Assessing Real-Time Object Detection in Rural Aerial Drone Footage: A Comparative Study of YOLOv5 and YOLOv8 with Transfer Learning

This study presents a comparative analysis of real-time object detection models, You Only Look Once (YOLO) v5 and v8, using drone footage captured in rural agricultural scenarios. The primary focus is to assess these models’ effectiveness in detecting objects within typical rural scenarios characterized by agricultural elements such as crops, livestock, and structures. The study employs transfer learning techniques with pre-trained models to adapt YOLOv5 and YOLOv8 for rural imagery. YOLOv8’s training time is 25 minutes, and its mean average precision (mAP) is 0.767. On the other hand, YOLOv5 has a training time of 20 minutes and an mAP of 0.735. Despite the slightly longer training time, YOLOv8x demonstrated superior inference speed, with an average total detection time of 16.5 ms compared to 17.6 ms for YOLOv5x. Additionally, this research highlights the effectiveness of transfer learning in achieving satisfactory results with minimal computational resources.