Study of the differences among hyperspectral estimation models under different measurement light source conditions

Abstract

ABSTRACT In hyperspectral remote sensing measurements, the spectral effective information extracted under different measurement light source conditions will produce differences, which in turn affects the robustness of the constructed models. Therefore, it is important to study the differences between hyperspectral estimation models under different measurement light source conditions to construct a general hyperspectral model of vegetation parameters under different light source conditions. In this paper, hyperspectral models are constructed from artificial and natural light sources, and the hyperspectral reflectance and the corresponding SPAD values of artificial and natural light sources under winter wheat, corn and tea tree were measured under the same time period and the same light measurement angle and height. The collected data were used to construct models for the four vegetation indices with better results after screening and to explore how to reduce model variability under different measured light source conditions. The results show that the difference between the artificial light source and natural light source models is significant (the coefficient of determination R2 is significantly different between the artificial light source and natural light source models), and the influence of different light sources on the construction of hyperspectral estimation models can be reduced by selecting vegetation indices. The linear regression equation model constructed by NDVI can effectively reduce the influence of different light sources on the construction of SPAD hyperspectral models and make the model more general. This study can provide new ideas and methods to investigate the differences in spectral models under different measurement light sources.