基于高光谱的甜菜冠层氮素遥感估算研究

Models of Estimating Sugar Beet Nitrogen Using Hyperspectral

利用野外便携式ASD Qualityspec光谱仪,实测了田间甜菜冠层光谱数据,采用植被指数对氮含量进行预测,发现估算精度较低,分析NDVI与VLOPT与氮含量的相关性,得出氮含量在很小的时候就达到饱和水平。根据4种预处理下的甜菜冠层光谱,分别采用偏最小二乘回归(PLSR)和主成分回归(PCR)建立甜菜氮含量估算模型,比较不同预处理和不同回归方法对估算精度的影响。结果表明:对PLSR来说,一阶导数处理的光谱数据建立的模型精度最好(RMSE=2.34g/kg,RE=19.6%),平滑、MSC和SNV建立的估算模型次之;对PCR来说,平滑处理的光谱数据建立的模型精度最好(RMSE=2.34g/kg,RE=19.4%)。总的看来,不同预处理对估算模型精度有一定的差异,但PLSR和PCR两种回归方法对甜菜氮含量估算模型影响不大。

This paper analyzes the beet canopy spectra under four pretreatment were used partial least squares regression(PLSR)and principal component regression(PCR)to establish beet nitrogen content estimation model,compare different methods of pretreatment and different regression estimation accuracy impact on PLSR,the first order derivative of the spectral data processing model established best accuracy(RMSE=2.34g/kg,RE=19.6%),smoothing,estimation model followed by MSC and SNV established;for PCR toHe said precision spectral data smoothing model established best(RMSE=2.34g/kg,RE=19.4%).Overall,there are some different pre-treatment model to estimate the accuracy of differences,but the two regression PLSR and PCR methods to estimate the nitrogen content of beet little effect model.