One of Australia’s largest harbours, Darwin Harbour has five times more water than the Sydney Harbour. High tides which swell more than seven meters obscures visibility. What’s more, Darwin Harbour is home to apex predators like saltwater crocodiles and several species of shark.
These occupational challenges make world for the Department of Primary Industry and Resources for the Northern Territory Government harder. The Department’s goal is to ensure fisheries resources are sustainably managed and developed for future generations. To do so, they must count the number of fishes in the waters to assess the abundance of critical species.
Beat around the Bush
Evidently, this work isn’t a walk in the park.
“If you’re in the water with a crocodile you aren’t taking a calculated risk. You’re going to be a statistic. That’s it. If you’re in the water and he’s there, he wants you and you’re gone,” emphasised Wayne Baldwin, Research Technical Officer, Northern Territory Fisheries.
The scientists have had to find ways around these obstacles. One approach was the use of a baited remote underwater video (BRUV). By immersing the camera underwater, the team was able to visualise the situation in the water without having to put themselves in physical danger. The team had to identify different fish species and track their behaviour through the video.
Even so, the solution was not perfect. Wayne explains, “We’ve had quite a few problems with sharks coming in and taking the baits away. Tawny sharks have learned how to open our baits and suck it all out before we have a chance to collect any video.”
Hungry sharks aside, watching 500 hours of video footage was not ideal either. Given the diversity of fish available and the murkiness of the water made classification work harder.
Overcome the Blues
However, with the marvels of developing technologies, the Northern Territories Fisheries team had no reason to feel blue.
Thanks to computer vision, hours of watching murky footage was a thing of the past.
In a collaboration with Microsoft, they worked on developing computer vision to identify fish in underwater images. The computer was trained to recognise the different species of fish.
But it wasn’t long before they ran into another problem: How would the AI program respond to camouflaging fish, like gold-spotted cod?
Microsoft machine learning engineer, Steve van Bodegraven, said, “We went in and talked to them about how they work and the challenges they face… From that we tried to figure out how we could help. Everything we do is explorative, so we don’t necessarily have solutions out of the box.”
Following a period of three months of analysing images, the scientists were greeted with positive results. The system is now able to identify fifteen different species of fish like black jewfish and golden snapper.
“We threw a few test images of fish it’s never seen before and it’s managed to pull those out and differentiate them from the fish it does know about. Once we had that first positive indication of a fish, we really felt we were onto something. From there it was just a matter of finding the right tools to improve and optimise,” said Dr Shane Penny, Fisheries Research Scientist.
AI helped the team save precious time. With time saved, they could use it to analyse the data it collected and improve the sustainability of the region’s fish stocks.
Another research assistant, Samantha Nowland thinks that the AI system could be a game changer in the marine environment. Others could also use this technology to improve their understanding of aquatic resources and ensure sustainable management.
Within the Northern Territories Fisheries team, the focus is now on trend analysis, management plans and expanding their reach.
Other fisheries departments in Australia have taken an interest in the AI technology too. Some also have ideas to use AI for monitoring other animal species such as Australia’s iconic Kookaburra.
For AI, solutions are as deep as the sea.
Read the original story here.