scispace - formally typeset

Journal ArticleDOI

Automatic Recognition of ISAR Images of Multiple Targets and ATR Results

01 Jan 2014-Progress in Electromagnetics Research B (EMW Publishing)-Vol. 61, pp 43-54

TL;DR: This paper proposes an ATR procedure for targets flying in formation and shows that ISAR is an adequate tool for ATR even if an image is contaminated by radar reflections from neighboring targets.
Abstract: Inverse synthetic aperture radar (ISAR) imaging is an effective method to identify unknown targets regardless of weather and illumination conditions. Research results published regarding this topic have focused mainly on imaging and automatic target recognition (ATR) of single targets. However, targets generally fly in formation, so the applicability of ISAR images to ATR of multiple targets must be studied. This paper proposes an ATR procedure for targets flying in formation. ATR accuracy derived using five targets composed of point scatterers and the measured radar signal of a Boeing 747 aircraft was as high as that of the solo flight in terms of SNR and the size of the training database; this result shows that ISAR is an adequate tool for ATR even if an image is contaminated by radar reflections from neighboring targets.

Content maybe subject to copyright    Report

Progress In Electromagnetics Research B, Vol. 61, 43–54, 2014
Automatic Recognition of ISAR Images of Multiple Targets
and ATR Results
Sang-Hong Park
*
Abstract—Inverse synthetic aperture radar (ISAR) imaging is an effective method to identify unknown
targets regardless of weather and illumination conditions. Research results published regarding this topic
have focused mainly on imaging and automatic target recognition (ATR) of single targets. However,
targets generally y in formation, so the applicability of ISAR images to ATR of multiple targets must
be studied. This paper proposes an ATR procedure for targets ying in formation. ATR accuracy
derived using five targets composed of point scatterers and the measured radar signal of a Boeing 747
aircraft was as high as that of the solo flight in terms of SNR and the size of the training database;
this result shows that ISAR is an adequate tool for ATR even if an image is contaminated by radar
reflections from neighboring targets.
1. INTRODUCTION
An inverse synthetic aperture radar (ISAR) image [1–7] represents the two-dimensional radar cross-
section (RCS) distribution of a target. An ISAR image of the target can be generated by synthesizing
compressed radar signals at various observation angles. Due to its applicability regardless of weather and
illumination conditions, it is used in many military applications such as classification of non-cooperative
aircraft, and battlefield awareness [8, 9].
The main impediment to ISAR imaging is to identify ying targets that are moving in formation;
this is because each target may have complicated motion components, which can degrade the quality of
its ISAR image. Efficient algorithms such as range-Doppler [10] and time frequency analysis [11] were
proposed to image a single target, but when multiple targets occur in a single radar beam, imaging each
target separately is a difficult task. To solve this problem, [12] proposed a segmentation method that can
separate ISAR images of aircraft flying in formation; in this method, coarse alignment proposed in [13]
was improved by using a combination of a polynomial and Gaussian basis functions, and segmentation
was improved using an image processing technique.
However, the ISAR image segmented by the segmentation method is a mere projection of the target
and can be contaminated by radar signals reflected from neighboring targets. For this image to be useful,
the segmented image must be identified by the automatic target recognition (ATR) technique and its
accuracy must be evaluated. Many approaches to image stationary targets or those engaged in solo
flight have been presented and their ATR accuracy have been published [14]. However, little research
on segmenting ISAR images of targets flying in formation has been reported due to the sensitive military
applications of this topic.
This paper proposes a procedure for ATR of an ISAR image by fusing a method to segment
the ISAR image, a scenario-based method to construct the training database, and an efficient polar-
mapping classifier [1]. ATR results are presented by using the ISAR images segmented from three
targets composed of point scatterers (SCs) and the measured radar signal from a Boeing 747 aircraft.
ATR accuracy degraded < 1% in spite of the image quality degradation; this result demonstrates that
Received 8 July 2014, Accepted 21 August 2014, Scheduled 11 September 2014
* Corresponding author: Sang-Hong Park (radar@pknu.ac.kr).
The author is with the Department of Electronic Engineering, Pukyong National University, Busan, Korea.

