Canadian student uses AI to identify birds

The work of a student at the University of Alberta (Edmonton, Canada) demonstrated how artificial intelligence technologies can facilitate the study of birds, according to the бюллетень of the university.

Priscilla Adebandji, a computer science student, spent the summer experimenting with AI to improve the analysis of video footage tracking red-winged blackbirds and their nests.

Her solution, if fully implemented, would save the hours required to manually identify individual birds, confirmed Professor Ivana Schopf, who studies the impact of parasite infections on avian behaviour.

Current methods involve recognizing birds and their behaviour by their calls. This requires hours of footage review by two staff members (for consistency), who need a certain level of knowledge and experience, she noted.

Although existing software can track the movements of animals like mice in laboratory settings, doing so in the wild is more challenging due to non-ideal conditions, according to Schopf. Another problem is video quality, as nests are well concealed in swamp vegetation.

Schopf turned to Adebandji’s supervisor, Professor Thibo Lutelle, in search of a way to automatically detect birds without manual viewing, and provided recordings from two field seasons — a total of 30 hours of video.

“We felt there were many AI applications that could help, though we had no idea what to expect. We needed to determine what type of machine learning to use. We did a lot of preparatory work and research,” Luetelle said.

Adebandji had to address various issues, including false positives produced by existing AI models.

“Sometimes they mistook a bird for an airplane, and things like shadows were incorrectly identified as bears in the background,” the student explained.

Using computer vision tools and motion-detection algorithms in video analysis, she improved tracking quality to the point where birds could be counted, distinguished from other objects, and identified.

By the end of the summer, Adebandji’s software could determine the exact times when the birds arrive at and depart from their nests. According to her, automatic detection reduced the time needed for this task from eight hours to a couple of minutes.

Although the software still needs refinement, Schopf hopes to eventually use it to remove the labour-intensive process of manual data collection for her study. In her view, the solution has potential for wider application in animal research.

As reported in May, the well-known Bitcoin critic, billionaire Warren Buffett compared AI to an atomic bomb and expressed concern about the rapid development of technologies.