Detection of Huanglongbing Disease in Citrus Using Fluorescence Spectroscopy

Abstract

Huanglongbing (HLB) is an important citrus disease greatly affecting the citrus industry in Florida and other parts of the world. Early disease detection would control the spread of this disease through the application of suitable management measures. This study evaluates the application of fluorescence sensing for HLB detection of citrus leaves. A commercial handheld fluorescence sensor was used to collect yellow, red, and far-red fluorescence at ultraviolet (UV), blue, green, and red excitations from healthy, nutrient-deficient, and HLB-infected leaves of two different sweet orange cultivars, Hamlin and Valencia. Evaluation of the fluorescence sensing was performed under laboratory (controlled) and field conditions. The Nave-Bayes and the bagged decision tree classifiers were trained and tested to assess their performance in classifying the healthy and stressed (nutrient-deficient) leaves. Results revealed that the Nave-Bayes classifier yielded high classification accuracy under laboratory conditions (higher than 85%), while the bagged decision tree classifier yielded high overall classification accuracy under both laboratory and field conditions (higher than 94%). The bagged decision tree classifier performed better than the Nave-Bayes classifier, resulting in higher classification accuracy, although the computation time was at least 10 times greater than that of the Nave-Bayes classifier. In addition, feature extraction using forward feature selection indicated that fluorescence features such as yellow fluorescence (UV excitation) and simple fluorescence ratio (green excitation) contributed toward differentiating healthy leaves from nutrient-deficient and HLB-infected leaves.