选取典型芦苇湿地基于芦苇叶片实测高光谱数据和叶面积指数(Leaf Area Index,LAI),在原始光谱的基础上进行了平滑(R)、一阶微分(FD)、倒数(RT)、对数(LT)、倒数一阶微分(RTFD)、对数一阶微分(LTFD)等六种光谱变换,利用竞争性自适应重加权算法(CARS)对不同变换下芦苇 LAI 特征光谱波段予以筛选,进而用筛选的特征波段采用逐波段组合法(BCI)构建芦苇LAI敏感光谱指数,利用随机森林(RF)、极端梯度提升(XGBoost)以及支持向量机(SVM)回归算法,构建芦苇 LAI 的高光谱估算模型.结果表明,采用CARS算法筛选不同变换光谱的特征波段构建模型,发现经过FD变换(R2=0.417,RMSE=0.905)的模型效果最优.在CARS基础上使用筛选过后的特征波段构建植被指数进行建模比较,模型效果最好的是XGBoost(R2=0.620,RMSE=0.826).
Hyperspectral Remote Sensing Estimation of LAI Reed Based on Characteristic Band Selection
Based on the measured hyperspectral data of reed leaves and leaf area index,six spectral transformations were carried out on the basis of the original spectrum,including smoothing(R),first-order differential(FD),reciprocal(RT),logarithmic(LT),reciprocal first-order differential(RTFD)and logarithmic first-order differential(LTFD),and the competitive adaptive reweighting algorithm(CARS)was used to screen the characteristic spectral bands of LAI under different transformations.Then,the band-by-band combination method(BCI)was used to construct the LAI sensitive spectral index of A.reed,and the hyperspectral estimation model of A.reed LAI was constructed by random forest(RF),extreme gradient boosting(XGBoost)and support vector machine(SVM)regression algorithms.The results show that the CARS algorithm is used to screen the characteristic bands of different transformation spectra to construct the model,and it is found that the model with FD transform(R2=0.417,RMSE=0.905)has the best effect.On the basis of CARS,the vegetation index was constructed using the screened feature bands for modeling comparison,and the model effect was XGBoost(R2=0.620,RMSE=0.826).
spectral transformationleaf area indexband by band combinationscompetitive adaptive reweighted sampling
万方数据
维普
