Introducing Facial Recognition Application for Seals | Science

Have you at any time seemed at a seal and assumed, Is that the exact same seal I saw yesterday? Perfectly, there could quickly be an app for that dependent on new seal facial recognition technology. Identified as SealNet, this seal encounter-finding program was formulated by a workforce of undergraduate learners from Colgate University in New York.

Using inspiration from other technological know-how adapted for recognizing primates and bears, Krista Ingram, a biologist at Colgate College, led the students in creating program that utilizes deep studying and a convolutional neural community to inform one particular seal face from yet another. SealNet is tailored to detect the harbor seal, a species with a penchant for posing on coasts in haulouts.

The group had to teach their application to identify seal faces. “I give it a photograph, it finds the encounter, [and] clips it to a standard size,” claims Ingram. But then she and her learners would manually detect the nose, the mouth, and the middle of the eyes.

For the job, staff members snapped far more than 2,000 pics of seals around Casco Bay, Maine, for the duration of a two-calendar year interval. They tested the software employing 406 unique seals and located that SealNet could effectively recognize the seals’ faces 85 percent of the time. The group has considering that expanded its databases to incorporate close to 1,500 seal faces. As the number of seals logged in the databases goes up, so too really should the accuracy of the identification, Ingram says.

As with all tech, on the other hand, SealNet is not infallible. The software package observed seal faces in other overall body components, vegetation, and even rocks. In 1 circumstance, Ingram and her pupils did a double get at the uncanny resemblance concerning a rock and a seal facial area. “[The rock] did glimpse like a seal confront,” Ingram suggests. “The darker sections were being about the exact distance as the eyes … so you can comprehend why the program identified a face.” For that reason, she states it’s generally finest to manually verify that seal faces recognized by the software program belong to a genuine seal.

Like a weary seal hauling itself onto a seashore for an involuntary image shoot, the query of why this is all needed raises by itself. Ingram believes SealNet could be a valuable, noninvasive device for researchers.

Of the world’s pinnipeds—a group that includes seals, walruses, and sea lions—harbor seals are viewed as the most extensively dispersed. Nevertheless know-how gaps do exist. Other strategies to track seals, these types of as tagging and aerial monitoring, have their constraints and can be really invasive or costly.

Ingram factors to web site fidelity as an component of seal conduct that SealNet could drop additional mild on. The team’s trials indicated that some harbor seals return to the same haulout sites yr after calendar year. Other seals, on the other hand, these kinds of as two animals the workforce nicknamed Clove and Petal, appeared at two distinctive sites together. Increasing scientists’ knowledge of how seals move all-around could bolster arguments for preserving specific parts, states Anders Galatius, an ecologist at Aarhus University in Denmark who was not associated in the job.

Galatius, who is liable for checking Denmark’s seal populations, says the software package “shows a lot of promise.” If the identification rates are enhanced, it could be paired with another photo identification technique that identifies seals by distinctive markings on their pelage, he suggests.

In the long term, following even more testing, Ingram hopes to create an app dependent on SealNet. The app, she states, could perhaps make it possible for citizen researchers to lead to logging seal faces. The plan could also be tailored for other pinnipeds and probably even for cetaceans.

This short article is from Hakai Journal, an online publication about science and modern society in coastal ecosystems. Go through more stories like this at hakaimagazine.com.

Associated stories from Hakai Journal: