Ein lernfähiger Fuzzy-Verteilungsschätzer zur adaptiven Analyse von Radardaten

Abstract

Presented is a new learning fuzzy distribution estimator (LFVS). It estimates similarities between pattern distributions and the distribution of a signal which is to be analysed. Learning capability gives a high flexibility to the system. Its application is not restricted to SAR data analysis. Compared to conventional estimation algorithms there are two significant advantages of the fuzzy approach. One is that the gradual similarity between density distributions can be established and the other is the estimation of reliability without additional computations. The fuzzy distribution estimator is applied to compress SAR raw data and it is a main modul of a new fuzzy classification system. This system allows a user adaptive classification of SAR image data. In addition to the classificaiton itself, the user gets pixel by pixel classification reliability. Pixels belonging to several classes - a typical problem of low resolution radar images - are well detected.