New research by researchers from institutes in France, Germany, Portugal, and South Africa proves that artificial intelligence (AI) can be used to train computers to recognize individual birds, in the research published in British Ecological Society journal Methods in Ecology and Evolution. In the research, they used thousands of identified images of birds they had collected to train and test Artificial Intelligence to identify individual birds.
Dr Andre Ferreira, at the Center for Functional and Evolutionary Ecology, France and lead author of the research said that the study helps to overcome one of the greatest shortcomings in the study of wild birds—reliably recognizing individuals.
The researchers trained the AI models to identify images of individual birds in the wild populations of some of the most commonly studied birds in behavioural ecology. Researchers after training the AI models tested it with images of the individual birds they had not detected before and had an accuracy of above 90% for the wild species and 87% for the captive zebra finches.
Individually identifying animals is one of the most expensive and time-consuming parts in animal behaviour studies, which limits the scope of behaviours and the size of the populations, researchers can study. Identification methods like attaching colour bands or stripes to the legs can be stressful to the birds.
AI models can identify images if they are trained with thousands of labelled images. Companies like Facebook, which have access to millions of pictures of different people that are voluntarily tagged by users can do human recognition with ease. But the difficulty in acquiring such labelled photographs of animals has created a bottleneck in research.
The researchers were able to overcome this challenge with camera traps and sensors. Most birds in the study populations carried a passive integrated transponder (PIT) tag, which is similar to the microchips implanted in pets. Antennae on the bird feeders were able to identify the bird from the tags and trigger the cameras. The limitations can be overcome with more datasets containing thousands of images of thousands of individual birds over long periods, which they are currently trying to collect.
