Artificial Intelligence and Unmanned Aerial Vehicles in Digital Livestock Farming: Hair Goat Detection, Tracking and Counting in Mountainous and Rough Terrain Using Deep Learning and Computer Vision

Abstract

The need to increase the production of high-quality animal products due to the rapid increase in global food demand has brought about the need for the use of technology in modern animal husbandry practices. Automatic monitoring and management of animals is of great importance in increasing productivity, especially in small ruminant farming under extensive conditions. At this point, combining high-resolution images obtained from unmanned aerial vehicles (UAV) and deep learning algorithms has the potential to provide effective solutions in remote monitoring of flocks. In this study, it was aimed to automaticaly detecting, tracking and counting hair goats using deep learning algorithms on high-resolution images obtained from UAV. In this context, five different models, namely YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l and YOLOv8x, which are among the most current state-of-the-art You Only Look Once (YOLOv8) architectur, were trained on UAV images obtained from real animal monitoring flights. According to the findings, the YOLOv8s architecture showed the highest performance in terms of both bounding box detection and segmentation performance, with an F1 score of 0.95 and mAP50 value of 0.99. Consequently, it has been revealed that the proposed deep learning-based approach can be an effective, low-cost and sustainable solution for UAV-supported precision livestock farming applications.