Purpose-built facial recognition software aims to identify individual kākā

Friday, 5 November 2021

Researcher Dr Racheal Shaw has created a feeder to correctly position kākā for their headshots. One correctly poses for its head shot, and another has a quick taste of the lens.

Two researchers are in the early stages of a project to identify individual kākā by their features – and then train facial recognition software to do the same.

Researcher Dr Rachael Shaw is working on the design for a bird feeder, based at the fenced Zealandia sanctuary, which positions the birds at the right angle for pictures to be taken in profile as they feed. The images will be used to create a facial recognition software to identify individual kākā.

Being able to track them would allow researchers and conservationists to get a better idea of the total population, how far they travel, and how information spreads in a community, Shaw said.

But between their high intelligence and cheeky personalities, the kākā aren't making it easy. From drinking upside down, to licking the lens, to detaching and throwing the whole camera down a bank, Shaw has her work cut out for her.

**READ MORE:

North Island kākā are classed as at risk (recovering) by the Department of Conservation.

* Facebook shutting down face-recognition system, deleting 1 billion faceprints

* Language Matters: Choosing a family language policy that works for your whānau

Researcher Rachael Shaw says she has a lot of questions that could be answered with a better understanding of the kākā population.

* Tuatara are slow and endangered, but their super-fast sperm could save them yet

* Police facial recognition: new study calls for greater oversight in NZ

A kākā uses the bird feeder at Zealandia, which allows pictures to be taken in profile of them as they come to feed.

North Island kākā are classed as at risk (recovering) by the Department of Conservation, and South Island kākā as nationally vulnerable.

Shaw guesses there are 180 to 200 kākā using the Zealandia feeders, and a few hundred living in wider Wellington.

Shaw is a lecturer in behavioural ecology at Te Herenga Waka, Victoria University of Wellington, and she’s been part of kākā research at Zealandia for the past few years – particularly their problem-solving abilities.

“It will hopefully enable us to have a good idea of the population structure of birds living in Wellington, their survival, and their behaviour outside the fence,” she said.

Not all the birds were banded; although it was hard to estimate the size of the total population, Shaw guessed there were 180 to 200 using the Zealandia feeders, and a few hundred living in wider Wellington. Only around 30 to 40 per cent had identification bands, which were put on when they used Zealandia nesting boxes.

Pictures were captured inside a box containing a motion-activated camera and a liquid nectar dropper, which had to be positioned at the right angle to capture a profile shot as it drank.

The resulting images would show differences in beak texture and shape. “They moult and change their feathers on a seasonal basis, so you want to pick a feature that will stay constant.”

Work on the feeder had been constant for the past three weeks. “Hopefully by the end of February we’ll have a good idea of how feasible this might be,” Shaw said.

The images would be fed into an algorithm designed by artificial intelligence lecturer Dr Andrew Lensen.

“The ideal case,” Lensen said, “is if we have pictures of birds, and we also know what their band colours are.”

Through a process of continuous tweaking and “iterative refinement” – a feedback loop of correction until the program could correctly identify each bird, the result checked against its band colours – which would give confidence in its ability to correctly identify the non-banded birds too.

It would hopefully answer another big question; did these features change over time?

In a world where AI gets a bad rap, with mistrust of large corporations and the misuse of personal information making headlines, Lensen said this was an opportunity to show people a homegrown example of artificial intelligence being used for good.