You must have a javascript-enabled browser and javacript and stylesheets must be enabled to use some of the functions on this site.
 
   ESA       
   
Introduction

 

Fully Automatic Land Cover Maps Generation Using Polarimetric SAR Data

Fabio Del Frate(1), Marco Del Greco(1), Chiara Pratola(1), Cosimo Putignano(2) and Domenico Solimini(1)

(1) Tor Vergata University, Via del Politecnico, 1, 00133, Italy
(2) GEO-K S.r.l., Via del Politecnico, 1, 00133, Italy

Abstract

Ongoing SAR missions are already providing a huge amount of polarimetric data and many more are expected in the next future from new planned missions. Such a scenario suggests that effective mining and exploitation of the available information may require fully automatic procedures. Processing of polarimetric data for classification purposes has been carried out by a variety of supervised algorithms which span from Bayesian Maximum Likelihood to Fuzzy Logic to Support Vector Machines to Multi-Layer Perceptrons (MPL). A training phase performed under human supervision is usually required by these procedures, thus preventing them from running in a fully automatic mode. Target decomposition provides a way to discriminate the observed surface types in an unsupervised fashion. This method yields classification maps directly in terms of scattering mechanisms, rather than in terms of final end user classes.

This study proposes a novel fully automatic approach for producing land cover maps from polarimetric SAR images. The method is based on the implementation of four basic steps: in the first one an unsupervised clustering algorithm is applied, in the second one the singled out clusters are labelled based on their objective electromagnetic scattering properties, the third step performs an automatic selection of representative pixels of the image, which are used for the fourth step where the final generation of the map is carried out by means of an automatically trained technique. Step one has been implemented by means of three different algorithms, i.e., SOM (Self Organizing Maps), k-means and H-A-alpha decomposition. Step 2 is based on results from canonical scattering models and, in some cases, on the concurrent use of ancillary data. Step 3 requires the statistical distributions of backscattering within the pixels belonging to each class, while step 4 is carried out using MLP Neural Networks.

The described technique has been applied to two different data set. In the first case study, the measurements have been collected by the DLR E-SAR Synthetic Aperture Radar system onboard the Dornier DO 228 aircraft, which imaged the Frascati-Tor Vergata test site at very-high spatial resolution (~2 m) at L-band in a fully polarimetric mode. The second case study considers polarimetric ALOS images taken over the same test site. The performance of each specific implementation of the fully automatic processing chain in discriminating among the main land cover classes of the observed area (residential, asphalt, arboreous, permanent and seasonal crops of various kinds) is shown and discussed for both data sets.

 

 

  Higher level                 Last modified: 07.05.06