The Early, Rapid, and Non-Destructive Detection of Citrus Huanglongbing (HLB) Based on Microscopic Confocal Raman

Abstract

Citrus huanglongbing (HLB) as a devastating disease seriously affects the advance in agriculture, so early detection or accurate diagnosis is the key to control its spread. This paper reported a method for early, rapid, and non-destructive detection of HLB using microscopic confocal Raman (MCR). The spectra of healthy (HE), HLB-asymptomatic (HA), and HLB-symptomatic (HS) leaves were very different at 730–810 cm−1, 866 cm−1, 942 cm−1, 1082 cm−1, 1250 cm−1, 1455 cm−1, and 1510–1630 cm−1, which could be clearly distinguished mutually. Meanwhile, the spectra of relative compounds inside leaves were connected to further analyze spectral differences. The contents of glucose, sucrose, carotene, and chlorophyll in HA leaves were distinctly decreased, but increased in starch and polyphenols compared with HE leaves. In addition, three types of leaves could be well classified by principal component analysis (PCA) whose cumulative percentage variance (CPV) accounted for about 91.01% (three principal components). Partial least square discriminant analysis (PLS-DA) also demonstrated the good clustering effect with an accuracy of 97.2%. Finally, BP-artificial neural network (BP-ANN) model was utilized to evaluate datasets (75% for training, 25% for testing). The low root mean square errors (RMSE 0.0616) and high squared correlation coefficients (R2 0.9598) values showed the high prediction accuracy and stability of the classification model. These results indicated that MCR had excellent practical values for horticulturists to constantly and early detect HLB, which was conducive to prevent and timely control of the spread of HLB.