects from images and their location, size, and classification.
Normally, a deep learning system is trained on thousands or millions of known objects, but the team at Yamagata University trained the system with only data from 21 known Nazca geoglyphs, they are divided into “elements”.
Any known geographical figure is made up of dozens of these elements — such as a head, torso, arms, or legs. And so the new deep learning system can only find parts of the geograph.