Near-Infrared Spectrum Analysis and Pls-Da of Huanglongbing in Citrus, Navel Orange and Pomelo

Abstract

A key part of managing citrus greening disease is detecting and removing infected trees. Hyperspectral and near-infrared detection technologies can meet the requirements of rapid, nondestructive detection of Huanglongbing (HLB) disease and are expected to become the main methods. Vision-near-infrared spectroscopy serves as a fast, non-invasive tool for detecting plant diseases. In this study, the following six types of leaves were identified via fluorescence quantitative PCR: mild, moderate and severe HLB-infected leaves; potassium-deficient leaves; zinc-deficient leaves; and normal leaves. HLB-infected, nutrient-deficient (potassium or zinc deficiency) and healthy leaf samples were collected. After preprocessing the NIR spectral data, three different strategies—direct spectral data splicing, spectral normalization splicing and model averaging—were combined with partial least square discriminant analysis (PLS-DA) and multiple linear regression. A multifeature extraction model and a multiclassifier algorithm based on near-infrared (NIR) spectra were integrated to detect HLB in honey grapefruit. The model achieved over 97% correct identification of HLB for citrus leaves of the same variety as those in the correction set, meeting the requirements for field detection. Models for navel orange and Xishi pomelo demonstrated strong applicability and could thus be used for predicting HLB susceptibility of these three crops, with a correct recognition rate above 79%.

Keywords: Huanglongbing, HLB, partial least square discriminant analysis, PLS-DA, near-infrared spectrum, pomelo