Generic Object Recognition in High Resolution Sar Images

GENERIC OBJECT RECOGNITION IN HIGH RESOLUTION SAR IMAGES
A. Popescu1, M. Costache2, J. Singh3, M. Datcu1,3, and G. Schwarz3
1)

University Politehnica Bucharest
2)
Télécom ParisTech
3)
German Aerospace Center DLR

ABSTRACT
This paper presents a non-parametric modeling scheme for
high resolution SAR data, based on Short Time Fourier
Transform which is able to integrate the radiometrical and
morphological properties of the data, for object recognition,
scene and target indexing, addressing the problem of large
data base queries and information retrieval.. The method is
assessed by using a Bayesian Support Vector Machine
image search engine based on a hierarchical learning model.
The method allowed for the recognition of over 30 different
classes, both homogeneous and heterogeneous urban objects
with high levels of details. Qualitative and quantitative
measures for evaluation are presented and discussed.
Index Terms— Short Time Fourier Transform, object
recognition, high resolution SAR data
1. INTRODUCTION
Many users of Earth observation SAR images are only
interested in individually selected application tasks. Thus,
instead of applying universal data analysis tools, SAR users
often aim at interactively controlled feature determination
based on object size, object motion, double bouncing, or
occultation.
In contrast to these typically geometry-driven
approaches, we propose a combined radiometry/
morphology-driven approach that allows us to detect,
identify and classify arbitrary and unexpected objects being
“machine visible” in high resolution SAR images.
The proposed approach is motivated by the
characteristics of typical high resolution remote sensing
SAR images. On the one hand, we see extended surface
areas such as forests or ocean surfaces: their main criteria
are overall brightness and texture. On the other hand, we
can also see an enormous number of small-scale details
especially in urban areas, where three-dimensional effects
and multiple scattering have to be...