Land Cover Classification of Palsar Images by Knowledge Based Decision Tree Classifier and Supervised Classifiers Based on Sar Observables

Abstract

Abstract-The intent of this paper is to explore the application of information obtained from fully polarimetric data for land cover classification. Various land cover classification techniques are available in the literature, but still uncertainty exists in labeling various clusters to their own classes without using any a priori information. Therefore, the present work is focused on analyzing useful intrinsic information extracted from SAR observables obtained by various decomposition techniques. The eigenvalue decomposition and Pauli decomposition have been carried out to separate classes on the basis of their scattering mechanisms. The various classification techniques (supervised: minimum distance, maximum likelihood, parallelepiped and unsupervised: Wishart) were applied in order to see possible differences among SAR observables in terms of information that they contain and their usefulness in classifying particular land cover type. Another important issue is labeling the clusters, and this work is carried out by decision tree classification that uses knowledge based approach. This classifier is implemented by scrupulous knowledge of data obtained by empirical evidence and their experimental validation. It has been demonstrated quantitatively that standard polarimetric parameters such as polarized backscatter coefficients (linear, circular 48Mishra, Singh, and Yamaguchi and linear 45 • ), co and cross-pol ratios for both linear and circular polarizations can be used as information bearing features for making decision boundaries. This forms the basis of discrimination between various classes in sequential format. The classification approach has been evaluated for fully polarimetric ALOS PALSAR L-band level 1.1 data. The classifier uses these data to classify individual pixel into one of the five categories: water, tall vegetation, short vegetation, urban and bare soil surface. The quantitative results shown by this classifier give classification accuracy of about 86%, which is better than other classification techniques.

Coupling of Thresholding and Region Growing Algorithm for Change Detection in Sar Images

“…where To confirm the effectiveness of the NDR operator over widely used ratio operator, a following separability index is used [18].…”

Remote Sensing Image Classification Based on Stacked Denoising Autoencoder

“…Since parametric classifier requires knowing the distribution of data in advance, this is often difficult to achieve in remote sensing images. Therefore, the nonparametric classifier has been widely used, including artificial neural network, expert system, Support Vector Machine (SVM), decision tree, and so on [2][3][4][5][6]. All of the above methods, however, require analysis and extraction of a manually designed feature, and the overall classification accuracy is to be improved.…”

Variational Sar Image Segmentation Based on the G0 Model and an Augmented Lagrangian Method

“…SAR images have been applied to increasingly wide fields such as land cover classification [2], target detection [3] and information retrieval [4]. Segmentation is a key step for interpreting Synthetic Aperture Radar (SAR) images [1].…”

“…However, as the ML estimation of the G 0 distribution is hard to get, so we adopt the log-moment based method to estimate the parameters in each region. This can be done by using Equation (2). The log-moment for the foreground and background can be computed by using (3), which leads to the following equations:…”