Ages of giant panda can be accurately predicted using facial images and machine learning

Abstract

To forecast giant panda (Ailuropoda melanoleuca) population dynamics in the wild, it is crucial to comprehend their age distribution. Traditional methods for estimating the age of panda are costly, time-consuming, and inaccurate. Additionally, these methods only forecast an age group rather than a real age, and lack a uniform standard. However, advances in deep learning and computer vision have given rise to fresh approaches to this problem. Classification models can be improved by using ordinal regression, which uses ordinal correlations across ages to reduce the non-stationary nature of aging tasks. In this study, we collected 8002 images from 272 pandas in various environments, whose ages ranged from 0 to 38. We applied a five-fold subject-exclusive (SE) protocol to train seven Convolutional Neural Networks (CNN) based on ordinal regression. Experiments were conducted on the Panda Age Dataset (PAD Full) and the Lite Panda Age Dataset (PAD Lite). The results were very encouraging and achieved a Mean Absolute Error (MAE) of 2.51 and 2.41, respectively. Our findings demonstrate that this new tool can noninvasively predict the age of giant pandas in captivity and the wild. Continued development of computer vision technology will drive progress in ecology and conservation.