SAR Image Classification Based on MAP via the EM Algorithm

With the analysis of statistical models of scattering, a novel classification algorithm suitable for synthetic aperture radar (SAR) images is presented based on maximum a posteriori (MAP) criteria via expectation maximization (EM) algorithm. Multiplicative noise and high dynamics should be taken into account in SAR image classification. Our main work is to select a Gaussian mixture model and then use it in a MAP classification algorithm. The model is assumed that the whole distribution could be separated into finite parametric density distributions, and then the maximum likelihood parameters of each proportional distribution can be estimated by EM iterative computation. In our experiment, the stable factor and momentum item are introduced to ensure correct convergence and accelerate the convergence respectively. Finally, experimental results obtained from the MAP classification algorithm show that this approach is effective for SAR image classification using MSTAR database