For birdwatchers it is a reality that some species are devilishly difficult to distinguish, particularly sparrows and monotonous birds called "little brown jobs." Distinguishing individuals is almost impossible. Now, a computer program that analyzes photos and videos has accomplished that feat. The advance promises to reveal new information on the behavior of the birds.
"We spent a lot of time with binoculars, crouched down, looking at the birds and their legs," says Iris Levin, a behavioral ecologist at Kenyon College who was not involved in the new work. The reason: For years, researchers have identified birds by putting colored bands on their legs. They use these bands to identify birds in the wild, and in photographs and videos in the laboratory. The task can often be laborious, says Levin.
Specially equipped tags can make your job easier by including GPS and proximity sensors that record when animals interact. Passive Integrated Transponder (PIT) tags are also used to prevent shoplifting and identify pets, ping connected antennas when a bird lands within a few inches. Behavioral ecologist Claire Doutrelant from CNRS, the French national research agency, and her colleagues have been adding these little tags to the rings of sociable weaver birds (Philetairus socius) since 2017.
Sociable weavers work together to build large nests in southern Africa, often in acacia trees. Nests can weigh up to 1 ton and house up to 200 birds in individual chambers. Their cooperative behaviors also include raising chicks and defending against snakes and hawks. To study these behaviors, researchers identify and track hundreds of individual birds.
Feeder antennas keep track of birds that live in the colony. But it has not been possible to obtain more granular information, such as which birds contributed the most to community activities. And Doutrelant and her colleagues can't put antennas around the nest - birds are wary of them, and their cameras are too close to each other for reliable data collection.
Then team member André Ferreira, Ph.D. A student at the University of Montpellier, he decided to try a kind of artificial intelligence. Called a convolutional neural network, the tool reviews thousands of images to discover what visual characteristics can be used to classify a given image; then use that information to classify new images. Convolutional neural networks have already been used to identify various species of plants and animals in the wild, including 48 types of African animals. They have even accomplished a more complicated task for elephants and some primates: distinguishing between individuals of the same species.
An algorithm tracks individual birds as they cooperatively build their huge communal nest.
Ferreira fed the neural network with several thousand photos of 30 sociable weavers who had already been tagged. "No one had devised an efficient method to collect these training data sets," he says. To take the photos, she installed cameras near bird feeders equipped with radio frequency antennas. As soon as the birds landed, a small computer recorded their identity using their PIT tag, and a camera took photos of their backs every 2 seconds. (The rear view is the part of the bird that is seen most frequently while nesting or feeding.)
After just 2 weeks, Ferreira had enough photos to train the neural network. "We weren't sure if it would work," recalls Doutrelant. "We have observed these birds a lot and have never been able to recognize them without the colored rings." But when they were given photos they hadn't seen before, the neural network correctly identified individual birds 90% of the time, they report this week in Methods in ecology and evolution. Doutrelant says it is as accurate as humans trying to spot colored rings with binoculars.
Ferreira then tested the approach on two other bird species studied by Damien Farine, a behavioral ecologist at the Max Planck Institute for Animal Behavior. The tool was so accurate in identifying captive zebra finches and big boobs in the wild. Both species are extensively studied by ecologists.
But Gail Patricelli, a behavioral ecologist at the University of California, Davis, sees some limits on the approach. For example, with species that are difficult to capture and tag, it could be difficult to obtain the thousands of identifiable photographs necessary to train the neural network. She studies the greater grouse, a species in decline, and tries to avoid manipulating them because it stresses the birds. Another potential limitation: When birds move, the neural network may not recognize them and would need to be retrained. Ferreria is collecting photos of other features, such as the appearance of the head, to improve the tool.
The biggest limitation with the current neural network, says Ferreira, is that it tries to identify each bird as one it already knows, so it cannot recognize a new individual. Ferreira is now working with Farine to test a different type of neural network that could do that; would need to be trained in images of many more birds. If the dataset were large enough, the tool could be used even by researchers who have not tagged their birds. "I think this would be a complete game changer," says Farine.
Despite those limitations, Patricelli calls the new work "exciting" and says it opens up possibilities for studying many other bird species and behaviors. "The fact that this algorithm was able to distinguish them, when they look very similar to the naked eye, is definitely surprising."
