Semi-supervised Deep Learning for Estimating Fur Seal Numbers

Abstract

Having estimates of animal species is of growing importance for conservation and ecological reasons, given the increasing concern about the impact of climate change on fauna worldwide. However, it is difficult and sometimes dangerous to count animal numbers in the wild. Counting and detecting animals from drone images can be expected to become a crucial part of conservation policies based on obtaining up-to-date estimates of population numbers. This paper proposes a deep learning approach, the Faster- RCNN algorithm, to count fur seals on the Alejandro Selkirk Island using drone images. Using a semi-supervised approach, the experimental results show the overall precision to be 0.86. This preliminary research shows that machine learning for remote sensing via drone images is helpful for estimating fur seal numbers and could be extended to other areas where it is important to quickly estimate animal populations for the purpose of ecology and conservation.