.png){kind=icon}
.png){kind=icon}
Return
Hyperspectral imaging for predicting maize leaf nitrogen content: A preprocessing and feature extraction-based study
.png){kind=icon}
OA
AI
DOI：10.1016/j.atech.2026.102089.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate monitoring of maize leaf nitrogen content (LNC) is a core prerequisite for precise crop nutrition diagnosis and refined smart agricultural management. Hyperspectral technology enables non-destructive LNC inversion, and optimizing spectral preprocessing and feature selection algorithms is critical to improving model accuracy and stability. To meet the demand for precise maize LNC prediction, this study proposed a combined spectral preprocessing method FD-PolyBaseline-SNV, introduced the Scattering Ratio index to quantify noise reduction effects, and developed a Traversal of feature number-Recursive Feature Elimination (TERFE) algorithm to screen high-quality feature subsets. Hyperspectral data and measured LNC of 120 leaves were collected at the maize big trumpet and tasseling stages, respectively. Eight preprocessing methods were compared, and two feature selection strategies (Competitive Adaptive Reweighted Sampling, CARS; TERFE) were combined with PLSR, SVR and RFR to construct a multi-combination inversion model system. Results showed that FD-PolyBaseline-SNV significantly enhanced the Pearson correlation between spectra and LNC, decreasing the Scattering Ratio to 1.34% and 1.32% at the two stages, with better performance than other preprocessing methods. TERFE-SVR screened the optimal feature subset for LNC characterization. The optimal model (FD-PolyBaseline-SNV + TERFE-SVR) achieved R² = 0.88, RMSE = 0.74 mg/g, RPD = 3.68 at the big trumpet stage, and R² = 0.84, RMSE = 0.70 mg/g, RPD = 2.98 at the tasseling stage, remarkably outperforming the CARS-based and raw-spectrum models. This technical system allows efficient and accurate non-destructive inversion of maize LNC. It provides important practical value for precise nutrition diagnosis and intelligent field management in precision agriculture.
Keywords:
Hyperspectral imaging
Preprocessing
Feature extraction
Leaf nitrogen content
Maize
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
5.7
Papers:
2.4K
Citations:
2.5K