A research team has shown for the first time that computers can be trained with AI to recognize individual birds without any tags or marks. This is not possible even for the most sharp-eyed of ornithologists as one bird of a species looks very similar to another bird of the same species.
“We show that computers can consistently recognize dozens of individual birds, even though we cannot ourselves tell these individuals apart. In doing so, our study provides the means of overcoming one of the greatest limitations in the study of wild birds—reliably recognizing individuals,” said lead author Dr. André Ferreira from the Center for Functional and Evolutionary Ecology (CEFE), France.
Currently, to identify individual animals and study their behavior, researchers use methods like attaching color bands to the legs which are difficult for us and stressful for animals and birds as well. This was a big limitation in studying individual behavior. But with the new AI model, this problem solved.
The lead researcher Ferreira while in South Africa was trying to study the co-operative behavior of the sociable weaver, a bird that works with others to build the world’s largest nest. But he faced difficulties in identifying individual birds and study their behavior.
So, he with researchers from France, Germany, Portugal, and South Africa built the AI model to recognize individuals simply from a photograph of their backs while they were busy nest-building.
He with researchers from Max Planck Institute of Animal Behavior in Germany showed that this model can be applied to two of the most commonly-studied species in Europe: wild great tits and captive zebra finches.
To build this model, the research team built feeders with camera traps and sensors. Almost all the birds in the study populations carried a passive integrated transponder (PIT) tag. Antennae on the bird feeders were able to read the identity of the bird from these tags and trigger the cameras.
The team then trained the AI model with the images captured using the above method to recognize individual birds. This study uses a type of deep learning known as convolutional neural networks, these are optimal for solving image classification problems.
This model have two main drawbacks. It can’t identify a bird which wasn’t shown to it before. As time passes, the birds age and their body changes, it is not know how efficiently the model can identify individuals at that time.
Anyhow, the model is a very big problem solver for behavioral analysis of individual animals and will probably help us in finding a lot of new things about birds and animals.
Journal Reference
André C. Ferreira et al, Deep learning‐based methods for individual recognition in small birds, Methods in Ecology and Evolution (2020). DOI: 10.1111/2041-210X.13436
