On the Use of YOLOv8 for Detection and Classification of Mammals Species in Wildlife Environments in the Ecuadorian Amazon

The preservation of global biodiversity has become a critical issue in recent decades, with significant increases in endangered species due to human activities. The use of Deep Learning models for the automation of animal monitoring plays a fundamental role in species conservation. This study aims to de-velop a robust classifier to detect six mammals of the Ecuadorian Amazon (Alouatta seniculus, Leopardus pardalis, Panthera onca, Puma concolor, Tayassu tajacu, and Tapirus terrestris) using the YOLOv8 computer vision model. A dataset of 11,708 images was collected from the iNaturalist repository, ensuring high-quality data through a rigorous cleaning and annotation process. To achieve a model that maximizes trade-offs between detection speed, accuracy, and computational burden, various versions of YOLOv8 were experimented with. The YOLOv8m model with data augmentation emerged as the best performer, with a 4.5% improvement in accuracy over other models.