Marine Biologists Are Using Microsoft’s AI To Watch Darwin’s Fish Stocks, And It Can Name Them By Sight

Marine Biologists Are Using Microsoft’s AI To Watch Darwin’s Fish Stocks, And It Can Name Them By Sight
Picture: NT Fisheries

Microsoft has rolled out and AI open source project to monitor and count fish. It’s a collaboration with Azure Services and newly acquired dev platform GitHub. The system also stops marine biologists getting eaten by crocodiles.

Microsoft has deployed artificial intelligence (AI) underwater to watch and record fish in and around Darwin Harbour.

Watching fish stocks in the waters around Darwin has its unique challenges. Namely, this:

Image Picture: Getty Images

So marine biologists aren’t all that keen to spend hours in the water holding a camera. Instead, Department of Primary Industry and Resources scientists do their work via underwater cameras attached to buoys.

Since the cameras are already in place, and identifying and counting fish for hours on end isn’t exactly why you spent all that time and money on a marine science degree, the setup provided a unique chance to run an AI experiment.

See if you can spot the global conservation potential:


That’s an open source AI solution developed at a time when global fish stocks face greater pressure than ever before. Microsoft helped build it and posted it on GitHub, the developer platform it officially bought last month for $US7.5 billion.

The AI platform progressively learns to identify different varieties of fish, and can analyse hours of video in minutes.

Darwin Harbour hosts an estimated 415 known species, but DPIR fisheries scientist, Dr Shane Penny, said that two particular species have been of early interest to the team; golden snapper and black jewfish.

“These are two commercially and recreationally important species in the Northern Territory, but research has proven that they had been overfished around the greater Darwin area,” said Penny.

The UN’s 2018 fisheries and aquaculture report was an alarming one for both the fishing industry and conservationists. Between 1961 and 2016, the average annual increase in global food fish consumption was 3.2%. Fish population growth was just 1.6%.

Consumption of meat from all land-based animals combined climbed 2.8 %, so the switch to fish is on, making it more crucial than ever to watch what we eat.

Watching you, humans
Using Microsoft’s Azure AI services, the NT system was up and running within a month. Six months later, it can now identify a fish in a video with 95 to 99% accuracy.

Part of the management plan for the species it is targeting is to identify and mark out protection zones, with underwater video installed to further monitor numbers.

It’s possible the system can also be used to monitor feral fish in freshwater systems, or even cattle movements across the NT.

DPIR chief information officer, Rowan Dollar, is keen to explore potential regulatory applications, such as keeping an eye on the commercial catch.

“We could look into setting up a camera on a trawler that’s out at sea and doing on-the-fly identification of the catch, so we can start measuring by-catch,” Dollar said.

“We can start being able to identify that in real-time, to help better manage those fisheries.”

Here’s some more information on the project: