Illumination normalization using 3D structure on airborne high resolution imagery: An application on Guyanese rainforest using UAV hyperspectral collected data

Organised by Laboratory of Geo-information Science and Remote Sensing

Mon 30 November 2015 13:00 to 13:30

Venue Atlas, gebouwnummer 104
Room 1

By Andrei Mîrț (Romania) 


Illumination normalization is a novel algorithm that mitigates effects like shadowing, caused by varying illumination conditions, on airborne high resolution hyperspectral data. The method has been designed for forestry scenes and it requires a 3D model of the top of the canopy. The algorithm calculates the ratio of direct sunlight to diffuse skylight and then uses geometrically estimated illumination coefficients to balance illumination over the whole scene. The illumination incident angle and portion of the visible sky are determined from the 3D top of the canopy model and used in a linear regression on a constant reflectance region, i.e. a single tree crown manually selected. The algorithm was tested on a Guyanese rainforest dataset. The normalized data showed a decrease in standard deviation for EVI and NDVI and improved, in average, self-similarity for tree crowns for SAM and MD classifiers.

Keywords: deshadowing; hyperspectral; UAV; rendering equation; canopy; forestry.