LIHU‘E — Researchers at the University of Hawai‘i at Manoa announced this week they’ve developed a new conservation tool, allowing them to track and identify individual dolphins and whales using facial recognition technology.
While some species’ individuals — like the southern right whale — are actually best identified by their faces, the tech used is actually a modified facial recognition algorithm, allowing it to pinpoint each species’ most variable features.
“Each species in the dataset has a different primary identifying mark,” said Philip Patton, lead researcher and doctoral student at the UH Manoa Hawai‘i Institute of Marine Biology.
“It turns out for most of the species in the dataset, the primary identifying mark is their dorsal fin, the one that you see when they come up for air. And we use the shape of the dorsal fin, the nicks and notches on it, the scarring patterns, pigmentation, etc., to identify those for those species.”
While marine biologists and conservationists have used algorithmic recognition technology for years, the UH Manoa model is unique in that it works on 24 different dolphin and whale species, creating an almost catch-all system.
“It’s tricky building a single model that’s able to deal with these differences and similarities in the recognition problem on all these different species,” Patton said.
“And so, the team that developed this approach incorporated some clever tricks to encourage the model to learn the similarities and differences among the individual recognition across all the different species. And, as a result, you get a little bit of a boost in performance for each individual species — at least we think. But it seems to work slightly better than just relying on the single species to get more data.”
Once the images are added to the dataset and separated by species, researchers work in tandem with the algorithm to sift through the results, providing added accuracy and ensuring as few errors as possible are made.
“(I’ll say to the algorithm), ‘OK, tell me which are the five closest individual animals to this one,’ and then I’ll choose which is the actual match — I’ll get the ID after the fact,” Patton said.
“So in that case, it’s not automatic, but it’s just kind of a time saving thing, especially if you have a catalog with like 5,000 individuals and 100 pictures for each individual. That’s a lot of photos for a human to sift through over time. But if you have a tool that can just tell you, ‘these are the five most likely matches,’ then you can do the sorting yourself and do it really quickly.”
By more easily identifying individual marine animals within a population, Patton hopes this technology can help conservationists better respond to ecological concerns as they arise.
“The first step in any sort of conservation or management plan is just understanding the demographics of the population,” Patton said.
“Is the population growing? Is it shrinking? Is it stable? What’s the survivable probability of an adult in the population? …If you’re a conservation organization and you’re trying to figure out which species you want to prioritize in terms of management actions, it’s good to know which species needs the most help.”
•••
Jackson Healy, reporter, can be reached at 808-647-4966 or jhealy@thegardenisland.com.
