Non-Destructive Detection of Nutritional Elements in Fresh Tea Leaves Using Hyperspectral Technology Combined with a Multi-Stage Feature Selection Strategy
Experimental study demonstrates non-destructive prediction of four nutrients in fresh tea leaves via hyperspectral modeling, indicating robust potential for real-time field monitoring.
Key Points
To establish a quantitative, non-destructive hyperspectral prediction framework for nitrogen, phosphorus, potassium, and carbon in fresh tea leaves while eliminating spectral interference caused by moisture.
Combined Savitzky-Golay (SG) smoothing with External Parameter Orthogonalization (EPO) to suppress water-related spectral interference.
Implemented multi-stage feature selection pipelines (SG-EPO-VCPA-IRIV-SVM_RFE and SG-EPO-BOSS-SVM_RFE) coupled with XGBoost, backpropagation neural networks, and support vector regression.
Validated the models using independent validation sets (N=20), interpreted key spectral features with Shapley Additive Explanations, and tracked nutrient changes under exogenous GABA treatments.
Models achieved high predictive accuracy across all nutrients: R² was 0.896 for nitrogen, 0.954 for phosphorus, 0.913 for potassium, and 0.928 for carbon.
Prediction error remained low, with root mean square error of prediction (RMSEP) values of 0.062 for N, 0.075 for P, 0.438 for K, and 0.157 for C.
Testing on 20 independent field samples confirmed the model's reliability for rapid in situ nutrient monitoring following exogenous GABA applications.