Given that they don't have fingerprints, how do you tell one great white shark from another? The answer lies in the trailing edge of the dorsal fin, which has a pattern that's unique to each animal. This still means, however, that biologists have to manually compare fin photos in order to see which shark is which. Wouldn't it be easier if there was fin-pattern-matching software, like there is for fingerprints? Well, it has been attempted before, and now there's a new effort underway.
The project began with Dr. Sara Andreotti, a marine biologist with South Africa's Stellenbosch University. She had built up a database of over 5,000 fin photos, which she used to differentiate between great whites that she had observed along the South African coast. As you can imagine, though, manually comparing any one pattern to the thousands of others was becoming a very tiresome task.
With that in mind, she turned to machine learning specialist Prof. Ben Herbst and software engineer Dr. Pieter Holtzhausen. Working with her, they developed a program called Identifin.
Users start by shooting a fin photo of a shark that they wish to identify, loading it into the program, then indicating the top and bottom ends of the fin's trailing edge within the image – this helps the software to pick it out against the visually-chaotic background of the water's surface. From there, Identifin traces the profile of that edge, then compares that outline with all the other edge patterns in its database. Results are displayed in order of match probability.
Although the software can still use some refinements, Andreotti is already finding it very useful for collecting tissue samples in the field.
"Previously, while at sea, I had to try and memorize which shark is which, to prevent sampling the same individual more than once," she says. "Now Identifin can take over. I will only need to download the new photographic identifications from my camera onto a small field laptop and run the software to see if the sharks currently around the boat have been sampled or not."
A paper on the research was recently published in the journal Marine Biodiversity.
Source: Stellenbosch University
And yes, it is a huge improvement in efficiency compared to manual comparisons of literally thousands of pictures of hundreds of individuals (been there, done that, foisted it off on undergrads...)!
The great-granddddy of them all is the work by Mizroch on humpback whale fluke patterns, which goes back to 1990.