Research on the Feasibility of Using Artificial Intelligence in Livestock Census in Mongolia
Abstract
Advances in artificial intelligence (AI) and computer vision are creating new possibilities for automated livestock monitoring and enumeration in the agricultural sector. This study investigates the feasibility of identifying, detecting, and counting horses under free-range grazing conditions on the Mongolian steppe by using the YOLOv8 convolutional neural network architecture. Because horses are widely dispersed across open pastures and exhibit high mobility, traditional census methods are both labor-intensive and time-consuming. This study evaluated the feasibility of detecting and counting horses across varying geographic locations, ambient lighting conditions, and diverse imaging scenarios. Unmanned Aerial Vehicles (UAVs) were deployed to capture aerial imagery of horses across different pasture regions throughout Mongolia. The raw imagery was preprocessed to construct a comprehensive training dataset, after which a deep learning model was developed and evaluated using the YOLOv8 architecture. Model performance was assessed using standard metrics, including precision, recall, an F1-score, and a Mean Average Precision (mAP). Experimental results confirmed that the YOLOv8 architecture provides a strong balance between detection speed and accuracy. The model demonstrated strong performance in detecting horses, achieving precision of 0.948, recall of 0.945, F1-score of 0.97, and an mAP of 0.991. These findings demonstrate the viability of automating the horse enumeration process and suggest potential for further evaluation in Mongolia’s annual livestock census. This study also contributes to the advancement of livestock management and monitoring systems, laying the groundwork for further development of remote sensing solutions in pastoral settings.
© 2026 Bat-Erdene Nyandag, Khishigdalai Ulaankhuu, Unurtseren Baasansuren, Tserendulam Tserengombo, published by Polish Society of Agricultural Engineering
This work is licensed under the Creative Commons Attribution 4.0 License.