Implementation of YOLOv5 for Estimating the Number of Oil Palm (Elaeis guineensis Jacq.) Trees

Abstract

Detection and estimation of oil palm tree populations are essential aspects of plantation management. Conventional methods that rely on manual recording have limitations in terms of efficiency, accuracy, and cost. Therefore, this study implements the YOLOv5 model as a deep learning-based approach to detect oil palm trees from aerial imagery. The dataset used in this study was obtained from aerial photographs of oil palm plantations owned by PT Kerry Sawit Indonesia (KSY) 1 and underwent preprocessing, labeling, and training using Google Colaboratory. The model was evaluated using test data from aerial photographs of the experimental plantation at the Indonesian Oil Palm Research Institute. The evaluation results indicate that the YOLOv5 model achieved an accuracy of 96.85%, precision of 98.48%, recall of 98.32%, and an F1-score of 98.40%. Compared to previous methods, YOLOv5 demonstrates a strong balance between detection speed and high accuracy, making it a more effective solution for large-scale oil palm tree population estimation.