Counting Individual Birds in a Colony with Machine Learning and Drones #drone #droneday
From New Scientist:
Jarrod Hodgson and his colleagues at the University of Adelaide in Australia had previously used aerial images from drones to count seabirds and found that the drones had a more comprehensive view of the colonies than the people trying to count them on the ground.
However, neither could provide an exact count of the individual birds. “We couldn’t test for accuracy,” says Hodgson.
So he bought many hundreds of plastic duck decoys to simulate a flock of greater crested terns on a beach in South Australia – much to the amusement of his fellow researchers.
“A shipment arrived and my colleagues had just thrown them all in my office – I couldn’t even get to my desk,” he recalls.
After the team set the fakes up on the beach, drones began taking aerial photographs of the “colonies”, one of which had more than 1000 plastic ducks. Humans counted the fake birds in these images, and also made counts from vantage points on the ground.
The highest-quality drone images – taken from altitudes of 30 metres – allowed for counts that were more than 90 per cent more accurate than those made by humans on the ground.
The team also developed a machine learning system to count the proxy birds in the photographs automatically. After training, it proved to be about as accurate as humans. The work will be presented at the International Congress for Conservation Biology in Colombia this week.
Welcome to drone day on the Adafruit blog. Every Monday we deliver the latest news, products and more from the Unmanned Aerial Vehicles (UAV), quadcopter and drone communities. Drones can be used for video & photography (dronies), civil applications, policing, farming, firefighting, military and non-military security work, such as surveillance of pipelines. Previous posts can be found via the #drone tag and our drone / UAV categories.
from Adafruit Industries – Makers, hackers, artists, designers and engineers! http://ift.tt/2uPF5EK
via IFTTT