Machine Learning Estimation of Large Herbivore Populations Using Drone Imagery

Abstract

Introduction: Large herbivores’ population surveillance is important for biodiversity protection, wildlife conservation, and proper planning of habitats. Traditional methods of population estimation generally involve significant time and effort in the field, while also requiring trained personnel; moreover, thick vegetation, large territories, and animal movements could decrease the efficiency of these methods. Aerial images taken from drones represent an alternative, noninvasive, and flexible approach to gathering information about wildlife. However, due to the significant amount of data gathered during drone surveys, there is a necessity of developing automatic techniques for animal detection and population estimation. Materials & Methods: In this work, we propose a technique based on machine learning for large herbivores’ population estimation from drone imagery. Machine learning and deep learning object detection models are employed in automatic identification and counting of the large herbivores from aerial images. The performance of the model is analyzed using detection accuracy, precision, recall, F1-Score, and counting error. Results: The suggested algorithm shows the possibility of using machine learning for automated detection and counting of the large herbivores from the drone aerial images, thereby minimizing the need for laborious and time-consuming manual surveying. The high resolution of the aerial images makes the animal detection much clearer. Automated detection allows for fast processing of large volumes of image data. Conclusion: The combination of the aerial images with the automated animal detection makes the drone-based monitoring useful for estimating the population of large herbivores. It can be used for wildlife censusing, population monitoring, habitat management, and conservation planning