44 Park
a 2D ISAR image is appropriate for ATR even when each image is contaminated by radar reflections
from neighboring targets.
2. PRINCIPLES AND SUMMARY OF THE CLASSIFICATION METHOD
2.1. Signal Modeling and Imaging Algorithm
For the radar signal, we assume the monostatic chirp radar signal that is widely used for high resolution
radar imaging. The transmitted chirp signal is given by [1]
r(t)=A
0
e
j2π
f
0
t+
Bt
2
2τ
× rect
t
τ
, (1)
where r(t) is a transmitted signal at time t, A
0
its amplitude, f
0
the start frequency, B the bandwidth,
τ the pulse duration, and rect = 1 for t τ/2 t t + τ/2 and 0 otherwise [1]. For a single target
composed of L SCs at aspect angle θ, the received reflected signal is [1]
g(θ,t)=
L
l=1
A
l
e
j2π
f
0
(td
l,θ
)+
B(td
l,θ
)
2
2τ
× rect
t d
l,θ
τ
, (2)
where A
l
is the amplitude of scattering center l and d
l,θ
the time delay between the radar and scattering
center l.ForK targets flying in formation, (2) is changed to [12]
g(θ,t)=
K
k=1
L
l=1
A
k,l
e
j2π
f
0
(td
k,l,θ
)+
B(td
k,l,θ
)
2
2τ
× rect
t d
k,l
τ
, (3)
where A
k,l
and d
k,l
correspond to A
l
and d
l,θ
in (2) for target k. Reflections from neighboring targets
interfere with the radar signal returned from each target.
In deriving the whole ISAR image of the multiple targets, the reflected signal is matched-filtered to
obtain range profiles (RPs) at each θ. Then RPs are aligned by minimizing the entropy cost function [3–
5] defined by
H
G
m
,G
m+1
=
N1
0
¯
G(s, n)ln
¯
G(s, n), (4)
where
¯
G(s, n)=
|G
m
(n)| + |G
m+1
(n s)|
N1
0
(|G
m
(n)| + |G
m+1
(n s)|)
, (5)
G
m
(n)andG
m+1
(n)areRPsm and m +1, and N is the total number of range bins. The s that
minimizes the 1D entropy is the shift that best aligns RP m + 1. Then, to remove residual phase errors
that range between ± (range bin)/2, the phase error vector that minimizes the following 2D entropy
function is found [12]:
Ent =
M
i=1
N
j=1
¯
I(i, j)
ln
¯
I(i, j)
, (6)
where
¯
I(i, j)
=
|I(i, j)|
2
M1
m=0
N1
n=0
|I(m, n)|
2
, (7)
and I is the ISAR image derived from the RPs with RP m multiplied by the mth element of the phase
vector. M is the number of RPs and N the number of range bins. Each RP is multiplied by each
element of the phase vector, and finally, SCs are separated in the cross-range direction by performing

Progress In Electromagnetics Research B, Vol. 61, 2014 45
a fast Fourier transform in each range bin. Because multiple targets are separated in the 2D range-
Doppler domain after this step, the ISAR image of each target can be clipped from the whole ISAR
image and the clipped image can be enhanced by applying range alignment (RA), phase adjustment
(PA) and fast Fourier transform (FFT) in each range bin to the clipped image.
To increase the effectiveness with which the ISAR image of each target is separated, we recently
proposed a new method that is composed of a new cost function for RA and a segmentation method [12].
Because constructive and destructive interference among various scatterers of multiple targets can cause
range profiles to be highly uncorrelated, the entropy cost function for RA can poorly align range profiles;
generally high-valued range bins were aligned in [13]. Therefore, for each range bin to have equal weight
in RA, we constructed a binary bulk image of the range profile history and proposed a new cost
function [13] which is the sum of all pixels that occur in the following polynomial:
f(t)=
P
i=0
p
i
t
i
+
G
i=0
a
i
exp
t b
i
c
i
2
, (8)
where P is the order of the polynomial, p
i
its parameters, G+1 the number of Gaussian basis functions,
and a
i
, b
i
, c
i
are their parameters. The parameters were found using a combination of the gradient-
based method and particle-swarm optimization (PSO) [15]. As the segmentation algorithm, we used
the conventional CFAR detector:
T (x)=
x ˆμ
c
ˆσ
c
, (9)
where x is the amplitude of the test pixel, ˆμ
c
the estimated mean of the clutter amplitude, and ˆσ
c
the
estimated standard deviation of the clutter amplitude. If T (x) exceeds a threshold, the corresponding
binary pixel value is 1; it is 0 otherwise. In setting T (x), if it is too large, the target pixel can be
removed and if too small, the clutter can be detected. In our numerous tests, T (x) between 2 and 4
was appropriate. Then, image closing defined as image dilation followed by image erosion was applied
to the binary image to ll the target region as follows:
C(A, B)=(A B) B, (10)
where C is the closing operation, the dilation operation, and the erosion operation (details in [12]).
The connected set of the binary was clipped and multiplied by the whole ISAR image to segment each
target from the ISAR image of the whole set of targets.
2.2. Proposed Classification Procedure
In classifying ISAR images, ISAR images of a test target must exist in the training database. This
requirement results in serious computational burden due the infinite number of aspect angles. For this
reason, we obtain the training database of the ISAR image by applying a method that is based on flight
scenarios of a target (Fig. 1) [1]. Assuming that a target moves at a given velocity in a given direction
starting from each point uniformly sampled in the 3D space (training space), the training database is
constructed using the ISAR images derived for each ight scenario.
In classification, various classifiers can be used. We use the polar mapping classifier (PMC) which is
invariant to variation in the scale and the translation of the ISAR image [16]. This method removes the
rotational variance by converting the rotation of the ISAR image to translation in the angular direction
by mapping the ISAR image in polar coordinates, then compressing it using principal component
analysis. The algorithm makes the final identification by comparing the compressed image to images in
a database.
Classification is conducted by fusing the principles mentioned above(Fig. 2). The test targetsfly
in formation starting from a random location in the training space (Fig. 1) and the reflected signal
is collected using (3). To derive the ISAR image of the whole set of targets, the collected signal is
matched-filtered and coarse range alignment is conducted using (8), PSO and the cost function in [12].
Then phase adjustment and cross-range FFT are applied to obtain the ISAR image of the test targets
under formation flight.
In segmenting each target, the binary image using (9) is constructed, image closing is conducted on
the binary image and finally the connected set of the binary was clipped and multiplied by the whole

