Antelope classification using Deep Neural Networks

Reliable and efficient monitoring of animals in their natural habitat is crucial to inform management and conservation decisions. Automation of species classification is important since manually classifying animals is expensive, monotonous, and time-consuming. Current studies have shown that Deep Neural Networks (DNNs) are effective at animal classification tasks. However, there is little research on distinguishing species like antelopes from each other. In this paper, an implementation of Deep Neural Networks in the classification of antelopes is proposed. Our model achieved a 93% Top-1 accuracy on the dataset, which is higher than most of the models discussed in the literature. This paper shows that DNNs, specifically vision transformers, can accurately classify antelopes. Out-of-distribution modelling using softmax confidence was implemented and an accuracy of 74% after introducing an out-of-distribution sample to the test dataset was achieved. Further work can be done to classify species like the lesser Kudu or Greater Kudu, which is more complex than the work of this study. For the out-of-distribution modelling, methods like using the energy score can also be explored to improve the accuracy.