Giant panda behaviour recognition using images

Monitoring giant panda (Ailuropoda melanoleuca) behaviour is critical for their conservation and understanding health conditions. Currently, captive usually monitored by caregivers. In previous studies, researchers observed behaviours short time spans over a period. However, both caregivers cannot monitor them 24-h using traditional methods of observation. other words, animal data are difficult to collect long periods prone errors when recorded manually. Some have used wearable devices such as accelerometer ear tags collar-mounted units with global position system (GPS) receiver contactless depth cameras video animals primates American white pelicans. the panda, an icon endangered species conservation, almost completely neglected in these studies. To effectively, fully automated recognition method based on Faster R–CNN two modified ResNet was created. The network able detect bodies faces images. One trained classify into five classes, walking, sitting, resting, climbing, eating recognise whether panda’s eyes mouth were opened or closed. Experiments conducted 10,804 images collected from 218 pandas various environments illumination experimental results very encouraging achieved overall accuracy 90% 84% subtle facial motions. proposed provides effective way captivity.