水稻叶片氮含量反演偏最小二乘模型设计

Partial Least Square Regression Model for Retrieving Paddy Rice Nitrogen Content with Band Depth Analysis and Genetic Algorithm

针对高光谱偏最小二乘模型(PLSR)反演作物氮含量时易出现数据冗余和模型复杂的问题,尝试结合波段深度分析和遗传算法(GA)建立水稻氮含量PLSR反演模型。基于去包络线处理的水稻高光谱数据(350nm^750nm),选取波段深度(BD)、波段深度比(BDR)、归一化面积波段深度(BNA)和归一化面积波段指数(NBDI)4种波段深度指数分别建立BDA-PLSR模型,进而采用遗传算法波段选择选取最适宜波段深度指数建立GA-PLSR模型,并将GA-PLSR模型与BDA-PLSR模型进行对比。结果显示,基于BNA的GA-PLSR模型在反演水稻氮含量中获得了最佳的结果(Adj.R2=0.67,RMSEP=0.20,RPD=1.84)。研究证明,利用波段深度分析建立的PLSR模型能一定程度上解决数据冗余问题,进一步采用遗传算法进行波段选择能更有效挖掘光谱信息,提高模型精度。

To reduce the data redundancy and complexity of partial least square regression （PLSR）model in retrieving nitrogen content of crops,this article tries to combine band depth analysis （BDA）and genetic algorithm （GA）to build PLSR models for rice nitrogen content retrieval.Based on the continuum-removed spectrum over 350nm^750nm of paddy rice,BDA is employed to derive band depth indexes,including band depth （BD）,band depth ratio （BDR）,normalized band depth index （NBDI）and band depth normalized to area （BNA）,and they are used to build BDA-PLSR models.GA is then utilized to select BDA-derived index most highly correlated with the nitrogen content to build GA-PLSR models,and it is then compared with the BDA-PLSR models.Results show that the nitrogen contents are best estimated by the GA-PLSR model based on BNA （Adj .R2 =0.67, RMSEP=0.20,RPD=1.84）.It is concluded that the combination of BDA and PLSR could reduce the data redundancy,and further selection by GA could explore spectral information effectively and improve the nitrogen content estimation accuracy.