Comparative Analysis of YOLOv8s and Faster R-CNN for High-Resolution UAV RGB Oil Palm Health Detection: Accuracy versus Inference Speed Trade-Off

Abstract

Accurate and rapid detection of oil palm health conditions using UAV imagery is essential for supporting precision agriculture and large-scale plantation monitoring. However, challenges such as overlapping canopies, complex background textures, varying illumination, and severe class imbalance often reduce the reliability of automated detection systems. This study presents a comparative evaluation between YOLOv8s, a one-stage object detector, and Faster R-CNN, a two-stage detector, for identifying healthy and unhealthy oil palm trees using high-resolution UAV RGB imagery. The dataset consists of 2,303 annotated images collected from drone surveys and divided into training (70%), validation (20%), and testing (10%) subsets under a controlled experimental design. Both models were trained and evaluated using identical preprocessing pipelines and annotation formats to ensure fairness in comparison. Performance was assessed using precision, recall, F1-score, mean Average Precision (mAP@50 and mAP@50–95), and inference time. Experimental results show that YOLOv8s achieves superior performance with 0.987 precision, 0.998 recall, 0.977 mAP@50–95, and extremely fast inference speed of 1.1 ms per image. In contrast, Faster R-CNN achieves comparable detection accuracy at 0.981 precision and 0.993 recall but with significantly higher computational cost, reaching 875 ms per image. These findings indicate that YOLOv8s provides an optimal balance between accuracy and efficiency, making it more suitable for real-time UAV-based monitoring systems, while Faster R-CNN is more appropriate for offline and high-precision analytical tasks. The study contributes a standardized benchmarking framework for deep learning-based oil palm health detection and provides practical insights for selecting appropriate models in smart agricultural applications.