The giant panda is one of the world’s most adored animals. Native to the mountain ranges of central and southwest China, the bears with distinctive dark eyes and black and white coats were put on the International Union for Conservation of Nature (IUCN) endangered list in 1990. Their status was upgraded to vulnerable species in 2016, and the WWF estimates there are now about 1,800 pandas living in the wild.
Maintaining healthy species populations naturally begins with breeding, but pandas pose a particular challenge due their short mating season — the window for optimal reproduction success is open only one day per year.
To help find the “right time,” researchers from Sichuan University, Sichuan Key Laboratory of Conservation Biology for Endangered Wildlife and Sichuan Academy of Giant Panda trained a neural network, CGANet (Convolution module, bidirectional Gated Recurrent Unit module, Attention module), to automate the mating success prediction process for pandas based on their vocal sounds.
In the vocal data acquisition stage, researchers collect, trim and normalize vocal sound data of mating pandas. The next step is automatic mating success prediction: researchers extract 43 acoustic features per second from the audio segments and feed this data into the CGANet deep neural network, which learns discriminative acoustic features and uses relevant identified features to predict mating success. Finally, this information is provided to local wildlife conservation workers who can explore options such as artificial insemination once the best mating time is determined.
Experiment results demonstrated the effectiveness of the proposed CGANet in learning more discriminative vocal features for automatic panda mating success prediction. For future study, researchers plan to continue increasing panda vocal sound data and testing the practical effectiveness of the method. They also plan to explore the potential for multi-modal data input, such as acoustic and visual data.
Beyond mating season, AI is also being used in panda facial recognition. The IEEE 2019 paper Giant Panda Face Recognition Using Small Dataset from Nanyang Technological University, Sichuan Normal University and Sichuan Key Laboratory of Conservation Biology for Endangered Wildlife introduced a novel panda facial recognition algorithm that has shown encouraging results in early performance evaluations.
The paper Audio-based automatic mating success prediction of giant pandas is available on arXiv.
Author: Yuqing Li | Editor: Michael Sarazen