Researchers in China and Singapore are planning the train an AI to identify pandas, the furry ones, not the Python libraries!
A unique challenge
Pandas have few biological markers to make it easy to identify individual animals. They also tend to lead mostly solitary lives in the wild and that makes tracking and monitoring panda populations tricky. It also complicates things for their patient keepers, who need to regulate their feeding and care schedules. Wan Yongqing, a Beijing photographer who has worked on panda photos for over a decade admits that all the pandas look identical to him.
A team from the Chengdu Research Base of Giant Panda Breeding are working with staff from Nanyang Technological University and Sichuan Normal University on the process. Head of the Chengdu Base, Zhang Zhihe, hopes that the project will help with conservation management. There has been some good news, as bear populations have stabilised, and they are now registered as vulnerable rather than endangered. Accurate data on bear numbers has also been collected in a panda census for many years, but that data is not always at a granular enough level. Information on age, sex ratios, their birthrate and mortality may help with conservation decisions, especially when some wild bear groups have fewer than 10 members.
Pandas have massive public appeal. The live panda feed from the Chengdu Research Centre (which I wholly recommend!) has an impressive 68 million followers, almost all from the UK, USA and Japan. The antics and clumsiness of pandas has made them universally loved and a worthy symbol for conservation work globally.
From a business perspective, the work of these teams may also filter down to more accurate image recognition technologies in general, especially as much of the data used to train the machine will be from video content, rather than static images.