A HIERARCHICAL MODEL-BASED POLARIMETRIC SAR IMAGE DECOMPOSITION BASED ON COHERENCY MATRIX

The nonnegative eigenvalue constraint is a complete way to avoid the negative powers in model-based decomposition approaches. But in nonnegative eigenvalue decomposition (NNED), the volume scattering model is always subtracted prior to the remaining scattering mechanisms. Furthermore, the rank of the remaining coherency matrix is acquiescently assumed equal to two, that is to say, expect for the volume scattering the other scattering is weakly depolarized. However, the remaining scattering maybe strongly depolarized owing to the complexity of ground targets. In some sense, in NNED, we assume that the volume scattering contribution is always dominant Obviously, this is not applicative for all pixels, and the decomposition should be hierarchical according to the dominant scattering mechanism So in this paper, for the pixels that volume scattering dominates, the general NNED is adopted. For the pixels that surface or dihedral scattering dominates, a new decomposition framework is developed, in which the dominant scattering mechanism can be automatically obtained according to the alpha angle as well as the three corresponding power coefficients. Particularly, we also present an alternative way to alleviate the negative powers by adaptive fitting between model and data, by which the decomposition will be more accurate and stable than the most existing approaches. The results of the proposed scheme show great improvements using the AIRSAR data set.