Get reliable rangeland science

Livestock excretions play a crucial role in nutrient cycling within pasture ecosystems. However, traditional field observation methods require significant human effort and time. In this study, we developed the Dung Detector (DD) model, which utilizes unmanned aerial vehicle (UAV) images and the You Only Look Once (YOLO) v 5 object detection approach, to identify cattle dung in pastures. We have also evaluated the spatial distribution of cattle dung pats in these pastures. The DD model consists of five paddocks, namely Obihiro (OBH), Shintoku (STK), Minokamo (MNO), Miyota (MYT), and Yatsugatake (YGK). A custom dataset containing 1,504 images segmented from UA V orthomosaic images was used for training. The accuracy of the DD model was assessed by comparing it with ground truth data obtained from 2-3 quadrats (10 m × 10 m) in each paddock. Accuracy (F-score) of the DD model in each plot ranged from 0. 432 to 0.861, with better results observed in paddocks characterized by simpler grass species and lower surface gr ass height (SSH). The spatial distribution of cattle dung pats detected by the DD model showed a heterogeneous distribution pattern within the plots due to differences in where grazing livestock stayed due to fences, shaded forests, and water troughs.