By Larry Powell
 |
An adult female grizzly (Ursus arctos). "BearID," as the program is called, captures a bear’s face in a photo image, rotates, extracts and embeds it in order to classify the individual. Facial recognition techniques have long been used to identify primates, including humans. But, up 'til now, there's really been no effective way of identifying wild species like the grizzly (brown) bear who, unlike the zebra or giraffe, lacks unique and consistent body markings. In co-operation with two US software developers, four scientists from the University of Victoria bought their idea to reality. They tested their system on grizzlies at two locations - Knight Inlet, BC, and Katmai National Park, Alaska. After taking thousands of pictures, they were able to positively identify 132 individuals with almost eighty-four percent accuracy. |
 |
| An adult female in another colour phase. All images by Melanie Clapham, U of Victoria, Canada. |
The technology enables wildlife monitoring on larger scales and in higher resolution than before. And it can be applied, not only to the grizzly, but to many other mammals, as well. This, in turn, could allow conservationists and lawmakers to tackle global challenges such as biodiversity and habitat loss.
Knight Inlet, one of the two locations for the research, is home to a First Nations Lodge where bear-watching forms part of the local eco-tourism industry.
A band official there, Dallas Smith is impressed with the results.
“This amazing technology will help us identify individual bears and better understand their movement and interactions throughout our territories, which will enable us to build better management plans around habitat protection. It will also help us manage and mitigate the impact of wildlife viewing, as well as positioning ourselves to more effectively and efficiently deal with bear-human conflicts that are becoming more and more prevalent.”
These new ways of using facial recognition technology, referred to as a "deep learning approach," were published recently in the journal, Ecology and Evolution.
