Estimating leaf nitrogen concentration based on the combination with fluorescence spectrum and first-derivative

Leaf nitrogen concentration (LNC) is a major indicator in the estimation of the crop growth status which has been diffusely applied in remote sensing. Thus, it is important to accurately obtain LNC by using passive or active technology. Laser-induced fluorescence (LIF) can be applied to monitor LNC in crops through analyzing the changing of fluorescence spectral information. Thus, the performance of fluorescence spectrum (FS) and first-derivative fluorescence spectrum (FDFS) for paddy rice (Yangliangyou 6 and Manly Indica) LNC estimation was discussed, and then the proposed FS+FDFS was used to monitor LNC by multivariate analysis. The results showed that the difference between FS (R2=0.781, SD=0.078) and FDFS R2=0.779, SD=0.097) for LNC estimation by using the artificial neural network (ANN) is not obvious. The proposed FS+FDFS can improved the accuracy of LNC estimation to some extent (R2=0.813, SD=0.051). Then, principal component analysis was used in FS and FDFS, and extracted the main fluorescence characteristics. The results indicated that the proposed FS+FDFS exhibited higher robustness and stability for LNC estimation (R2=0.851, SD=0.032) than that only using FS (R2=0.815, SD=0.059) or FDFS (R2=0.801, SD=0.065).

叶片氮浓度（Leaf nitrogen concentration, LNC）是估算作物生长状况的核心指标，已被广泛应用于遥感领域。因此，借助被动或主动技术精准获取LNC具有重要意义。激光诱导荧光（Laser-induced fluorescence, LIF）可通过分析荧光光谱信息的变化，用于监测作物LNC。本研究探讨了荧光光谱（Fluorescence Spectrum, FS）与一阶导数荧光光谱（First-derivative Fluorescence Spectrum, FDFS）在水稻（扬两优6号与曼利籼稻）LNC估算中的应用性能，并采用所提出的FS+FDFS组合结合多元分析方法开展LNC监测。结果显示，利用人工神经网络（Artificial Neural Network, ANN）开展LNC估算时，FS（决定系数（Coefficient of Determination, R²）=0.781，标准偏差（Standard Deviation, SD）=0.078）与FDFS（决定系数R²=0.779，标准偏差SD=0.097）的性能差异并不显著。所提出的FS+FDFS组合可在一定程度上提升LNC估算精度（决定系数R²=0.813，标准偏差SD=0.051）。随后，本研究对FS与FDFS进行主成分分析（Principal Component Analysis, PCA），提取核心荧光特征。结果表明，相较于仅使用FS（决定系数R²=0.815，标准偏差SD=0.059）或FDFS（决定系数R²=0.801，标准偏差SD=0.065），FS+FDFS组合在LNC估算中展现出更优异的鲁棒性与稳定性（决定系数R²=0.851，标准偏差SD=0.032）。