Land Cover Classification of Full Polarimetric PALSAR Images using Decision Tree based on Intensity and Texture Statistical Features

Abstract

Although there are numerous land cover classification methods, still some restraints presents while labelling distinct classes to which it actually belongs to, without any past available information For an SAR image backscattering coefficient and its texture are significant characteristic to portray an image. In this paper, a classification technique for PALSAR image using decision tree based on intensity and its texture statistical features has been developed. The statistic texture features like homogeneity, mean, entropy, variance, contrast, correlation, dissimilarity, and second moment is analyzed and their capability to classify SAR image into diverse land cover classes has been evaluated. The Seperability index idea is used to analyze the prominence of texture features in classifying each land cover class from remaining classes. The proposed classification method is applied on ALOS PALSAR HV polarized image. The decision tree based classifier uses these data to classify individual pixel into one of the four categories: water, bare soil, urban and vegetation. The quantitative results shown by the proposed method gives overall classification accuracy of about 95.88% and kappa coefficient of 0.9490.