Bird-identifying AI could put an end to leg bands
If you saw a finch one time, chances are you'd have great difficulty picking it out from a large group of finches later on. A new artificial intelligence-based system can do just that, though, potentially making life much easier for both biologists and the birds that they study.
Ordinarily, if a wildlife biologist wants to track an individual bird, they have to capture it, put an identity band on its leg, release it, then later recapture it to read that band. Needless to say, doing so is quite a hassle for the scientist, and very stressful to the bird. There are now also remotely readable GPS tags, although these still have to initially be attached to the animal.
Seeking a better alternative, an international consortium of research groups has created an AI system that can identify individual birds based on nothing but photos.
It was trained on a database of images of thousands of birds, each animal displaying distinctive patterns in their plumage. In this way, the system learned what to make note of in subsequent images, in terms of unique features.
The AI was then tested on wild populations of great tits and sociable weavers, and a captive population of zebra finches. In all cases, it utilized a camera installed at a feeding station, which started by getting an initial close-up shot of each bird. When that bird later returned and got photographed again, the system was able to match that photo up to the first one, determining that both shots were of the same animal.
So far, the system has proven 87 percent accurate at identifying individual finches, and over 90 percent accurate with the wild birds.
In order to gauge that accuracy, most of the birds had already been equipped with passive integrated transponder tags, not unlike those implanted in dogs and cats. When that tag was read by antennae at the feeding station, the system recorded the tag's individual code, plus it triggered the camera to take a photo. This means that all the photos were of animals that had also been identified by their tags – in practical use, of course, the system would utilize nothing but the photos.
It should be noted that the scientists have yet to determine how the AI may be affected by changes in birds' appearance over time, such as when they go through feather-molting cycles.
Taking part in the study were researchers from the University of Porto (Portugal), the Max Planck Institute of Animal Behavior (Germany), the CNRS institute (France), the University of Paris-Saclay, the University of Konstanz (Germany), the University of Montpellier (France), and the FitzPatrick Institute of African Ornithology (South Africa).
The AI system is described in a paper that was recently published in the journal Methods in Ecology and Evolution.