10_225o - Approximation of the KA distribution by the GA 0 distribution

In the field of processing and analysis of Synthetic Aperture Radar (SAR) images, the returned signal can be modelled as the product of the inherent speckle noise and the terrain backscatter. For amplitude SAR images, the data can be fitted with several distributions depending, among other considerations, on the degree of homogeneity of the areas under study. In this work, unless otherwise stated, linear detection (amplitude) data will be used. In zones where the backscatter can be considered homogeneous, for example: crops, pastures, etc., the Γ distribution is a good model for the returned signal. The KA distribution gives a good fit for homogeneous areas as well as for heterogeneous areas (for example forest on flat terrain) but there are numerical problems caused by the presence of Bessel functions. The Γ distribution does not explain data from heterogeneous zones. When the area under study is extremely heterogeneous, as it is the case of cities, or forest on undulated terrain, the Γ distribution and the KA distribution fail to fit these data. In this case, the GA distribution behaves very well. Taking also into account that this distribution fits equally well homogeneous and heterogeneous areas too, and that its use is more computational and theoretically tractable, it is desirable to substitute the GA distribution for the KA distribution. In this work the feasibility of this substitution will be studied. To this end, a correspondence between the parameters of both distributions is proposed in order to approximate, in some sense, the KA distribution by the GA 0 distribution. The minimization of a distance between both densities in order to obtain such correspondence will be considered, and the goodness of fit of between KA distributed data by the GA distribution model will be measured, using the χ adherence test in Monte Carlo experience.