Time-saving tool detects, identifies animals in research photos

n

n n n n

n n

n n n n

n n

n n n n

n n

n n n n

n n

n n n n

n n n

GIFU–A Gifu University team developed a computer program that can accurately detect, identify and count animals in photos, saving researchers the trouble of having to analyze thousands of pictures for subject matter.

The researchers enabled the program to “learn” data from 117,457 photos taken by cameras set up to study the habitats of wild animals. The cameras had a mechanism to start recording or shooting after sensing an animal’s presence with infrared light.

Masaki Ando, an associate professor who specializes in wildlife ecology and management at the university, and other staff members took the photos at 20 locations in a forest in Gero, Gifu Prefecture, between February 2014 and November 2016.

The team added data, such as presence or absence, species and numbers of animals, to the photos.

But some photos showed no animals, so the researchers had to perform the time-consuming task of checking every picture to determine the species and the numbers.

Kunihito Kato, head of the university’s AI research promotion center and an associate professor of image recognition, made the computer program learn the data.

The program used singled out photos showing 18 kinds of animals, including Japanese deer, serows, raccoon dogs, foxes and humans, with a 99-percent accuracy.

In addition, it could count the numbers of four types of animals–Japanese deer, serows, boars and Asiatic black bears–with an accuracy topping 80 percent.

“The program does not reach the level of accuracy of researchers, but its level exceeds those of amateurs,” Ando said. “We would like to place many cameras in a wider area and create a mechanism to understand the habitats of wild animals.”

He said such efforts would make it much easier to plan the protection or capture of wild animals.

n n