Equalization of Shooting Conditions Based on Spectral Models for the Needs of Precision Agriculture Using UAVs

Unmanned aerial vehicles (UAVs) are widely used as data sources for monitoring of farm lands. As distinct from satellite imagery, in which satellites often have a sun-synchronous trajectory, UAV data can be characterized by significant variability of shooting conditions, which complicates data analytics. We consider the problem of equalization of the shooting conditions for a hyperspectral image using specific spatial image zones (clues), for which the values obtained under the target conditions are known. It is shown that the affine model of the irradiance incoming to a sensor on the test dataset is more accurate than the linear one. For analytical calculation of the parameters of the affine model, the presence of instability in the spectral regions, in which the images of clue regions have close values, is shown. A regularized numerical method that is free of such a disadvantage is proposed for estimation of the parameters of the affine model. The affine model is used to propose a new equalization method that makes it possible to bring images obtained under original conditions closer to images obtained under target conditions, reducing the error by a factor of 4.6. For the experimental study of the models and the equalization method, we use a specific dataset consisting of the AVIRIS hyperspectral images obtained for a single area under significantly different conditions for illumination.