The technique uses artificial intelligence to analyze shark dorsal fins, which are unique to each shark. To train the system, researchers Ben Hughes and Tilo Burghardt used a data set of 240 photographs. While that might sound like a relatively small amount of data, it turned out to work with an accuracy of 81 percent.
This is not the first time such technology has been applied to tracking marine animals. A few years ago, researchers at Eckerd College in St. Petersburg, Florida used computer vision and signal processing techniques to identify dolphins based on their fin outlines.
Unlike that project, Hughes and Burghardt’s work chose to replace fin outlines as their main metric with the unique contours of a particular segment of a shark’s dorsal fin.
“The idea behind using portions of the fin contour is to make individual recognition robust,” Hughes told Digital Trends. “That robustness allows for local changes in fin shape, for example due to damage over time, as well as waterline occlusions — which means when part of the fin can’t be seen because it’s below the waterline. From a technical perspective it’s also robust to fin contour detection errors, that can occur as a result of automatically extracting fin contours from images.”
In the past, researchers tracking sharks have been able to uncover some extraordinary behavior. For instance, in 2005 researchers at the White Shark Trust discovered that a great white shark named Nicole swam from South Africa to Australia and back again over a nine-month period. The ability to use AI to better monitor sharks will hopefully reveal more of this kind of fascinating behavior.
“The objective of this finprinting identification system is to create an online international database that will be accessible to white shark scientists around the world at first, and then, in a second stage, open it up to non-scientists,” Michael Scholl, CEO of the Save Our Seas Foundation, told Digital Trends. “It has become impossible to manually and visually manage databases [for the identification of sharks], which include hundreds of thousands of images, and thousands of individuals, without a automated identification system. Technology now allows for very effective identification and database management tools which will make the life of researchers more efficient and effective.”
