Winnipeg news: Artificial intelligence can now pick Beluga Whales, calves included, out of aerial drone images after seeing just 34 labelled photos, and two Manitoba researchers helped build the tool.
By Maplestime News Desk | Winnipeg, Manitoba | November 20, 2026
Sources: Phys.org | Florida Atlantic University via Newswise (Nov. 19, 2026) | Frontiers in Marine Science | Last updated: November 20, 2026
Quick Answer: Researchers from Florida Atlantic University, Fisheries and Oceans Canada in Winnipeg and the University of Manitoba built an AI model that detects Beluga Whales in drone images using only 34 human labelled photos. The model reached a mean average precision of 19.2, a 20% relative improvement over the best supervised model, and it was far better at spotting calves.
Key Takeaways
- The model was trained on a dataset of 7,655 high resolution aerial drone images with 27,118 beluga annotations.
- The images came from a summer 2022 drone survey at Creswell Bay, Nunavut.
- The semi supervised SEMI-DETR model reached mean average precision 19.2 using just 34 labelled images, 1% of the set.
- Calf recall jumped to 27.8% from 1.3% for the supervised baseline.
- Manitoba authors Cortney A. Watt and Mark Gillespie work at the DFO Freshwater Institute in Winnipeg and the University of Manitoba.
How Does the Beluga Whales AI Work?
The team compared five models across labelling levels from 1% to 50%. The winner, a semi supervised model called SEMI-DETR, learns from unlabelled images as well as labelled ones. That is the trick. The Beluga Whales detector does not need a human to label every whale in every frame. Labelling thousands of drone photos by hand is slow, expensive work, and this approach cuts the labelling burden by 5 to 10 times.
With just 34 labelled images, 1% of the dataset, SEMI-DETR reached a mean average precision of 19.2. That is a 20% relative improvement over the best supervised model trained the old fashioned way.
Related: Port of Churchill Expansion 2026 — Owner Says $3B Beats Kinew’s $80B LNG Plan
Why Are Beluga Calves So Hard to Spot?
Calves are the weak point in whale surveys. They are smaller, harder to tell apart from the water, and easy for a model to miss. The supervised baseline found almost none of them, with calf recall of just 1.3%.
The new model changed that. Calf detection improved 24.3% relative to the supervised baseline, and calf recall rose to 27.8%. That matters because calves are the future of the population. A survey that cannot see calves cannot tell you whether the Beluga Whales population is growing or shrinking.

What Does This Mean for Manitoba Research?
The Manitoba connection runs through Cortney A. Watt and Mark Gillespie, of the DFO Freshwater Institute in Winnipeg and the University of Manitoba. Their Beluga Whales work puts Winnipeg at the centre of a conservation tool that could be used far beyond the Arctic, anywhere researchers need to count animals from the air without labelling every frame by hand.
Senior author Hanqi Zhuang put it this way: “Artificial intelligence is opening a new path for conservationists to make sense of the enormous volumes of imagery that modern monitoring technologies can collect.”
One date to keep straight: the peer reviewed paper itself was published on September 4, 2026, in Frontiers in Marine Science. The November 19 news peg is the Florida Atlantic University and Phys.org release that brought the work to a wider audience.
Frequently Asked Questions
How many labelled photos did the AI need to spot Beluga Whales?
Just 34 human labelled photos, 1% of the dataset. The SEMI-DETR model reached a mean average precision of 19.2 with that tiny labelled set.
When was the beluga whale AI research published?
The peer reviewed paper was published on September 4, 2026, in Frontiers in Marine Science. The November 19, 2026 news came from the Florida Atlantic University and Phys.org release.
Who are the Manitoba researchers behind the study?
Cortney A. Watt and Mark Gillespie, of the DFO Freshwater Institute in Winnipeg and the University of Manitoba.
Maplestime will provide updates from beluga whale research as announcements are made.
Have a tip about beluga whale research? Email [email protected]
Have a correction? Email [email protected]
Should Manitoba be investing more in AI tools that protect Arctic wildlife, or is this clever tech chasing a problem that field biologists already handle? Tell us in the comments. And share this with every Canadian who loves the belugas of the North.
Enjoying Maplestime? Support independent Canadian journalism — no paywall, no corporate owners.








