Facial recognition software system is good for more than just tag photos of your friends on Facebook . The engineering is now being used to distinguish item-by-item whales from above .
Christin Khan , a life scientist who studies North Atlantic ripe giant for theNational Oceanic and Atmospheric Administration , was inspire by a specific particular she often capture in her aerial picture of the animate being . The whale Khan studies have distinctive pale markings on their fountainhead shout callosities , which are areas of raised inspissate skin . The light color comes from the thousands of " white lice"—small crustaceans called cyamids — that call the callosities home . These spots are alike to thumbprints in that their unparalleled patterns can be used to tell apart whales from one another .
Determined to reverse her observation into a pragmatic instrument for inquiry , Khan took her idea to the net . She offer users on the data contender siteKagglea $ 10,000 prize supply by the software companyMathWorksto modernise an precise facial recognition software for right whales . intimately 500 hoi polloi vied for the top prize , and it was finally awarded to the data science companyDeepsense.io , whose package was able to identify the whales with about 87 percentage accuracy . The initial data exercise set was compile of 4500 whale pictures , and the squad says they will need more photographs for amend the engineering .
In addition to its applications in whale research , the software could also have a more direct impact on whale conservation . When a hulk becomes entangle in a net , fishing crews could use it to ID the trapped giant and pucker information about its wellness chronicle to inform a line of activity . According to Khan , it will be another twelvemonth before the research team is inclined to utilize the software in the field . Similar identification technology is presently being pursued for hunchback whales , which can be tell apart by their distinctive fanny flukes rather than the marking on their back .
[ h / t : portion ]
