Data driven PolSAR unsupervised classification based on adaptive model based decomposition

In this paper, we propose a data-driven unsupervised PolSAR image classification algorithm using adaptive model-based decomposition. By estimating a mean orientation angle and a degree of randomness for canopy scattering, the adaptive decomposition method provides a refined expression of volume scattering component in which no scattering reflection assumption is required. The classification algorithm combines affinity propagation and iterated Wishart classifier. In terms of being data-driven, it can automatically determine the number of clusters. Experimental results on the NASA/JPL AIRSAR L-band PolSAR data of San Francisco demonstrate the effectiveness of our algorithm.