At present, in the study of bird individuals, field biologists should first capture it, then tie something that can identify it on its legs, and then capture and read it again after release. This is quite cumbersome for scientists, and it can also have an impact on birds.

in search of better alternative solutions, an international research team has created an artificial intelligence system that can identify birds through photos. The system has been trained on image databases of thousands of bird species, each of which displays a unique pattern of feathers. In this way, the system learned the unique features to pay attention to in subsequent images.

the team tested on wild great tit, social weaver and zebra finches in captivity. In all cases, it first takes an initial close-up of each bird using a camera installed at the feeding station. When the bird returns and is photographed again, the system is able to compare the photo with the first one to determine that the two photos are the same animal.

so far, the system has been proved to have an accuracy rate of 87% in identifying individual finches and more than 90% in wild birds.

to measure this accuracy, most birds have been equipped with passive integrated transponder tags, which are different from those implanted in dogs and cats. When the tag is read by the feeding station’s antenna, the system records the tag’s personal code and triggers the camera to take photos. This means that all the photos are the same animals identified by the tags – in practice, of course, the system only uses photos.

researchers from the University of Porto (Portugal), the Max Planck Institute for animal behavior (Germany), the CNRS Institute (France), the University of Paris Thackeray, the University of Constance (Germany), the University of Montpellier (France) and the Fitz Patrick Institute of African birds (South Africa) participated in the study.