DETECTION AND COUNTING OF GRAZING CATTLE FROM AERIAL IMAGES USING CNN
CNN АШИГЛАН АГААРЫН ЗУРГААС БЭЛЧЭЭРИЙН МАЛЫГ ИЛРҮҮЛЭХ, ТООЛОХ
DOI : https://doi.org/10.35633/inmateh-75-39
Authors
Abstract
This study explores the use of deep neural networks for detecting and quantifying the cattle population in Mongolia using drone imagery, addressing the limitations of traditional methods that are labor-intensive and time-consuming. A custom dataset of aerial images featuring grazing cattle in Mongolia was developed, focusing on winter and spring seasons, to train and validate a model based on state-of-the-art object detection algorithms. Specifically, the You Only Look Once (YOLOv8) architecture was employed to detect cattle across diverse environmental conditions. Model performance was evaluated using widely accepted metrics, including precision, recall, F1 score, and the mean average precision (mAP). The findings demonstrate the effectiveness of the proposed approach, with the YOLOv8 model achieving a mAP of 97.3% at an IoU threshold of 0.5, highlighting its potential for efficient cattle detection and monitoring in Mongolia's unique environmental contexts.
Abstract in English