How Google’s AI helped Australian researchers with conservation efforts

Editor’s note: Amanda Hodgson is a marine mammal scientist based in Perth, Australia . Her core research objective is to develop innovative methods to research, monitor and conserve marine megafauna and their critical habitats.

Have you ever seen a dugong in real life? These big, friendly-looking vegetarians (affectionately called “sea cows”) can be found in coastal waters around the world from East Africa to Australia. Unfortunately, their coastal habitat is also favored by us for development and fishing activities — leading to dwindling populations over the years.

As a marine mammal scientist, my main research is around monitoring marine megafauna and their critical habits. Most of that research has focused on dugongs. Dugongs are a really good indicator of ecosystem health, meaning if a seagrass meadow has a thriving dugong population, it’s likely doing quite well. And if there aren’t so many dugongs, then we need to look into how we can help the seagrass ecosystem (and the other animals depending on it) get healthier.

What makes dugong monitoring quite difficult is that the best way of counting them is from the air. This means I have spent many hundreds of hours staring out of the window of a small plane spotting and counting animals. In 2007, when I was one of the first in the world to investigate the potential for using drones for fauna surveys. And although the drones were safer and provided better data, we were left with thousands of still images from each survey to manually look through.

At that time, during long hours of processing those drone images, I remember dreaming about a solution that could automate all of our image processing needs: mapping the images, finding the animals in the images and ensuring accuracy. Luckily, the expansion and accessibility of AI in recent years has made this a much more realistic dream.

So we started working with AI experts and geoprocessing experts to create a software that could help us automate that image analysis process: WISDAM. By using Google’s open-source TensorFlow technology, we were able to automate the image processing and save us a tremendous amount of effort. WISDAM allows us to both automatically and manually look through those images and find the animals in there, as well as map and assess the background environment within the images.

WISDAM has allowed people who aren’t aerial survey experts by trade to conduct wildlife surveys. There’s always been a great need to develop research tools that are accessible and affordable in the developing countries that form a large proportion of the dugongs’ range. Researchers, community groups and NGOs around the world are now using WISDAM to monitor dugongs in their local areas — from the Great Barrier Reef to the Seychelles to Saudi Arabia. In certain places like the Seychelles, it’s almost impossible to conduct surveys with piloted planes. Now, for the first time, they’re using drones and WISDAM to locate and count their dugong population.

Since we first developed WISDAM, its capabilities have expanded beyond dugongs and can now support surveys of turtles, whales and other marine animals. Our hope is to keep developing this technology, and eventually expand it to different habitats and parts of the world — so one day, anybody will be able to help monitor wildlife and conservation efforts.