海角大神

Observing lemurs in the jungles of Madagascar is no easy task.

鈥淲e find the group,鈥� explains Stacey Tecot, an assistant professor at the University of Arizona鈥檚 School of Anthropology, 鈥渁nd then we watch them for a little bit, we get our bearings ... and then we start to collect our data.鈥�

Doing so is an all-day process of recording each individual, more or less continuously. But lemurs typically live in 鈥渢roops鈥� of up to 15 individuals. To get solid data, Dr. Tecot tells 海角大神 in a phone interview, 鈥測ou really have to know that who you're watching is who you think you're watching.鈥�

Biologists usually tell individuals apart using distinctive traits such as scars, or give them tags or radio collars. But in a recent study, scientists tested an easier, higher-tech way to keep track of lemurs: facial-recognition software. 聽

鈥淭his is ,鈥� says Tecot, one of two senior authors of a paper published Friday that describes the study. 鈥淚t's something that can be used during training so everyone can learn who individuals are, but then also make sure that you're being accurate in your identification.鈥�

Using a database of faces also makes it possible to track individuals as they enter and leave a population, 鈥渟o there's implications for long-term research as well.鈥�

Tecot explains that she and the paper鈥檚 other lead author, Rachel Jacobs, both struggled to distinguish lemurs in the course of their research. They began exploring non-invasive means to identify different lemurs, and 鈥渇ace-recognition just kept coming back, because they have such distinct faces, and color patterns.鈥�

After reading about the work of Anil Jain, a computer science professor at Michigan State University whose work focuses on biometrics and face recognition, 鈥渨e realized this is something that was actually possible.鈥�

While facial-recognition programs help Facebook tag photos and police identify suspects, adapting the systems to a different primate proved challenging. 鈥淭he texture characteristics of the human skin is different from that of the lemur skin,鈥� Professor Jain tells the Monitor in a separate interview. 鈥淭here's a lot more facial hair.鈥�

The team taught their program, called LemurFaceID, using a 鈥渢raining database鈥� of 462 photos of 80 different red-bellied lemurs.

Over the course of 100 trials, the system recognized individuals correctly 98.7 percent of the time, with about a 2-percent margin of error. Jain expects its accuracy to improve further as it analyzes more images.

Ultimately, the study鈥檚 authors aim to bring this system to the field. 鈥淩eally the idea is that it could be an app in the mobile phone, and the field worker, instead of [needing]聽a bulky laptop, or a camera, can just point the mobile phone camera towards the lemur, and know whether we聽have seen the lemur before,鈥� Professor Jain tells the Monitor.

He points out that many Android phones already have facial-recognition capability for humans. Bringing LemurFaceID to the platform is 鈥渁 question of resources and time.鈥�

Population counts and observation data are seen as a necessary part of conservation efforts for species ranging from birds to frogs. Lemurs are also in dire need of protection: Twenty-six members of the genus lepilemur are currently on the International Union for the Conservation of Nature鈥檚 鈥淩ed List鈥� of threatened and endangered species. In 2015, the BBC reported that hunting and deforestation within 25 years.

If deployed, LemurFaceID could help reverse this trend. 鈥淭his could be used by all different kinds of people,鈥� says Tecot, 鈥渋ncluding anyone who's tracking wildlife trafficking or something like that.鈥�

But while the animals鈥� plight creates a clear need for better research tools, new technologies present a learning opportunity for scientists in Madagascar, where lemurs make their home.

As Dr. Tecot explains, 鈥渙ur long-term plan is to work with the ecological monitoring team in Madagascar that's run by Malagasy researchers and continue to build the database and develop the software, as well as help train people in how to manipulate and adapt these algorithms for other species.鈥