Monitoring of free-range rabbits using aerial thermal imaging

Abstract

Unmanned Aerial Vehicles (UAV) imagery is a mature technology, which has found use in a number of applications in agriculture and environmental sciences. However, its application for monitoring and classification of livestock and wild animals has not yet been developed. This study presents a robust methodology to count wild and free-range rabbits and monitor their population. The aims of this study were to 1) test the capacity of the methodology in counting small nocturnal animals such as rabbits in the field, 2) assess the rabbit's density at different sites and different periods of the year and 3) record the temporal pattern of rabbits’ activity during the night hours, with the overall aim to provide a reliable and accurate tool in management studies. For this purpose, a UAV equipped with a thermal camera was used to perform night flights on the island of Lemnos, scanning selected sites and collecting aerial nadir thermal imagery data of the ground. The derived thermal images were analysed using deep learning techniques towards counting the individual animals in each image and the results were compared with manual counting conducted by a researcher. The results revealed that the deep learning approach for automated counting and rabbit recognition overall achieved comparable results to physical counting, with the final model yielding an F1-score of 0.87. However, there were differences between seasons in the methods’ accuracy. This method could be a helpful tool in assessing populations of small nocturnal animals and other free-range livestock animals.

Keywords

• UAV
• Thermal_images
• Deep_learning
• Animal_population
• Animal_detection
• Free-range_rabbits