基于高光谱维数约简与植被指数估算冬小麦叶面积指数的比较

高光谱遥感反演LAI时,由于实际样本数远小于光谱维数,易导致基于全谱段建立的模型不稳定。针对该问题,该文提出将基于原始光谱反射率与LAI相关性和基于光谱曲线特征的2种波段选择方式分别与主成分回归(PCR)或偏最小二乘回归(PLSR)结合的高光谱维数约简方法,估算冬小麦LAI。并选择归一化植被指数(NDVI)、增强型植被指数(EVI)、重归一化植被指数(RDVI)、修正土壤调节植被指数(MSAVI)和三角形植被指数(TVI)5种代表性植被指数,利用2009、2010年实测大田冬小麦冠层高光谱和LAI数据,将提出的基于维数约简的方法与基于植被指数的LAI估算方法进行了比较,独立样本集验证结果和交叉验证结果均表明,提出的基于维数约简的方法比基于植被指数方法的估算精度高,在交叉验证结果中,基于维数约简的方法R2最高达到0.818,相应RMSE为0.685。该研究可为后续基于高光谱的LAI估算提供参考。

Comparison of winter wheat LAI estimation methods based on hyperspectral dimensionality reduction and vegetation index

The actual number of samples is much less than the dimensionality of hyperspectrum. The LAI estimation model based on the whole spectrum is unstable. This paper presented a dimensionality reduction based LAI estimation method which was the combination of two band selection methods and principal component regression (PCR) or partial least squares regression (PLSR). The first band selection method is based on the correlation coefficient between original spectral reflectance and LAI, and the second band selection method is based on characteristic of spectral curve. Five representative vegetation indices including Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Renormalized Difference Vegetation Index (RDVI), Modified Soil-Adjusted Vegetation Index (MSAVI) and Triangular Vegetation Index (TVI), were chosen. Using field measured winter wheat canopy hyperspectral data and LAI in 2009 and 2010, the proposed dimensionality reduction based LAI estimation methods and vegetation index based LAI estimation methods were compared. The experimental results indicated that the overall accuracy of dimensionality reduction based LAI estimation methods were higher than that of vegetation index based LAI estimation methods, generally. The model combing the proposed first band selection method and PLSR got the highest overall estimation accuracy with R2=0.818, RMSE=0.685 in cross-validation experiments. At the same time, the models combing the proposed two band selection methods with PCR respectively also reached high overall LAI estimation accuracy.