Ganoderma disease detection in oil palm plantations using C and L band SAR backscatter time series

Abstract

Ganoderma disease poses a significant threat to the oil palm industry in Malaysia and Indonesia, resulting in substantial economic losses. This study aims to utilize a time series of C-band and L-band backscatter information to identify Ganoderma-infected oil palm plants and compare the results within a designated region of Malaysia. The analysis incorporates 30 backscatter coefficients, backscatter cross ratio, the coefficient of variation (CV) from 2019 to 2022, and second-order texture matrices using Gray Level Covariance Matrix (GLCM) as input features. Three classification algorithms were tested: random forest, XGBoost, and quadratic discriminant analysis (QDA) for both L-band and C-band backscatter coefficients separately. Results indicate that L-band backscatter information from ALOS-2 PALSAR-2 outperforms C-band backscatter information from Sentinel-1, with QDA achieving the best performance, resulting in an F1-score of 0.86 for infected trees and 0.88 for healthy trees.