UAV-based integration of RGB, thermal, and structural features with machine learning for multi-class basal stem rot (BSR) severity detection in oil palm

Basal Stem Rot (BSR) caused by Ganoderma boninense is the foremost threat to global oil palm productivity, yet its early and scalable detection remains profoundly challenging. This study presents an integrated UAV-based framework that combines RGB and thermal imagery with top-view structural palm features - crown area, frond number, and frond angle to classify BSR severity levels (T0–T3) using machine learning. A total of 1278 field-verified oil palm trees were assessed, and the Synthetic Minority Oversampling Technique (SMOTE) was applied to address substantial class imbalance. Vegetation indices (VARI, ExG, GLI), thermal pixel intensities, and canopy structural attributes were extracted and Principal Component Analysis was enabled in Machine Learning before training 30 classification models. Among these, the Ensemble Bagged Trees classifier achieved the most robust and consistent performance, recording 94.20 % accuracy for both validation and testing phases, with high per-class precision up to 98.5 % and recall up to 99.7 %. VARI demonstrated the strongest and most consistent spectral response to disease progression, while ExG and GLI exhibited unstable patterns due to canopy shadowing and radiometric variability. The findings highlight the potential of integrating multisensor UAV data with ensemble learning to develop an accurate, scalable, and cost-efficient BSR severity mapping system, supporting improved surveillance and precision disease management across commercial oil palm plantations.

Text 123048.pdf - Published Version Available under License Creative Commons Attribution Non-commercial. Download (11MB) Official URL or Download Paper: https://www.sciencedirect.com/science/article/pii/...

Actions (login required)

View Item