Automated individual identification of wildlife using deep neural networks

Abstract

Automated re-identification of individuals in endangered species has gained traction in nature conservation initiatives. Scholars in computer vision community have explored the use of the deep convolutional networks in the re-identification and classification of chimpanzees and gorillas from image signals. However, no work reports on the re-identification of lions and cheetahs. We provide an implementation of deep neural networks for individual re-identification of the big cats, lions, and envisage replicating the success obtained in classification and re-identification tasks reported in chimpanzee and human beings using face images. We present a comparison of the performance obtained from different deep neural network architectures on individual lion re-identification using face image features, and also search for the best loss function between pair-based loss function and class aware loss functions. This endeavor is aimed at assisting conservation initiatives to monitor, report and manage biodiversity on the population of lions with minimal manual labor.