Estimation of kiwifruit leaf nitrogen balance index based on hyperspectral and successive projections algorithm
By investigating the relationship between kiwifruit leaf nitrogen balance index(NBI)and hyperspectral reflectance,this study established a suitable remote sensing estimation model to provide a theoretical foundation for guiding precise nitrogen management and growth monitoring of kiwifruit in the Xianyang region.Taking Xuxiang kiwifruit in Yan-gling of Xianyang City,Shaanxi province as the main research object,the hyperspectral reflectance and leaf nitrogen bal-ance index were measured.Through the first derivative,second derivative,continuum removal and standard normal distribution spectral transformation,the relationship be-tween five different spectra including the original spectrum and leaf nitrogen balance index was analyzed.Further-more,through the successive projections algorithm,the redundant information was eliminated and the characteristic wave-lengths were screened out.Based on the characteristic wavelengths of different spectra,single factor regression model,ran-dom forest regression(RF)model,support vector regression(SVR)model and partial least square regression(PLSR)model were used for modeling,and the model accuracy was compared.The results showed that when the NBI value was dif-ferent,the change trend of the related indices of kiwifruit leaves was similar.The reflectivity of the visible light band showed a downward trend with the increase of the NBI value,while the change trend of the reflectivity of the near-infrared band was opposite,showing an upward trend with the increase of the NBI value.Partial spectral transformation increased the number of bands passing the 0.01 level significance test and improved the correlation with NBI values.The number of sensitive bands of continuum removal spectra increased by 190,and the maximum absolute value of the first derivative spec-tral correlation coefficient was 0.77.The successive projections algorithm could minimize the redundancy of data,and the highest dimensionality reduction ratio was as high as 99%.It greatly improved the computational efficiency and the accuracy of the model.Compared with the single-factor regression model,the multi-factor machine learning model had a higher abili-ty to estimate the kiwifruit nitrogen balance index.SNV-SVR performed best,with a coefficient of determination(R2)of 0.82 and a relative percentage difference(RPD)of 2.34.In the future estimation of kiwifruit nitrogen balance index,the model constructed in this study can be given priority.
万方数据
维普
