Artificial intelligence model trained to identify birds in new images

Artificial intelligence could help researchers in studying avian behaviour

For the first time, researchers have developed an AI model that is able to identify individual small birds in new images.

Identifying individual animals in animal behaviour studies is an important step to answering many questions in evolutionary biology. However, it is usually a very time-consuming job and limits the size of populations and scope of behaviours that researchers can study.

Moreover, current identification techniques, such as attaching coloured bands to birds' legs, are also sometimes stressful to the animals that are being studied.

The new study, published in the British Ecological Society journal Methods in Ecology and Evolution, describes the process for using AI technology to individually identify birds.

"We show that computers can consistently recognise dozens of individual birds, even though we cannot ourselves tell these individuals apart," said Dr André Ferreira, a PhD student at the Centre for Functional and Evolutionary Ecology (CEFE) in France, who is also the lead author of the study.

As part of the study, the team, which included researchers from South Africa, Germany, Portugal and France, collected thousands of labelled images of three small bird species: great tits (Parus major), sociable weavers (Philetairus socius) and captive zebra finches (Taeniopygia guttata).

To capture photographs of the birds, the researchers first built bird feeders with sensors and camera traps. Most birds who were photographed carried a tag similar to microchips implanted in dogs and cats. As a bird entered the feeder, the installed sensors read its identity through the tags and then triggered the camera to capture the image.

The researchers then used these images to train their AI model to recognise individual birds. The model was tested using new images of individual birds in a different context.

According to the researchers, their AI model displayed an accuracy of over 90 per cent in re-identifying sociable weavers and wild great tits in new images collected in contexts that differed from the ones originally used to train the model, and an accuracy of 87 per cent in identifying captive zebra finches.

While the current AI tool can only re-identify individual birds that it has been shown before, the researchers said that they are now trying to create more efficient AI models, which can identify individuals even if they have never seen them before.

"Overall, our work demonstrates the feasibility of applying state-of-the-art deep learning tools for individual identification of birds, both in the laboratory and in the wild," the researchers said.

"The ability to conduct individual recognition of birds without requiring external markers that can be visually identified by human observers represents a major advance over current methods."