Análisis polarimétrico a través de “SAR Polarimetry Target Analysis” como apoyo para obtener coberturas de la tierra con la metodología Corine Land Cover.

Abstract

The constant research and development carried out by the scientific community to develop algorithms to obtain polarimetric information from targets captured with SAR imagery has become a useful tool for obtaining information on land cover; thus, in this research, a fully polarized image of the RADARSAT 2 programme (Quad Pol) taken on 1st September 2008 was used, located on the Bogota, which was processed with the CATALYST Professional program, formerly PCIGeomatics, in order to determine the characteristics or polarimetric signatures of land cover obtained with the CORINE Land Cover methodology (continuous urban fabric, annual or transient crops, confined crops, clean pastures, dense natural forest, planted forest, dense grassland, dense shrubland and inland waters). For this purpose, SAR image filtering processes were performed to reduce noise or speckle, such as the Lee filter 5 X 5 and BoxCar 3 X 3, and the following polarimetric decomposition algorithms were applied: Krogager, Cloude & Pottier (Entropy, alpha, beta, and anisotropy), Freeman & Durden, PHDW, Touzi and Van Zyl, as well as matrix transformation algorithms: C4R6C and C3R3C. The result layers were visually analyzed in order to infer particular characteristics and relevant as (intensity, texture, compactness, shape) for each type of land cover. Subsequently, these result layers were deployed in the SAR Polarimetry Target Analysis (SPTA) module. However, only the two filtered files and two matrix transformation files could be deployed and analyzed due to that they are totally polarized. From these four layers, polarimetric characteristics of the above-mentioned land cover were obtained by the co-polarization methods, Cloude & Pottier (H/A/α/β) and Freeman & Durden. Finally, the supervised classification was carried out through object analysis by the methods of Vector Support Machines (VSM) and Random Forests (RF) on SAR images with decomposition of Krogager, Freeman & Durden and PHDW, obtaining Kappa coefficients of 0. 634, 0. 600 and 0. 637 respectively, those that fall into the very good quality classification range.