Adaptive scale segmentation algorithm for polarimetric SAR image
Polarimetric SAR (PolSAR) data can be characterised by scattering matrix, which contains four elements. It is difficult to merge all the elements of the scattering matrix for segmentation. On the other hand, a PolSAR image contains objects of various scales, so segmentation on a single scale may lead to over-segmentation and under-segmentation. To address these two problems, an adaptive scale segmentation algorithm for PolSAR image is proposed. First, extract a synthetic gradient image that contains scattering information of all polarisation channels. Second, to prevent the over-segmentation of the watershed algorithm, extract the markers of the synthetic gradient image adaptively. Finally, combined with the result of PolSAR classification, the segmentation scales are adaptively selected for the segmentation of the synthetic gradient image. The results show that the proposed algorithm can not only accurately extract the segmentation boundaries of different objects but also overcome the over-segmentation of large-scale objects and under-segmentation of small-scale objects. Compared to fixed scale segmentation algorithms, the segmentation result of the proposed algorithm has the best quantitative index, that is, the lowest global heterogeneity evaluation index.