Deep learning-based UAV framework for automated morphological and growth analysis of feedlot cattle

Accurate and continuous monitoring of cattle development is essential to optimize feedlot management and determine the most profitable slaughter period. Traditional weighing procedures are labor-intensive, intrusive, and difficult to scale, highlighting the need for automated, non-invasive longitudinal monitoring frameworks. In this study, we propose a UAV-based computer vision framework that integrates high-resolution aerial imagery with deep learning techniques to track the morphological development of beef cattle throughout the production cycle, rather than directly estimating body weight. A subset of images was annotated to fine-tune YOLOv11s for cattle detection, while instance-level body contours were extracted using SAM 2.1 without additional training. An automated filtering strategy retained only well-posed standing animals, enabling consistent extraction of morphological features such as body length, width, and Width-to-Length ratio (W/L), a scale-invariant proxy of body condition over time. Longitudinal, population-level analysis of Nellore cattle monitored over 112 days revealed a clear sigmoidal growth pattern, well described by a logistic model, and the growth curve uncovered three biological phases: lag, rapid development, and plateau, allowing the framework to qualitatively determine an appropriate economic marketing window. These results demonstrate the practical value of the proposed framework for scalable, data-driven precision livestock management in commercial feedlot environment.