Oil Palm Health Assessment Using Embedded System on UAVs Images

Tan, Shaw Yang (2025) Oil Palm Health Assessment Using Embedded System on UAVs Images. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Tan Shaw Yang - FULL TEXT.pdf Restricted to Registered users only Download (10MB)

Abstract

This project has achieved its aim to develop a lightweight deep learning model for detecting and classifying oil palm tree health conditions using UAV images, addressing challenges in the oil palm industry such as health classification and resource management, while supporting sustainable farming and aligning with UN Sustainable Development Goals like zero hunger and climate action. The methodology involved training three YOLOv8 variants—YOLOv8n, YOLOv8s, and YOLOv8m—using Python libraries like Ultralytics on a dataset across five health categories (healthy, smallish, mismanaged, yellowish, dead), split into 70% training, 20% validation, and 10% test sets, with models deployed on a Raspberry Pi 5 for practical application. The results showed that YOLOv8n was the optimal model, achieving an average precision of 99.73%, an average recall of 96.61%, an average F1-score of 98.12%, and an average accuracy of 99.81%. It outperformed existing methods such as MOPAD and M-CR UNet. In terms of testing performance, YOLOv8n had the fastest inference time of 139.70 seconds and a detection rate of 131 trees per second. In comparison, YOLOv8m demonstrated a slower inference time of 781.53 seconds due to its larger architecture. These results highlight YOLOv8n as the proposed method for resource-constrained devices, while still ensuring reliable and efficient health assessment for sustainable agricultural practices.