Forward Modeling and Atmospheric Compensation in hyperspectral data: Experimental analysis from a target detection perspective

Abstract

Taking into account atmospheric effects is crucial in target detection of airborne/satellite hyperspectral images. In regard to this, two physics-based approaches to atmospheric radiative transfer modeling are considered here: Atmospheric Compensation (AC) and Forward Modeling (FM). An experimental analysis is presented that encompasses target detection both relying upon an atmospherically compensated reflectance image and by generating predicted radiance target spaces through a forward modeling approach. Real hyperspectral imagery that embodies a very challenging, cluttered, mixed pixel detection problem is used to compare AC and FM approaches from an operational target detection perspective. On this data, detection in the radiance domain through FM has proven to be as effective as the standard AC plus reflectance domain processing. Experiments have also highlighted several aspects of FM approach (e.g. its intrinsic simplicity, flexibility, and applicability) that should be considered when performing target detection, especially for targets affected by high variability.