Early Detection of BSR Disease in Oil Palm Trees through Hyperspectral Analysis with MLP-Based Algorithm

Abstract

Basal Stem Rot (BSR) disease, caused by Ganoderma boninense, poses a significant threat to Malaysia's oil palm industry, resulting in substantial yield losses. Detecting BSR effectively is crucial for maintaining stable palm oil production. The current method, reliant on manual visual inspection by experienced personnel, proves time-consuming. The combination of unmanned aerial vehicles (UAVs) and machine learning offers a more efficient solution. This study introduces a novel approach to automate BSR detection using UAV imagery, enhancing time efficiency and the detection process. The proposed method involves two key stages: pre-processing hyperspectral image (HSI) and employing an artificial neural network for disease detection. The Multilayer-Perceptron (MLP) model is introduced to learn spectral features across various infection stages. The model is trained using ground truth data collected by trained surveyors. The HSI dataset includes samples from 2 healthy trees, 5 Stage A (mild infection), 5 Stage B (moderate infection), and 3 Stage C (severe infection). Performance evaluation includes support vector machines (SVM), 1D convolutional networks (1D CNN), and multiple vegetation indices such as Normalized Difference Vegetation Index (NDVI), Normalized Difference Red Edge (NDRE), Optimised Soil-Adjusted Vegetation Index (OSAVI), and Merris Terrestrial Chlorophyll Index (MTCI). The MLP model exhibits the highest accuracy at 86.67%, outperforming SVM and 1D CNN (66.67% and 73.33%, respectively). While vegetation indices primarily detect Stage C trees and struggle to differentiate between Healthy, Stage A, and Stage B trees, the MLP model offers a balanced performance with moderate training time and quicker inference time. This demonstrates the model's effectiveness in detecting BSR, even at an early infection stage.