Accuracy Improvement of Basal Stem Rot Disease Identification in Oil Palm Plantation Using Unmanned Aerial Vehicle and Machine Learning

Abstract

The basal stem rot disease caused by Ganoderma sp remains a majority disease in oil palm plantations, and there is no effective treatment. The technical culture becomes a majority treatment to prolong the oil palm life. The identification and classification accuracy of healthy and infected by BSR (unhealthy) oil palm is needed to support the technical culture of treatment. This study is based on Santoso's previous study (2020), which identifies and classifies healthy and unhealthy oil palms using remote sensing from an image of a multispectral camera with three bands and machine learning. This study aims to improve the interpretation accuracy of healthy and unhealthy oil palms by adding ten vegetation indexes from three bands (red, green, and near-infrared/NIR) of a multispectral camera and applying sixteen models of machine learning. The results showed that the random forest and stochastic gradient boosting had improved 87.18% of the interpretation accuracy by 79.49% in the previous research and 0.69 kappa value from 0.48 in the previous research. This study's accuracy and kappa value improvement may be caused by adding variables from the vegetation indexes that become variable importance besides the red band in the fitting model. The model in this study needs to validate for identifying and classifying healthy and unhealthy oil palm caused by BSR in the area with low and moderate incidence.