Automated Oil Palm Tree Detection and Counting from Satellite Imagery Using YOLOv8 and Integration into the PreciPalm Platform

Abstract

Accurate estimation of oil palm tree populations is crucial for fertilizer planning, yield prediction, and overall plantation management. In practice, tree census is still often performed manually, which is labor-intensive, costly, and prone to error. This study develops an automated detection and counting system that combines very high-resolution satellite imagery (WorldView-3) with a deep learning approach using the YOLOv8 model. The workflow involved data preprocessing, annotation, augmentation, model training, and validation. The best-performing configuration achieved high accuracy, with mAP@0.5 of 0.995 and an F1-score of 0.989. The trained model was integrated into the PreciPalm platform, enabling block-level census reporting and spatial mapping of each tree. Beyond the technical performance, the system offers practical benefits: plantation managers and smallholders can access more reliable data in a shorter time, supporting timely decision-making in nutrient management and crop care. These findings highlight the role of digital innovation in reducing census costs, improving efficiency, and advancing sustainable precision agriculture in the oil palm sector.