The 18th International Radar Symposium IRS 2017, June 28-30, 2017, Prague, Czech Republic
978-3-7369-9343-3 ©2017 DGON
1
Efficient Object Classification and Recognition in SAR Imagery
Ievgen M. Gorovyi and Dmytro S. Sharapov
Department of Microwave Electronics, Institute of Radio Astronomy of NAS of Ukraine
4 Chervonopraporna Str., Kharkov 61002, Ukraine
gorovoy@rian.kharkov.ua, sharapov@it-jim.com
Abstract: SAR is a very popular instrument for imaging of the ground surface. Possibility
of high-resolution image formation makes it superior tool for various information
extraction tasks. In the paper, a problem of automatic target recognition is
comprehensively analyzed. An idea of azimuth and range target profiles fusion is
proposed. It is demonstrated, that usage of a proper image preprocessing with
appropriate feature extraction steps allow to achieve a competitive recognition accuracy
while keeping a low-dimensionality of feature vectors. Experimental results are discussed
for a publicly available MSTAR dataset.
Keywords: synthetic aperture radar; automatic target recognition; SAR image, feature
extraction; support vector machines; object classification
1. Introduction
Synthetic aperture radar (SAR) is a widely used instrument for various remote sensing tasks
[1]-[3]. Among them, the problem of automatic target recognition (ATR) has attracted a
considerable attention of SAR research community [4]-[5].
It is known, that SAR images significantly differ from optical images [6]. Basically, two main
challenges arise. Firstly, existence of the speckle noise due to a coherent acquisition of
backscattered signals. Secondly, a varying reflectivity of man-made objects as a function of
viewing angle. In this context, a multi-look processing has a great importance [1]-[3].
The problem is that the human visual system (HVS) often fails in real target identification.
This explains a high interest in automatic methods for SAR ATR.
A typical SAR ATR problem can be divided into three key stages (Fig. 1).
Figure 1. Key steps of ATR
The first stage is detection. Commonly, some target candidate regions are extracted from the
initial SAR images [6]-[7]. Typically, there are a lot of false positives after this step. This can
be easily explained due to the fact, that some clutter objects like buildings, trees, bridges also
introduce high reflectivity level. This explains the necessity of the second step: discrimination.
This stage can be considered as a two-class classification problem [6]-[7]. Finally, the third