In Section 2, the basic PolSAR and PolInSAR models, AMD, Shannon entropy characterization, and the DPM model are briefly introduced as the basic theory underpinning this paper.
In the case of PolInSAR data, the backscattered signal can be described by a six-element complex target vector corresponding to the polarimetric performances at the two antennae:
The PolInSAR homogeneous region is statically characterized by a 6 x 6 covariance matrix :
In a homogeneous region of PolInSAR data, the Shannon entropy expression S[[??]] can be decomposed into a sum of three terms:
The algorithm first calculates the Shannon entropy characteristics of the PolInSAR data and uses the three terms [S.sub.I][??], [S.sub.P][??] and [S.sub.[mu]][??] as the multichannel data to extract local histograms, which are the basic inputs for the DPM-based histogram clustering procedure.
1) We first extract the Shannon entropy characteristics from the PolInSAR data and use the three Shannon entropy terms as multichannel input images to extract local histograms.
This decomposition is applied to one polarimetric channel of the PolInSAR data.
In this section, we illustrate the effectiveness of the proposed classification algorithm using E-SAR PolInSAR data from the German Aerospace Center (DLR).
We proposed a data-driven unsupervised classification algorithm for PolInSAR data.
The main purpose of the proposed algorithm is to classify data from the point of scattering mechanism, so we only use the Shannon entropy characterization of the PolInSAR data to extract local histogram features for DPM model-based histogram clustering, and interferometric PolInSAR information is not fully used.