| Autors: Matov, P. Y., Lazarova, M. K., Filipova-Petrakieva, S. K. Title: Hybrid Model for Wild Animals Detection Based on Camera Traps in Bulgarian Mountains Keywords: Bulgarian mountains; camera traps;hybrid model; wild species Abstract: The disappearance of wildlife species in the wild is a key problem for the normal functioning of the global ecosystem. A hierarchical wild animal detection based on a hybrid model utilizing the YOLOv8n architecture is proposed. The suggested object detection pipeline consists of two stages – a base detection stage and a specialized classification stage. First, a modified YOLOv8n model performs initial detection and coarse classification of wildlife into four primary taxonomic groups. Then, refinement of the classification within each taxonomic group is accomplished in order to identify specific species with higher precision. The experimental evaluation is based on an annotated dataset comprising 1882 camera trap images from the Bulgarian mountains, divided into four classes: carnivores, deer, small mammals, and other. The experimental results of the proposed hybrid hierarchical model demonstrate superior performance on all metrics compared with other state-of-the-art models. References Issue
Copyright IEEE Digital library |
Вид: публикация в международен форум, индексирана в Scopus
