National Conference on Recent Advances in Electronics & Computer Engineering, RAECE -2015, Feb.13-15, 2015, liT Roorkee, India An Approach to Use Polarimetric Signature for Land Cover Classiication Ajay Kumar Maurya 1, Tasneem Ahmed 2 , Dharmendra Sin�h 1 , Raman Balasubramanian 3 I Department of Electronics & Communication Engineering, Department of Mathematics 3 Department of Computer Science & Engineering Indian Institute of Technology Roorkee, Roorkee-247667, India Email: ajO131pec@iitr.ac.in, tas19dma@iitr.ac.in, dharmfec@iitr.ac.in, balarfma@iitr.ac.in Abstract-The aim of this paper is to explore the information obtain from the fully polarimetric SAR data. Fully polarimetric SAR data contain both magnitude and phase information therefore it contains great potential about target classiication. By using both parameter amplitude and phase we can distinguishes different types of scattering mechanism. For fully utilization of polarimetric SAR data, polarization signatures are used which utilizes the different orientation angle. Polarization signature is a 3-D plot of the received backscattered intensity as a function of ellipticity and orientation angle of antenna. Polarization signatures of urban area are equivalent to dihedral corner relector which shows the double bounce, polarization signature of water is equivalent to trihedral which shows single bounce and short vegetation shows the polarizations signatures equivalent to dipole at different orientation angle. In this paper, by utilizing the fully polarimetric ALOS-PALSAR data, polarization signatures are extracted at different angles and their capability to classify different land cover classes like; urban, water, short vegetation, tall vegetation and bare soil are explored. The scattering mechanism of generated elliptical and linear polarized images for above mentioned land cover classes is also analyzed and on the basis of their scattering mechanism, decision tree classiication (DTC) algorithm has been proposed and performance of the algorithm is also compared with other supervised and unsupervised classiication techniques. Keywords-P ALSAR; polarization signature; seperabiiy index; decision tree classication; I. INTRODUCTION The RADAR polarimetry is an energetically developing method of SAR image processing. RADAR polarimetry is popular for land cover classiication and extraction of information rom satellite data. Many decomposition methods were reported in the literature. Cloude and Pottier (1996-97) [1,2] introduced the concept of entropy (H), alpha (a) and anisotropy (A) which play an important role and worked as standard tool for target classiication. Entropy and alpha measures randomness and characterize scattering mechanism, but main drawback of this decomposition is random location of decision boundary. Classiication techniques mainly classiied into two categories, Supervised (maximum likelihood, minimum distance etc.) and unsupervised (K mean, ISO data) techniques. Every classiication technique has its own advantages and disadvantages. Main disadvantage of 248 supervise classiication techniques is due to single discriminative classiier which work on the pixel by pixel labeling. If during classiication any pixel remain unidentiied then it not assign to any class. In unsupervised classiication techniques natural clustering obtained during the classiication iteration and not necessarily that cluster belong to desired class because user not having any control on the clustering. Except these techniques, non parametric classiier such as decision tree classiier (DTC) has advantage to not only non linear relations between classes and features but also it can be applied on noisy data [ 3]. Non parametric classiiers also having some disadvantage such as it need large number of training data. Now, due to availability of ully polarimetric SAR data, amplitude and phase information is also provided which is very helpul to classiy the target more accurately and we can enhance more information about the target. Fully polarimetric SAR data (HH- horizontal transmit and horizontal receive, HV- horizontal transmit and vertical receive, VH- vertical transmit and horizontal receive, VV vertical transmit and vertical receive) provides more information in comparison to single or dual polarized data. Thus, we can also use circular and elliptical polarization combination to extract the more information about the target. Linear polarization having single plane horizontal or vertical in the direction of propagation but circular and elliptical polarization contains all the planes. Due to containing all planes its help to detect target at different orientation angle or alignment to radar line of sight [5]. For ully utilization of the SAR data, polarization signatures can also be utilized [4]. Polarization signature is a 3-D plot which shows the variation of backscattered intensity for different combination of orientation and ellipticity angle. Peak in the polarization signature represent maximum received backscattered for that orientation angle and valley in polarization signature represent mInImum received backscattered for corresponding polarization angle. To reduce the complexity only two combinations i.e. co- polarized signature and cross polarized signature are measured. Backscattered intensity for different target varies with change in orientation and ellipticity angle. For a particular polarization states one target give maximum backscattered intensity but other target for same polarization states give minimum intensity. If we obtain that orientation and ellipticity angle combination for which one class give