<![CDATA[Oil palm (Elaeis guineensis) is a strategic commodity for Indonesia’s economy, however, tree inventory processes in plantation areas are still predominantly manual, requiring considerable time and cost, and posing a high risk of human error. This study analyzes the effect of hyperparameter variations on the performance of the YOLOv11 algorithm for automated oil palm tree detection using UAV imagery. Four key hyperparameters batch size (16 and 32), number of epochs (100 and 150), learning rate (0.01 and 0.001), and optimizer (SGD and AdamW) were evaluated, resulting in 16 training configurations. The dataset, obtained from Roboflow, underwent annotation, augmentation, and preprocessing prior to model training. Model performance was assessed using precision, recall, and mean Average Precision (mAP), followed by additional evaluation at varying confidence and Intersection over Union (IoU) thresholds. Experimental results show that the optimal configuration batch size 16, 100 epochs, a learning rate of 0.001, and the SGD optimizer achieved an mAP50 of 98.3%, with precision and recall values of 95.3% and 94.1%, respectively. The model also demonstrated stable detection performance at a confidence threshold of 0.5 and an IoU threshold of 0.5. These findings highlight the significant effect of hyperparameter tuning on YOLOv11 detection performance and offer insights for enhancing automated tree-counting systems in the plantation sector, enabling more efficient and accurate operational workflows.]]>
Copyrights © 2025
