Assessing the Transferability of Models for Predicting Foliar Nutrient Concentrations Across Maize Cultivars

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(4), P. 652 - 652

Published: Feb. 14, 2025

Fresh sweet and waxy maize (Zea mays) are valuable specialty crops in southern China. Hyperspectral remote sensing offers a powerful tool for detecting foliar nutrients non-destructively. This study aims to investigate the capability of leaf spectroscopy (SVC HR-1024i spectrometer, wavelength range: 400–2500 nm) retrieve nutrients. Specifically, we (1) explored effects nitrogen application rates (0, 150, 225, 300, 450 kg·N·ha−1), cultivars (GLT-27 TGN-932), growth stages (third (vegetation V3), stem elongation stage V6), silking (reproductive R2), milk R3)) on (nitrogen, phosphorus, carbon) spectra; (2) evaluated transferability regression physical models retrieving across cultivars. We found that PLSR (partial least squares regression), SVR (support vector machine RFR (random forest regression) model accuracies were fair within specific cultivar, with highest R2 0.60 lowest NRMSE (normalized RMSE = RMSE/(Max − Min)) 17% nitrogen, 0.19 21% phosphorous, 0.45 19% carbon. However, when these cultivar-specific used predict cultivars, lower higher values observed. For model, which does not rely dataset, chlorophyll-a -b (Cab), carotenoid (Cxc), equivalent water thickness (EWT) 0.76 15%, 0.67 34%, 0.47 21%, respectively. prediction accuracy expressed as protein PROSPECT-PRO, was lower, an 0.22 27%, comparable models. The primary reasons this limited attributed insufficient number samples lack strong absorption features nm range confounding other biochemicals features. Future efforts needed mechanisms underlying hyperspectral incorporate transfer learning techniques into nutrient

Language: Английский

Prediction of the Quality of Anxi Tieguanyin Based on Hyperspectral Detection Technology

Foods, Journal Year: 2024, Volume and Issue: 13(24), P. 4126 - 4126

Published: Dec. 20, 2024

Anxi Tieguanyin belongs to the oolong tea category and is one of top ten most famous teas in China. In this study, hyperspectral imaging (HSI) technology was combined with chemometric methods achieve rapid determination free amino acid polyphenol contents tea. Here, spectral data samples four quality grades were obtained via visible near-infrared hyperspectroscopy range 400–1000 nm, detected. First derivative (1D), normalization (Nor), Savitzky–Golay (SG) smoothing utilized preprocess original spectrum. The characteristic wavelengths extracted principal component analysis (PCA), competitive adaptive reweighted sampling (CARS), successive projection algorithm (SPA). predicted by back propagation (BP) neural network, partial least squares regression (PLSR), random forest (RF), support vector machine (SVM). results revealed that content clear-flavoured greater than strong-flavoured type, first-grade product second-grade product. 1D preprocessing improved resolution sensitivity spectra. When using CARS, number for acids polyphenols reduced 50 70, respectively. combination CARS conducive improving accuracy late modelling. 1D-CARS-RF model had highest predicting (RP2 = 0.940, RMSEP 0.032, RPD 4.446) 0.938, 0.334, 4.474). use multiple algorithms can be used fast non-destructive prediction

Language: Английский