Hyperspectral inversion of leaf nitrogen content in wheat by integrating CWT-SPA feature optimization and XGBoost-SSA model

• A cascaded framework integrating SG, CWT, SPA, and swarm intelligence was developed. • Medium-scale CWT effectively enhanced nitrogen-sensitive spectral features. • SSA-XGBoost achieved high accuracy (R² = 0.79, RMSE ≈ 0.14 mg/g) and strong robustness. • The framework enables reliable and scalable nitrogen monitoring for precision agriculture. Leaf nitrogen content (LNC) is an essential physiological indicator for assessing the growth status of wheat. However, the accuracy and generalization of remote sensing-based monitoring models are often constrained by spatial and temporal variability. To overcome these limitations, this study proposes a cascaded optimization framework that integrates signal enhancement, feature selection, and intelligent optimization algorithms. First, the raw spectral data were preprocessed using Savitzky-Golay (SG) smoothing and a first-order derivative transformation, followed by a multi-scale continuous wavelet transform (CWT). Then, relevant spectral bands were identified through Pearson correlation analysis, and further dimensionality reduction was performed using the successive projections algorithm (SPA). Finally, two regression models were developed: an Extreme Gradient Boosting (XGBoost) model optimized with the Sparrow Search Algorithm (SSA) and an Extreme Learning Machine (ELM) optimized with the Artificial Hummingbird Algorithm (AHA). The XGBoost-SSA model demonstrated superior predictive performance on the test set, achieving a coefficient of determination (R²) of approximately 0.79, a root mean square error (RMSE) of approximately 0.14 mg/g, and a mean absolute percentage error (MAPE) of approximately 9.31%. On an independent external validation set, the XGBoost-SSA model also showed strong generalization capability, maintaining an R 2 of approximately 0.73, an RMSE of approximately 0.14 mg/g, and a MAPE of approximately 9.63%. These findings underscore the value of medium-scale CWT-SPA in spectral feature extraction and highlight the advantages of swarm intelligence algorithms in enhancing regression model performance. Overall, the proposed approach provides a reliable solution for high-precision nitrogen monitoring using hyperspectral remote sensing and supports data-driven applications in smart agriculture.