Mitigation of variable illumination effects in hyperspectral imagery
2 Excel Geo Ltd
This project will conduct a feasibility study aimed at improving the operational effectiveness (i. e. reduce costs, increase envelope, improve exploitation) of airborne hyper-spectral remote sensing.
The key objective will be to develop and evaluate techniques to mitigate the effects of variable illumination, in particular cloud shadow, on the recovery of surface reflectance.
The techniques will be developed and tested on a significant volume of empirical and modelled data.
The main areas of focus are;
* To determine if the at-sensor radiance spectra reveal information regarding the local illumination conditions via radiative modelling and empirical data collection.
This seeks to identify traits embedded in the at-sensor radiance that may provide information regarding the local illumination conditions.
Existing techniques include empirically derived ‘shadow indices’ that use a few selected bands; the innovation is to examine the entire spectrum to seek an improved technique.
* To exploit the statistics of the illumination field (determined above), including in shadow, to devise a mitigation approach.
This approach is an extension of previously published ‘invariance’ techniques designed to be robust to varying atmospheric and environmental factors.
These determine the subspace which contains the majority of variance caused by these factors.
These unwanted dimensions are then ‘projected out’ of the at-sensor signature.
These techniques have been limited to clear sky conditions by ‘off-the-shelf’ radiative transfer models.
The innovation is to include the attenuating effects of cloud in the determination of an invariant subspace thus enabling robust exploitation over a wider range of environmental conditions.
* To develop a empirical correction technique exploiting time series, cloud-free airborne and satellite imagery.
This approach is motivated by the increasing regularity and availability of satellite imagery.
Published techniques crudely replace shadowed pixels with the corresponding cloud-free pixel from an image acquired at around the same time.
This approach is not suitable to be applied to sources with differing spectral and spatial response (i.e. satellite and airborne).
The innovation is to develop a technique in which a correction for cloud shadow in the hyper-spectral image is derived from that of a satellite (e.g. Sentinel-2) multi-spectral image.