Auto ‘finprinting’ identifies individual sharks as they migrate
Article published online in English on the New Scientist website on October 5, 2016
Great white sharks migrate over huge distances, making it tricky to track specific individuals through the seasons. Now, a project hopes to automate their identification from photographs of their fins.
The technique, known as “finprinting”, uses the unique contours of a shark’s dorsal fin as a biometric – rather like a human fingerprint or iris. Researchers have scrutinised fins to identify sharks for years, sometimes using software to help, but the new project is an attempt to make the whole process automatic.
The system, developed by Ben Hughes and Tilo Burghardt at the University of Bristol in the UK, has been trained on 240 photographs of shark fins. It picks out recognisable portions of the fin contour, not just entire fins. This means that images of fins that later become partially damaged might still be useful for identification.
In tests, the software was able to analyse a picture of a shark fin and say, with an accuracy of 81 per cent, whether it belonged to a known individual or not.
The approach should help researchers keep tabs on revealing behaviour in shark migrations. In 2005, for instance, Michael Scholl, then at the White Shark Trust, and his colleagues reported an astonishing finding: a satellite tag on a great white nicknamed Nicole showed that she had travelled from South Africa to Australia and back within nine months.
This was a groundbreaking insight into the wanderlust of the mysterious species – but after that, observations of the shark ran cold. “In November, she left, but we didn’t have time to put a satellite tag on her, which was a big shame,” says Scholl. “She was never spotted again.”
Scholl still receives emails from shark enthusiasts asking what happened to Nicole. So far, he has had to reply that he does not know. The new system might one day spot her.
“It could be a fantastic tool,” says Greg Skomal, a fisheries scientist at the Executive Office of Energy and Environmental Affairs in Boston, Massachusetts.
Skomal is developing a catalogue of the growing population of great whites observed off the east coast of the US, an area popular with divers and windsurfers. His team spends “hundreds, if not thousands” of hours doing manual photo identification each year.
Hughes hopes to have the tool in the hands of researchers such as Skomal by the first quarter of 2017, using photos collected by scientists in the field. The plan is to later feed the AI system photographs contributed by members of the public who go shark watching or cage diving, says Scholl. That way, a far greater number of identifications could be recorded.
Automatic identification has already been tried out on other species by scientists working on Wildbook, the wildlife-data platform where the shark tool will be hosted. A separate Wildbook project recently used automatic identification techniques to analyse 15,000 photographs of Grevy’s zebras in Kenya. In this case, algorithms recognised individuals by their stripe patterns.
Tools to identify zebras, sharks and other animals make it possible to analyse these species’ social networks, too, says Tanya Berger-Wolf, a computer scientist at the University of Illinois, Chicago, and co-founder of the system behind Wildbook. She recently analysed the making and breaking of social connections between individual zebras, for example.
Berger-Wolf says that Hughes and Berghardt’s shark-identification system might need some tweaking before it can be used to identify large numbers of individuals. That would mean training the tool on a bigger set of photographs.
The system leads to the possibility that, one day, someone might upload a snap of Nicole’s fin. Scholl’s own photo database will be used to develop the tool, so that shark might well be found again in the future. For him, that would be a thrilling outcome.
“It’d be one of the happiest days of my life,” he says.View
- Category :
- Date :
Oct . 05 . 2016