IEEE - Aerospace and Electronic Systems - March 2021 - 60

Automatic Target Recognition on Synthetic Aperture Radar Imagery: A Survey
Table 3.

Most Referenced MSTAR SOC and EOC Evaluation Subsets
SOC

SOC ID of most referenced target subsets (involved targets)
1 (all target serial No for training and testing)

10-class

3-class

2 (2S1 - BMP2 (9563) - BTR60- BTR70 - D7 - T62 -
T72 (132) - ZIL131 - ZSU23/4 for training and
testing)

1 (all BMP2 classes - BTR70 - all T72 classes, both for training and testing)

EOC

Training targets

Testing targets

Description

depression
variation

2S1 - BRDM2 -ZSU23/4

2S1 - BRDM2 - ZSU23/4

Testing at 15 , 30 , and 45 depression angles

version
variants

BMP2 (9563) - T72 (132) - BTR60 -
T62

BMP2 (9566) - BMP2 (c21) -
BMP2 (812) - T72 (s7) -
BTR60 - T62

Testing at 15 against training targets and BMP2
version variants

noise
variation

2S1 - BMP2 (9563) - BTR60 -
BTR70 - D7 - T62 - T72 (132)- ZIL
131 - ZSU23/4

10-class with SOC ID 2

Testing at 15 with additive artificial Gaussian
noise at SNR ¼ f10; 5; 0; À5; À10gdB

occlusion
variation

10-class with SOC ID 2

10-class with SOC ID 2

Testing at 15 with additive occlusion level at
f10; 20; 30; 40; 50g%

resolution
variation

10-class with SOC ID 2

10-class with SOC ID 2

Testing at 15 with resolution variation
atf0:3 Â 0:3; 0:4 Â 0:4; 0:5 Â 0:5; 0:6 Â 0:6; 0:7 Â 0:7gm

Table 4.

Feature-Based Methods
No

Reference

SAR chip
size

Pre-processing

Main features

1

Belloni et al. [36]

variable

-

BRISK features, Hamming distance for
classification

2

Bolourchi et al.
[43]

128x128

Histogram equalization, averaging filter,
thresholding

LibSVM classification

3

Amrani et al. [40]

128x128

-

Gabor and HOG features, Mahalanobis
distance for classification

4

Tan et al. [46]

88x88

Histogram equalization, averaging filter,
thresholding, morphological operations

Target outline " dictionary, " a variant of
Hausdorff distance metric for classification

5

Clemente et al.
[45]

128x128

-

Krawtchouk moments, k-NN classification

determine the target's class. Further methods include
binary operations [47], sparse robust filters [48], and the
azimuth and range target profile fusion [49].

ATTRIBUTED SCATTERING CENTERS (ASC)
Scattering centers or scatterers are the dominating highfrequency radar reflections from surfaces such as corners
and flat plates. ATR techniques that exploit the scattering
centers model the scattering behavior predicted by physical optics and the geometrical theory of diffraction and
statistically estimate the object's location, geometry, and
polarization response. Each scattering center is a function
of frequency f and aspect angle w, and the total backscattering field is the sum of these individual scattering centers
[50], [51]:
60

E ðf; '; QÞ ¼

No
X

Ei ðf; '; ui Þ

(1)

i¼1

 ai
f
Ei ðf; '; ui Þ ¼ Ai Â j
fc
!
4pf
Â exp Àj
ðxi cos ' þ yi sin 'Þ
c
(2)


2pf
Â sin c
Li sin ð' À 'i Þ
c
Â expðÀ2pfg i sin 'Þ
where fc is the radar center frequency, c is the propagation
velocity, No is the number of scatterers with Q ¼
½u1 ; . . . :; ui ; . . . :; uNo  incorporating the parameter matrices

IEEE A&E SYSTEMS MAGAZINE

MARCH 2021
IEEE - Aerospace and Electronic Systems - March 2021

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - March 2021
