AUTOMATIC RECOGNITION OF MSTAR TARGETS USING RADAR SHADOW AND
SUPERRESOLUTION FEATURES
Jingjing Cui, Jon Gudnason, Mike Brookes
Department of Electrical and Electronic Engineering
Imperial College London
ABSTRACT
Automatic target recognition from high range resolution
radar profiles remains an important and challenging prob-
lem. In this paper, we present a novel feature set for this task
that combines a representation of the target’s radar shadow
with a noise-robust superresolution characterisation of the
target scattering centres derived from the MUSIC algorithm.
Using an HMM to represent aspect dependence, we demon-
strate that the inclusion of the shadow features results in
a significant improvement in recognition performance. We
evaluate our proposed feature set on a closed-set identifica-
tion task using targets from the MSTAR database and show
that it results in lower recognition error rates than previously
published methods using the same data.
1. INTRODUCTION
The automatic detection and classification of targets from
their radar signatures is an important and difficult problem
that has attracted considerable research effort. Algorithms
for target recognition from high range resolution (HRR)
radar signals generally use as their primary input either a
synthetic aperture radar (SAR) image or else a sequence of
one or more one-dimensional range profiles. The image-
based approaches generally have higher performance but
are much less robust to target motion because of their long
data acquisition time. Some image-based algorithms use
the pixel values of the image as their recognition features
[9, 10, 12] while others first transform the image to another
domain [4, 16]. An alternative approach for targets that are
large compared with the radar wavelength is to model the
radar return as emanating from a discrete set of orientation-
dependent points known as scattering centres [1]. In this ap-
proach, the SAR image is processed to generate an explicit
list of scattering centre positions and associated radar cross
sections on which the recognition features are based [2, 8].
In the same way, systems that act on the one-dimensional
range profiles can either use the raw [18] or transformed
This work was supported by the UK MoD through work funded by the
Defence Technology Center for Data and Information Fusion.
HRR Profiles
Aspect Angle [deg]
range [m]
90.5 91 91.5 92 92.5 93
35
30
25
20
15
10
5
SAR Image
cross range [m]
range [m]
5 10 15 20 25 30 35
35
30
25
20
15
10
5
Fig. 1. (a) HRR profiles (b) SAR image of T72 tank
[7] profile values as their features or else can process the
profiles to estimate the scattering centre locations and cross
sections [5]. Both SAR images and HRR profiles often ex-
hibit large variations for small changes in target orientation.
Target recognition systems must account for this aspect-
dependency by using a rotation invariant transform [4] or
by having multiple, orientation-dependent, target represen-
tations which may conveniently be embedded in an HMM
[5, 13, 16].
In this paper, we present a novel feature set for auto-
matic target recognition from a sequence of radar range pro-
files. Our feature set uses a noise-robust super-resolution
technique for identifying scattering centre locations and
combines this information with additional features that
characterise the shape of the radar shadow. Fig. 1 (b) shows
a SAR image of a T72 tank taken from the MSTAR [14, 15]
dataset. This image may be divided into three regions hav-
ing significantly different characteristics: (a) the target it-
self characterised by discrete scattering centres within a rel-
atively uniform background, (b) the target shadow with very
low signal levels and (c) a clutter region surrounding the tar-
get. As can be seen in this example, the shape of the shadow
region gives potentially useful information about the verti-
cal profile of a target that is sited on level ground. This
information is not available from the direct target returns
which are insensitive to vertical displacement. The shadow
information has been used by others to improve target detec-
tion [6] but is not generally used explicitly in target recog-
nition.
V - 589 0-7803-8874-7/05/$20.00 ©2005 IEEE ICASSP 2005