Абстрактный

Application of successive projections algorithm on spectral monitoring of rice leaves nitrogen contents

Ming-Bo Liu, Yan-Lin Tang, Xiao-Li Li, Jia Lou

Visible-NIR reflective spectrum was used to predict the nitrogen contents of rice leaves. Different preprocessing methods were used in pretreatment of the original spectra. The effective wavelengths were selected by successive projections algorithm (SPA) for original spectra and pretreated spectra.Multiple linear regression (MLR) models and Partial least squares regression (PLS) models were built respectively. SPA could reduce the dimensions of spectralmatrix efficiently. In the models established on SPA effective wavelength,MLR model and PLSmodel based on multiplicative scatter correction (MSC) pretreated spectrum had the best predicting effect with r=0.7943 and RMSE=0.4558. In PLS models established on all wavelengths, the best predicting effect model was that based on MSC pretreated spectrumwith r=0.8470 and RMSE=0.3953.