POLARIMETRIC-SAR CLASSIFICATION USING FUZZY MAXIMUM LIKEHOOD ESTIMATION CLUSTERING WITH CONSIDERATION OF COMPLEMENTARY INFORMATION BASED ON PHYSICAL POLARIMETRIC PARAMETERS, TARGET SCATTERING CHARACTERISTIK, AND SPATIAL CONTEXT

This paper shows a study on an alternative method for unsupervised classification of polarimetric-Syenthetic Aperture Radar (SAR) data. The first step was to extract several main physical polarimetric parameters (polarization power, coherence, and phase difference) from polarimetric covariance matrix (or coherency matrix) and physical scattering characteristics of land use/cover based on polarimetric decomposition (Cloude decomposition model). In this paper, we found that these features have complementary information which can be integrated in order to improve the discrimination of different land use or cover types. Classification stage was performed using Fuzzy Maximum Likelihood Estimation (FMLE) clustering algorithm. FMLE algorithm allows for ellipsoidal clusters of arbitrary extent and is consequently more flexible than standard Fuzzy K-Means clustering algorithm. Hoever, basic FMLE algorithm makes use exclusively the spectral (or intensity) properties of the individual pixel vectors and spatial-contextual information of the image was not taken into account. Hence, poor(noisy) classification result is ussualy obtained from SAR data due to speckle noise. In this paper, we propose a modified FMLE which integrate basic FMLE clustering with spatial-contextual information by statistical analysis of local neightbourhoods. The effectiveness of the proposed method was demonstrated using E-SAR polarimetric data acquired on the area of Penajam, East Kalimantan, Indonesia. Result showed classified images improving land-cover discrimination performance. Exhibiting homogeneous region, and preserving edge and other fine structures. Keywords: Cloudes polarimetric decomposition, FMLE clustering, polarimetric coherence, Polarimetric-SAR, unsupervised classification