Automated hierarchical classification of SAR images

This paper faces the problem of achieving a satisfactory classification of SAR data, based on statistical methods such as maximum likelihood. The proposed method performs a hierarchical classification, making an automatic feature selection, simulating the behaviour of a real exhaustive selection in the features space. The method, whose accuracies outperforms those obtainable with a classical ML one shot, uses a new statistical characterisation of classical features as sample mean and sample variance that exploits the spatial correlation between data