Cattle Identification and Detection using Vision Transformers and YOLOv8

Abstract

Modern livestock management need accurate cattle identification, which enable better monitoring, health tracking, and productivity optimization. The cattle detection locate and detect the presence of cattle within an image or video frame. For cattle detection, this research focuses on YOLOv8 model and for the cattle identification the focus is on use of vision transformers (ViT, DeiT, BEiT). We evaluated the performance of the proposed model using the Opencows2020 dataset. Experimental results indicates that ViT outperformed other models in identification tasks and achieve an accuracy of 99.79%. YOLOv8 effectively detected cattle based on coat patterns that shows its suitability for real-time applications. The findings shows the potential of deep learning in improved cattle management systems.