Spatially Variant Mixtures of Multiscale ARMA Model for SAR Imagery Segmentation

摘要

We propose a new model built on multiscale tree structure, spatially variant mixtures of multiscale autoregressive moving average (SVMMARMA) model, for unsupervised synthetic aperture radar (SAR) imagery segmentation. We derive an expectation maximization (EM) algorithm for learning the pixel labeling as well as the parameters of the component models. We also present the bootstrap sampling technique applied to the parameter estimation, which not only increases estimation precision, but also saves computation time greatly. Finally, we design classifier based on Euclidean distance of multiscale ARMA coefficients. Experiments results show this model gives better results than previous methods.