Evaluating AI for Palm Tree Disease Detection: A Comparative Study of YOLOv8 Object Detection and U-Net Segmentation Using UAV Imagery

Abstract. Agriculture is a cornerstone of economic stability and food security, particularly in regions like Morocco, where palm trees are vital to the environment and local livelihoods. The Figuig oasis, known for its extensive palm plantations, faces significant threats from the spread of palm tree diseases, which can lead to substantial agricultural losses. Early and accurate disease detection is critical to mitigating these impacts. This study evaluates the effectiveness of deep learning models YOLOv8 for object detection and U-Net for segmentation in detecting and segmenting healthy and diseased palm trees using Unmanned Aerial Vehicle (UAV) imagery. A dataset of 400 UAV images was annotated and divided into training (70%), validation (20%), and test (10%) sets. YOLOv8 achieved an accuracy of 78.48%, with a precision of 58.38% and a recall of 47.70%, demonstrating robust object detection capabilities but highlighting the need for improved recall to reduce false negatives. On the other hand, U-Net excelled in segmentation, achieving an overall precision of 0.8746, recall of 0.8713, and F1-score of 0.8727, with powerful performance in delineating diseased regions. The results underscore the complementary strengths of YOLOv8 and U-Net, with YOLOv8 offering efficient detection and U-Net providing detailed segmentation for precise health assessment. This study highlights the potential of integrating UAV imagery and deep learning for automated palm tree health monitoring, paving the way for early disease detection and sustainable agricultural practices. Future work will focus on optimizing model performance, expanding the dataset, and exploring advanced architectures to further enhance accuracy and recall.