Detection of Citrus Greening Using Microscopic Imaging

Abstract

Citrus greening reduces fruit production and quality and will likely result in rapid tree decline and death. Because citrus greening symptoms are usually observed on the leaf surface, detection of citrus greening leaf symptoms can significantly aid in scouting for infected trees and managing the disease, thus reducing its spread and minimizing losses for citrus growers. This article presents the microscopic image analysis using color co-occurrence method to differentiate citrus leaves with eight conditions: greening blotchy mottle, green islands, iron deficiency, manganese deficiency, zinc deficiency, young flush leaves and normal mature leaves. Thirty-nine statistical features were extracted from transformed hue (H), saturation (S), and intensity (I) images using the color co-occurrence method for each leaf sample. The number of extracted texture features was reduced by a stepwise discriminant analysis. A discriminant function based on a measure of the generalized squared distance was used for classification. Three classification models were performed using (1) all leaf conditions, (2) all conditions except young flush leaves and (3) all conditions except young flush leaves and blotchy mottle. The three classification models obtained accuracies of 86.67 %, 95.60 % and 97.33 %, respectively. The overall performance was demonstrated in a confusion matrix. The model HSI_14, which used all conditions except young flush and blotchy mottle, resulted in the best accuracy for positive (96.67 %) and negative (97.5 %) symptoms.