A Waterloo Engineering research team from the Vision and Image Processing (VIP) Lab has developed an AI-powered system that counts beluga whales faster and with greater accuracy than traditional methods.
Led by Dr. David Clausi, VIP lab co-director and a systems design engineering professor, the research team partnered with Fisheries and Oceans Canada to put their system to the test.
Aerial surveys are the most popular non-invasive approach for monitoring beluga whale populations. The surveys capture images of breeding and feeding regions and are used to track the populations’ numbers and health.
Visual analysis by the human eye of thousands of large-scale and highly detailed images is labour-intensive, time-consuming and prone to error — the whales in the images are quite small, extending only a few pixels, and are often obscured by imaging artefacts.
Clausi worked with Dr. Linlin Xu, research assistant professor in systems design engineering, and Muhammed Patel, a master’s student, to develop an AI-based algorithm that can identify areas within large-scale images that are likely to contain beluga whales and then highlight those areas in square boxes. This first step reduces the volume of work for further analysis significantly.
“Our system cut out about 95 per cent of the remote sensing images taken during aerial surveys,” Patel says. “In other words, only 5 per cent of the images contained beluga whales. That’s a huge reduction in material for analysis, helping the researchers focus on the photos that matter to their work.”
"Our collaboration with the VIP Lab has allowed us to develop an open-source tool that will help with the study and conservation of not only belugas but many other marine mammal species," said Dr. Marianne Marcoux, research scientist at Fisheries and Oceans Canada.
Read Can AI help save beluga whales? for the full story.