Early diagnosis of citrus Huanglongbing by Raman spectroscopy and machine learning

Abstract Timely diagnosis of citrus Huanglongbing (HLB) is fundamental to suppressing disease spread and reducing economic losses. This paper explores the combination of Raman spectroscopy and machine learning for on-site, accurate and early diagnosis of citrus HLB. The tissue lesion characteristics of citrus leaves at different stages of HLB infection was explored by Raman spectroscopy, and a scientific spectral acquisition strategy was proposed. Combined with machine learning for feature extraction, modeling learning, and predictive analysis, the diagnostic accuracies of principal component analysis (PCA)-Partial least-square and PCA-support vector machine models for the prediction set were 94.07% and 95.56%, respectively. Compared with conventional random detection method, the detection strategy proposed in this paper shows higher accuracy, especially in early HLB diagnosis with significant advantages.