Adelie penguin counting using very-high-resolution UAV images and deep learning-based object detection technique

Abstract

Climate changes in polar regions affect to environment that can be directly linked to the living conditions of mammals. Adelie penguin (Pygoscelis adeliae) is known as a species reflecting environmental changes such as sea ice condition and food supply in Antarctica, therefore monitoring the Adelie penguin is crucial to investigate the effects from environmental changes. In this study, we propose a deep learning-based object detection technique using very-high-resolution (VHR) images acquired from unmanned aerial vehicle (UAV) for counting individual Adelie penguin. The VHR images with a spatial resolution of less than 1 cm were acquires in Cape Hallett around Ross Sea, Antarctica using commercial UAV. The image acquisition using UAV has merits from shorter operation duration over large area than field investigation by researcher, preventing disturbance to penguins. Penguin counter software was developed using Google’s tensorflow object detection application programming interface (API), an open source framework, with an image segregation-aggregation approach. This automated method can be applied to other Adelie penguin colonies around Ross Sea and other species of penguins in Antarctica with additional training and testing procedures.