Modeling effects of illumination and plant geometry on leaf reflectance spectra in close-range hyperspectral imaging

While Hyperspectral Imaging (HSI) has been successfully applied for remote monitoring of vegetation, its use is still underdeveloped in close range settings, where a higher spatial and temporal resolution is applied to measure functional plant traits. Much more than remotely, leaf reflectance spectra in close range are very sensitive to plant geometry and specific alignment of the imaging system. In particular, the spectrum of each plant pixel heavily depends on its distance and inclination towards the light source and sensor. To deal with these effects, this work studies the influence of illumination and plant geometry on the recorded HSI in a specific indoor setup (PHENOVISION at VIB, Ghent, Belgium). Based on simple optical models, the reflectance spectra are modeled using multivariate linear regression. The obtained model coefficients are then used to correct the spectra. Finally, a commonly applied scatter correction method, the Standard Normal Variate (SNV) transformation is shown to remove the illumination and geometry effects