China Creates AI-Powered Facial Recognition For Pandas


The giant panda is a bear native to south central China. It is easily recognized by the large, distinctive black patches around its eyes, over the ears, and across its round body.

For humans, their uniform black and white markings makes them fairly anonymous and indistinguishable. But for computer vision, that is a different story.

Researchers in China has developed an AI-powered facial recognition app that can identify specific pandas with ease. Available at the Chengdu Research Base of Giant Panda Breeding in southwest China, the app is meant to help visitors identify any of the facility’s dozens of captive giant pandas, and find out more information about them.

The app’s developers hope the software can also be used by scientists to rack down the endangered species when in the wild.

According to Chen Peng, a researcher who co-authored the paper Giant Panda Face Recognition Using Small Database:

"The app and database will help us gather more precise and well-rounded data on the population, distribution, ages, gender ratio, birth and deaths of wild pandas, who live in deep mountains and are hard to track."

"It will definitely help us improve efficiency and effectiveness in conservation and management of the animals."

The app uses various elements to identify pandas, including ear shape and eye markings
The app uses various elements to identify pandas, including ear shape and eye markings

The facial recognition system uses various elements to identify one panda from another, including the shapes, sizes and markings of the animals' ears and eyes.

And just like other facial recognition systems, the app was built using a huge database of panda pictures. The developers of the app ran 120,000 images and tens of thousands of videos to train the algorithm.

This isn't the first time researchers create facial recognition system for animals. Previously, similar tools have been developed and used in conservation efforts for bears and lemurs, and the technology is also being introduced into farms.

Software that can identify individual pigs, sheep, and cattle has also been used to monitor animals’ health and activity.

But this AI technology for pandas, is meant to also track the feeding schedules of individual pandas, understanding their genealogy, as well as helping human-panda relations.

"You no longer need to worry about making the pandas angry by calling them by the wrong name,” said one of the researchers.

Giant pandas are endangered animals, and their conservation work requires a lot of human resources.

The basic population of wild pandas in China has been basically ascertained. According to a 2013 survey by the National Forestry and Grassland Bureau, there are more than 1,800 wild pandas. And using various efforts, China has organized professional panda analysis to obtain as many information about them.

However, previous studies about pandas in the wild was difficult to clarify.

According to the experts, the population structure of these endangered animals is mainly composed of population density, age composition, sex ratio, birth rate and mortality. The population of wild giant pandas in China rely on their living habitats at deep bamboo forests, making things difficult for scientists to track and monitor them.

"Efficient and accurate research on the population and distribution of wild giant pandas is an urgent and important task," the experts said.

Panda in the wild
Tracking and identifying pandas in the wild require a lot of human resources

The usual method of tracking giant pandas in the wild, include capture and recapture, human visual recognition, and collection of DNA from hair or feces.

For workers who perform these tasks in the field, such operations invade the territory of the giant pandas, and the cost of manual searches is high and can be dangerous.

With AI-powered facial recognition system, scientists can simply install surveillance cameras in areas where giant pandas live, allowing them to track, monitor and identify each panda without having to bother their habitats.

This should help scientists in studying the animal's population structure, and provide scientific support for wildlife conservation management.

“When we were doing the primary sorting of photos, because of the large number of contacts, the naked eye could find some photo collection and classification errors, which deepened our confidence in this research.” said Peng.

"I hope that the relevant technology can be applied to the national giant panda survey in the future."