Weighted Support Vector Machine Segmentation of SAR Images Based on MARMA model

Synthetic aperture radar (SAR) is a coherent sensing device. Existing algorithms for processing optical images are not suitable for SAR images because of speckles noise in SAR images. This paper introduces the support vector machine (SVM) segmentation of SAR images based on multiscale autoregressive moving average (MARMA) model, which can capture the statistical scale-dependency of SAR images. Firstly, the multiscale sequences of SAR image are constructed. Secondly, the paper investigates how to establish MARMA model and how to extract the multiscale stochastic characteristics of the different SAR texture images. Finally, the paper classifies the characteristics vector using generalized weighted SIM. Experiments show that the proposed algorithm is efficient.