Abstract:

Terrains with large variance are difficult to be discriminated by the traditional pixelbased classification methods for synthetic aperture radar (SAR) images. Aiming at solving this problem, an algorithm for terrain classification of SAR images based on the Bhattacharyya similarity between regions is proposed. Firstly, a proper image segmentation technology is applied to obtain homogeneous regions of the SAR image. Then, the Bhattacharyya similarity measuring the statistical proximity between regions is defined, of which the analytical expression referring to the Gamma distribution is derived. Finally, with the previous image regions as classification elements, terrain classification is implemented by a criterion of maximizing the Bhattacharyya similarity between regions. The experimental results on the real SAR image validates the superiority of the proposed algorithm to the pixelbased maximum likelihood classification method, support vector machine classifier and the regionbased method of minimizing the distance between regions.