A supervised classification of multi-channel high-resolution SAR data|
Dirk Borghys and Christiaan Perneel
Many methods have been proposed in literature
for the supervised classification of multi-channel (polarimetric
and/or multi-frequency) SAR data. Most of them are based on the extraction of a
set of features from the original SAR data. In this paper the influence of
these features on the results of the classification is examined in a
quantitative manner. A set of multi-channel (P, L, C and
X band) SAR data was acquired by an airborne system over a site in Southern
Europe. A ground-truth mission defined the classes for learning and validation.
A feature-based classification method, based on logistic regression, is used
for detecting each of the classes. Logistic regression combines the input
features into a non-linear function, the logistic function, in order to distinguish
that class from all others. For each class a 'detection image', with a
well-defined statistical meaning, is obtained. The value at each pixel in the
detection image represents the conditional probability that the pixel belongs
to that class, given all input features. The logistic regression is performed
using a step-wise method in which, at each step, the most discriminating
feature is added to the selected feature set, but only if its addition
contributes significantly to the detection. The logistic regression thus also
performs a feature selection. Moreover, logistic regression allows combining
input data with very diverse statistical distributions.
The main aim of the current paper is to investigate the usefulness of each
feature for the detection of the different classes.
View Full Text (pdf file, 1 MB)
Submitted: 21 June 2006
Revised: 26 Feb 2007
Accepted: 28 Feb 2007
Published: 21 Mar 2007
Responsible editor: Mario Caetano
Borghys D & C Perneel, 2007.
A supervised classification of multi-channel high-resolution SAR data.
EARSeL eProceedings, 6(1): 26-37