傅里叶变换中红外光谱谱区宽, 搜索空间大, 需要采用高效率和高质量的算法进行波长选择.敏感波段及其 组合的选择是简化分析模型和提高模型预测精度的关键技术之一.本研究以水稻孕穗期叶片干样的中红外光谱透 射率和叶片氮素含量为数据源, 通过协同偏最小二乘算法(siPLS)从宽谱区中初选出波段范围1583.3～992.2 cm -1 , 再采用迭代遗传算法(GA)从中选出了84个水稻叶片氮素含量预测的敏感波段.研究结果显示以此敏感波 段建立的偏最小二乘回归模型的预测均方根误差(RMSEP)和水稻叶片总氮含量的测量值与预测值之间的相关系 数分别为0.1186和0.9120, 该预测结果明显优于协同偏最小二乘法(siPLS)和光谱指数NFSA的预测结果, 说明傅 里叶变换红外光谱技术结合siPLS-GA-PLS算法能够实现水稻叶片氮素含量的预测

The Fourier transform mid-infrared (FT-MIR) spectra and nitrogen concentration were measured from dried leaves of rice in booting stage under various treatments of nitrogen fertilization. The synergy interval partial least square algorithm (si-PLS) combined with genetic algorithm (GA) was applied to the selection of optimal variables. 84 variables were determined to build the estimation model for leaf nitrogen content. The prediction accuracy of the PLS regression model based on these variables was better than that of the siPLS model or the model from spectral index NFSA. The RMSEP (root mean square error of prediction) is 0.1186 and r (correlation coefficient) is 0.9120. The results show that FTIR spectra combined with siPLS-GA-PLS algorithm can be used successfully to analyze leaf nitrogen concentration of rice.