A researcher built a tool to collect data to identify individual birds by setting up a motion-activated camera and using facial detection
As a birder, I had heard that if you paid careful attention to the head feathers on the downy woodpeckers that visited your bird feeders, you could recognise individual birds. This intrigued me. I tried sketching birds at my own feeders and found this to be true.
I knew that researchers had used machine learning techniques to recognise individual faces in digital images with a high accuracy. I thought — Would it be possible to apply those techniques to identify individual birds?
So, I built a tool to collect data: a type of bird feeder favoured by woodpeckers and a motion-activated camera. I set up my monitoring station in my yard and waited for the birds.
Image classification
It’s a hot topic in the tech world. Companies like Facebook, Apple and Google are researching to provide services like auto-tagging of friends in social media posts, using your face to unlock the cellphone, and visual search.
When I started researching, image classification research focussed on a technique that looked at image features such as edges, corners and areas of similar colour. Those approaches were 70 per cent accurate, using benchmark data sets with hundreds of categories.
Recent research has shifted toward the use of artificial neural networks, which identify their own features proving most useful for classification. Convolutional neural networks, the type that we are now using in our work with birds, are modified in ways that were modelled on the visual cortex.
Progress on bird ID
All the images were taken from the same perspective, and fell into limited categories. Only 15 species ever visited the feeder. Of those, only 10 visited often enough to provide a useful basis for classification. The limited number of images was a definite handicap, but lesser categories worked in our favour. It could recognise whether the bird in an image was a chickadee, a Carolina wren, a cardinal or other. The project achieved about 85 per cent accuracy. Identifying birds in images is an example of a “fine-grained classification” task, that is, the algorithm tries to discriminate between objects that are only slightly different. Many birds at feeders are roughly the same shape so telling the difference between two species can be quite challenging.
The challenge ups when you try to identify individuals. For most species, it isn’t possible. The woodpeckers have strongly patterned plumage but are still largely similar.
I found that the head feathers of downy woodpeckers weren’t a reliable way to distinguish between individuals, because those feathers move around a lot. They’re used by the birds to express irritation or alarm. However, the patterns of spots on the folded wings seemed to work just fine to distinguish. Those wing feathers were almost always visible in our images, while the head patterns depended on the angle of the bird’s head. In the end, we had 2,450 pictures of eight different woodpeckers. When it came to identifying individual woodpeckers, we achieved 97 per cent accuracy. However, it needs further verification.
How can this help birds?
Since many species are very specific in their habitat needs when it comes to breeding, wintering and migration, fine-grained data could be useful for thinking about the effects of a changing landscape. Data on downy woodpeckers could then be matched with information, like land-use maps, weather patterns, population growth, to better understand the abundance of a local species.
Recent studies suggest that it should be possible to train a classifier using a much broader group of images, fine-tune it quickly and with computational demands to recognise birds.
Projects like Cornell Laboratory of Ornithology’s eBird are working to monitor population dynamics, but the bulk of those data tends to be from locations where people are numerous, rather than from locations of specific interest to scientists. An automated monitoring station approach could provide a force multiplier for wildlife biologists concerned with specific species or locations. This would broaden their ability to gather data with minimal human intervention.
