Hyperspectral reflectance imaging combined with carbohydrate metabolism analysis for diagnosis of citrus Huanglongbing in different seasons and cultivars

Huanglongbing (HLB) is a highly destructive disease to citrus that is threatening the global citrus industry. It is a great challenge for HLB disease detection at various stages due to the long asymptomatic period and the similar symptom to nutrient deficient trees. This research was aimed to propose an effective method for HLB detection in different seasons and cultivars based on hyperspectral imaging coupled with carbohydrate metabolism analysis. It was found that sucrose accumulated more steadily than starch, glucose and fructose in infected leaves through the hot and cool seasons, but nutrient (Fe) deficient leaves presented a reverse pattern to HLB infected leaves. Spectral and textural features from optimal wavelengths and principle component images were well linked to the HLB fingerprint. The three-class classification models for healthy, HLB infected (asymptomatic and symptomatic), and nutrient deficient leaves achieved 90.2%, 96.0%, and 92.6% accuracy for the cool season, hot season, and the whole period, respectively, using least squares-support vector machine (LS-SVM) classifier. Additionally, the robustness of classification model was validated by a different citrus cultivar using model transfer strategy with the overall accuracy of 93.5%. These results demonstrated the potential of hyperspectral imaging combined with carbohydrate metabolic analysis for HLB detection in different seasons and cultivars.