Raman spectroscopy coupled with interpretable machine learning for early detection of citrus Huanglongbing

Citrus Huanglongbing (HLB), caused by Candidatus Liberibacter asiaticus, threatens global citrus production, while early diagnosis is hindered by a prolonged asymptomatic period and laboratory-intensive methods. This study establishes a non-destructive framework integrating portable Raman spectroscopy, reference compound validation, optimized machine learning, and model interpretability for early HLB detection. A total of 1025 Thai pomelo leaves yielding 3075 spectra were classified into four categories: healthy…

By the Truvace deskOctober 20, 20261 min read1 sourcePeer-reviewedTRV-2026-2052· cite & verify

In brief

Researchers collected 3075 Raman spectra from 1025 Thai pomelo leaves labeled as healthy, early Huanglongbing, symptomatic Huanglongbing, or nutrient deficient. They tested 72 preprocessing workflows across six classifiers and tuned models with four search strategies, and the best SVM reached roughly 96% accuracy.

Early diagnosis of citrus Huanglongbing is hard because infected trees stay asymptomatic for a long time and lab methods are labor-intensive. Linking model decisions to molecular signals makes the approach more biologically grounded, but the source only describes potential for field deployment, not completed field use.

Main points

The dataset covered 1025 Thai pomelo leaves and 3075 spectra in four classes, including nutrient deficiency as a confounder.
Unsupervised UMAP plus K-means clustering reached 91.72% ± 0.96% purity, indicating the classes were spectrally separable.
The SVM scored an F1 of 0.9597 ± 0.0146 after 72 preprocessing workflows and four hyperparameter search strategies.
SHAP analysis identified a cellulose-associated feature at 1096.8 cm-1 as the leading contributor to early HLB identification.

The gain

An optimized support vector machine classified healthy, early HLB, symptomatic HLB, and nutrient-deficient pomelo leaves from Raman spectra with about 96% accuracy, giving a non-destructive route to early detection.

The rundown

Spectral differences between classes appeared in carotenoid, chlorophyll a, carbohydrate, and polyphenol regions. Measurements of pure reference standards were used to support molecular assignments beyond literature-based interpretation.

SHAP analysis tied classification decisions to these molecular assignments. Pigment-associated features contributed most to overall classification, while a cellulose-associated band was most important for flagging early infection.

The authors present the framework as connecting molecular assignments, classification performance, and model interpretation, and say it has potential for field deployment.