Establishment of Rape Leaf Moisture Content Spectral Character Models Based on RSR-PCA Method

摘要

It was developed that the method of spectral analysis was used to quantitatively analyze the rape moisture content. The method of region stepwise regression (RSR) was proposed to select the characteristic wavelengths for rape leaf moisture content prediction. The spectrum curve was segmented into several regions by the middle points of adjacent zeros of derivative spectrum data. Each region included a spectral absorption peak or an absorption valley. Stepwise regression was applied to each region, where the correlation coefficient and root mean square error (RMSE) was taken as the evaluation standard to select the spectral characteristic wavelength regions for the model in each region. In order to avoid wrongly choosing characteristic wavelengths or neglecting the necessary information, applied further choice to the selected characteristic wavelengths according to the former research findings of our team and regularities of molecular spectrum absorption band distribution. The method of principal component regression analysis (PCA) was used to establish the model between the moisture content and the characteristic wavelengths of rape leaf. The method could diminish runtime and overcome the effect of multiple co-linearity while enhance model prediction precision. From the spectral date of rape leaves under different water stress conditions, it was found that the rape leaf moisture content had a significant correlation with the spectral reflectance at 460nm, 510nm, 1450nm, 1650nm, 1900nm and derivative of spectral reflectance at 702nm. The correlation coefficient between the estimated value and the real value is 0.92; the root mean square error is 0.37.