Accuracy Evaluation of Immature Oil Palm Detection Using Object-Based Image Analysis on UAV Imagery

Abstract

Rapid and accurate identification of oil palm health status is crucial for supporting plantation replanting decisions. This study aimed to compare the sensitivity of the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Red Edge (NDRE) in detecting senescence in oil palms using multispectral Unmanned Aerial Vehicle (UAV) imagery. A comparative quantitative approach was employed across two smallholder oil palm plantation blocks in Kamang Baru District, Sijunjung Regency, West Sumatra. Data were obtained through multispectral UAV image acquisition, followed by orthomosaic generation, manual digitization for ground truth, calculation of NDVI and NDRE indices, descriptive statistical analysis, accuracy evaluation using a confusion matrix (Overall Accuracy and Kappa Coefficient), and spatial pattern analysis using Nearest Neighbor Analysis. The results identified 320 individual trees, comprising 195 healthy plants and 125 plants exhibiting senescence. NDVI yielded clearer value separation between classes compared to NDRE and demonstrated higher classification accuracy, with an Overall Accuracy of 81.25% and a Kappa Coefficient of 0.604, whereas NDRE achieved an Overall Accuracy of 69.06% and a Kappa Coefficient of 0.357. Spatial analysis revealed that the senescent plants exhibited a clustered pattern. The novelty of this study lies in the empirical evidence demonstrating that NDVI is more sensitive than NDRE for detecting senescence at the study site, indicating that the effectiveness of vegetation indices is influenced by plant physiological characteristics, environmental conditions, and spectral responses at the observation location. These findings provide a scientific basis for selecting the most appropriate vegetation index to support health monitoring and replanting planning for oil palms using multispectral UAV technology.