Training the palmy paparazzi

Machine learning was an interest that I kind of fell into. So I was surprised it’s why I was recommended for a conservation project, especially so early in my IT career.

I was very appreciative, though, because the palm cockatoos project was such an amazing opportunity. Not only to contribute to such an important environmental project, but also to meet and work with such a passionate team of people.

The team at Weipa – led by Celina, a Threatened Species Researcher – had set up 10 cameras, each pointed at a nesting hollow and taking photos roughly every minute. So we took the huge bank of images they’d already manually reviewed and tagged, and used it to train a computer model called YOLOv5 to find the cockatoos. It’s an open-source, object detection model that kind of works like filters that use facial recognition on social media. Once it was trained, the model could identify which images were likely to have birds in them, and pointed out where the bird could be found in the picture, building trust with users. It automated the huge job of checking tens of thousands of pictures, which they’d previously been doing manually.

With my skills and background, I knew I couldn’t make a big difference going out personally into the field. But I’m good at making research tools more efficient, and that means the rest of the team could focus more energy on what they’re good at. And collectively, we can make a bigger impact.

Going to Weipa and seeing the research in action was both a highlight and a reality check. I’d never visited a mine site before, so it was really helpful to see the conservation work in person. The sheer size of the site was hard to comprehend until I saw it myself. I’d also wondered why the cameras weren’t closer to the hollows, but after seeing them in person, I could see that they’d definitely placed them as close as they could. So working with the team on the ground really helped me understand their challenges and reasons for doing things a certain way.

Other teams are now adapting the work we did for new conservation projects, too. The Weipa team are exploring using the same process to tag and analyse audio data like bird calls to isolate certain species. And teams in other parts of the world have been looking at how they can use our model for their own sites – some have more than 60 species they’re tracking, so a model like this would make their research and conservation work much more efficient.

Being recognised as a finalist for the QRC Award was so humbling. And I was really touched that a project focused on preserving the environment, rather than on mining itself, could be regarded so highly in an innovation category.