傅立叶变换中红外光谱估测水稻叶片氮素含量的研究

通过不同氮素水平的水稻田间试验，在分析测定了水稻叶片叶绿素、氮素等农学参数后，采用傅立叶中红外光谱仪测定了水稻孕穗期叶片干样的透射光谱，利用协同偏最小二乘算法（siPLS）分析选取了傅立叶变换红外光谱估测水稻氮素含量的敏感波段及其组合。结果表明，其最优主成分数是9个，最佳估测建模的波段组合分别为1350.89～1586.57, 1587.53～1822.40 和 3709.41～3943.72 cm-1；建立的水稻氮素预测模型的精度较高，交互验证均方根误差（RMSECV）和预测均方根误差（RMSEP）分别为0.1538和0.1933，预测值与化学分析获得的叶片总氮浓度之间的交互相关系数和独立检验相关系数分别为0.9393和0.6649，高于中红外光谱指数NFS和NFSA的预测精度。说明利用傅立叶红外光谱作为水稻氮含量的诊断技术是可能的，值得进一步验证和完善。

Determination of leaf nitrogen in rice using FTIR spectroscopy

The time and quantity of N fertilization is considered as a key technique for high yield and quality in crop pro-duction. Based on the rice field experiment with different N rates, the mid-IR transmittance spectra of the dried and ground leaf samples were determined by Fourier transform infrared (FTIR) spectroscopy, and then the estimation model for leaf nitrogen content was built with the obtained mid-IR transmittance spectra and synergy interval partial least square algorithm (siPLS). The optimal siPLS model was obtained with 100 intervals and 9 PLS components. The best combina-tions of spectral regions selected were 1350.89-1586.57, 1587.53-1822.40 and 3709.41-3943.72 cm-1. The value of RMSECV (root mean square error of cross-validation) is 0. 1538, and correlation coefficient (r) was 0.9393 in calibration set, and the RMSEP (root mean square error of prediction) was 0. 1933 and correlation coefficient (r) was 0.6649 for test data set. Furthermore, compared with spectral indices NFS and NFSA, this modal was more reliable and representative. It was suggested that FTIR spectroscopy may be considered as a diagnosis technology for leaf nitrogen content in rice.