Dell Technologies Realize magazine Issue 4 | Page 33

PHOTO BY DIRK DANIEL MANN/SHUTTERSTOCK warmer than their environment—so, with thermal imaging, they stand out really brightly against this cooler background.” Mounting thermal cameras on drones allows conservationists like Wich to survey vast areas of parkland or wilderness that would be impossible to cover on foot. But, once you have all this footage, someone still needs to identify the animal that each thermal blob represents. “We were gathering so much data that it quickly became very laborious to have a person go through all these thousands and thousands of images,” Wich says. “When Steve suggested he could help, I thought he meant offering to put some volunteer time into manually labeling the images. But then he explained how in astrophysics they’ve developed machine-vision algorithms for automatically classifying stars and galaxies from thermal data and that this might be applicable to identifying animals from our thermal imagery.” AUTOMATING ANIMAL IDENTIFICATION To retrain these algorithms for species instead of star clusters, Longmore founded the Astro-Ecology Group. First, his team gathered thermal camera footage from a nearby farm to see if these algorithms could be taught to identify the difference between the thermal signature of humans versus cows. When this was a success, they visited local safari parks and zoos to build up a database of numerous thermal images of multiple species in different environments. In order to be used for machine learning, these images would need to be labeled with the correct identifications. For this, they turned to the citizen science platform Zooniverse, which was created in 2009 to allow members of the public to assist astronomers by hand-classifying types of galaxies. Since uploading their first set of data to the platform in February 2019, Longmore’s group has received the help of over 4,000 volunteers, who’ve scoured around 27,000 images from Knowsley Safari Park for the “ We were gathering so much data that it quickly became very laborious to have a person go through all these thousands and thousands of images.” —Serge Wich, Liverpool John Moores University presence of animals, and tagged them as an example for the team’s automated system to follow. After this initial training with captive animals, the team next took their system out into the wild. So far, it has been used to successfully identify and count orangutans in Borneo, spider monkeys in Mexico, river dolphins in Brazil, and riverine rabbits in South Africa. “The riverine rabbit is the most endangered animal you’ve never heard of,” Longmore says. “It’s the 13th most endangered mammal on the planet, and there are only a few hundred living in an area of the South African desert the size of Austria. So the fact that we were able to get a confirmed identification of one was really exciting.” There have been challenges, however. While endangered animals may be much closer than 31