Segmentation of dual polarized SAR imagery of the west of China based on evidence theory

More areas in the west of China are mountains covered by cloud and mist all the year. Therefore mapping by negative remote sensing could not satisfy the requirements of thematic map. SAR is an alternative approach. Among single, dual and quad polarized imagery, dual polarized one is a compromised choice taking cost and information into account. A new algorithm based on the theory of evidence for dual polarized SAR images of west China is proposed in this paper. Firstly we get a single channel image from original dataset by PWF (Polarimetric whiten filter), and then over-segment the image. Secondly the borderlines of each segment are adjusted by fusing two evidences, pixels´ values and the values of their neighbourhood by applying evidence theory. Finally initial segments are merged based on evidence combination. In this paper, not only each channel but also the mean of neighbour areas is considered as a piece of evidence. In order to eliminate the factor of hypsography, the ratio of different polarization was added as a new piece of evidence. The experimental results of dual-polarimetric SAR images in Hengduan Mountain of West China show that the proposed algorithm performs better than traditional algorithm when noise rich images are processed because it is less sensitive to speckle noise so edge information is better reserved and it reduce the influence of hypsography.