Calibration of hyperspectral close-range pushbroom cameras for plant phenotyping

Abstract

Hyperspectral sensors are able to detect biological processes of plants which are invisible to the naked eye. Close-range cameras in particular support the identification of biotic and abiotic stress reactions at an early stage. Up to now, their full potential is only partially realized because geometrical factors as leaf angle, curvature and self-shading, overlay the signal of biological processes. Suitable 3D plant models constitutes an important step to removing these factors from the data. The matching of these 3D model and the hyperspectral image with sufficient accuracy even for small leaf veins is required but relies on an adequate geometric calibration of hyperspectral cameras.We present a method for the geometric calibration of hyperspectral pushbroom cameras in the close-range, which enables reliable and reproducible results at sub-pixel scale. This approach extends the linear pushbroom camera by the ability to model non-linear fractions. Accuracy and reproducibility of the method is validated using a hyperspectral senor system with two line cameras observing the reflected radiation in the spectral range from 400 to 2500nm. We point out new potentials arising from with the proposed camera calibration, e.g. hyperspectral 3D plant models, which have high potential for crop plant phenotyping.