An empirical (target-) BRDF normalization method has been implemented for hyperspectral data processing, following the approach of Kennedy, published in 1997. Correction results of this method highly depend on the successful application of an appropriate spectral pre-classification
which necessarily must be insensitive to reflectance anisotropy.
A standard classification output (as of ATCOR-4) is first evaluated for its suitability concerning anisotropy normalization. A hierarchical BRDF selection scheme is then set up, covering the most prominent target classes in the image. A classification algorithm is then evaluated on the basis of a standard spectral angle mapper (SAM) approach with the RSL’s spectral database SPECCHIO attached for reference spectra evaluation. Results show that the ATCOR-4 pre-classification output is highly sensitive to the reflectance anisotropy and therewith not suited for pre-classification. The SAM pre-classification is still under investigation, but first tests using reference spectra out of the reflectance data itself showed problems due to the high robustness against brightness differences
since a gradient must be estimated from targets of comparable brightness.