46 Park
Figure 1. Scenario-based construction method ([1 1 0] direction) [1].
Figure 2. Proposed classification procedure.
ISAR image. The segmented ISAR image is classified by the PMC. Classification accuracyis expressed
as the correct classification percentage
P
c
=
N
c
N
t
× 100%, (11)
where N
c
is the number of the correct identifications and N
t
the number of test sets.
3. SIMULATION RESULTS
3.1. Simulation Results Using the Point Scatterer Model
To demonstrate the effectiveness of the proposed method and to evaluate the ATR accuracy of the
segmented ISAR image, simulations were conducted using targets composed of point SCs. The
simulation used pulse repetition frequency (PRF) = 2 kHz, center frequency f
0
=10.0 GHz, bandwidth
B = 200 MHz (0.75-m range resolution). The signal-to-noise ratio was varied to study the effect of
noise. The parameters used in PSO were population size = 50, number of generations = 20, ϕ =0.5,
and c
1
= c
2
=1.49 (definitions in [3]). Five targets were used for simulation; they were constructed

Progress In Electromagnetics Research B, Vol. 61, 2014 47
(a) Boeing-737 (b) F-14
(c) F-18 (d) Rafale
(e) Su-35
Figure 3. Scatterer targets used for classification.
using the 3D CAD data of a Boeing 737, an F-14, an F-18, a Rafale, and an Su-35 (Fig. 3). Two
separate simulations were performed for each formation flight; one with three identical targets and the
other with randomly chosen targets. The relative distances between the centers of the aircraft were
60 m for the Boeing-737, 35 m for the Su-35 and 25 m for the others.
The training database was constructed based on formation ights in two directions, [1 10]and
[0 1 0] and we assumed that the aircraft flew at v = 300 m/s. Using N
g
=5whichisthenumberof

Citations
More filters

Journal ArticleDOI
TL;DR: The aim of the present research is to develop procedures improving quality of images and extract the silhouette contour line of the ISAR image, the main geometric parameter in the target recognition process.
Abstract: Inverse synthetic aperture radar (ISAR) technique provides high-resolution images of observed moving objects However, shadow-occluded effects and noise, which results in lack of fractions of the object geometry, often corrupt ISAR images To facilitate the image analysis and to solve the target identification problem in most practical applications additional image-processing procedures are strongly required The aim of the present research is based on ISAR geometry and signal modelling to develop procedures improving quality of images and extract the silhouette contour line of the ISAR image, the main geometric parameter in the target recognition process The problem is solved in two main stages First, ISAR geometric parameters are described and expressions for linear frequency modulated signal modelling and imaging are derived Second, image-processing algorithms for ISAR image improving and contour line extraction are developed All stages of image-processing and reconstructing algorithms are illustrated by results of numerical experiments

22 citations


Proceedings ArticleDOI
Abdelmalek Toumi1, Ali Khenchaf1Institutions (1)
21 Mar 2016-
TL;DR: An approach to achieve the aircraft target recognition task using Inverse Synthetic Aperture Radar (ISAR) images reconstructed using IFFT (Inverse Fast Furrier Transform) and MUSIC2D (multiple signal characterization) methods.
Abstract: We present in this paper an approach to achieve the aircraft target recognition task using Inverse Synthetic Aperture Radar (ISAR) images reconstructed using IFFT (Inverse Fast Furrier Transform) and MUSIC2D (multiple signal characterization) methods. The first goal of this work is to propose efficient features for target recognition using a polar mapping procedure, to make a polar signature, with well-designed classifier. The second goal of this work is to compare the recognition rate using the both reconstruction methods. Finally, the obtained results are given and compared for the two types ISAR images in order to validate our proposed feature vectors and retrieval scheme.

14 citations


Journal ArticleDOI
Andon Lazarov1, Ch.N. Minchev2Institutions (2)
TL;DR: Stages of the image processing algorithms intended for image improving and contour line extraction are discussed and an algorithm for target recognition is developed based on neural network architecture.
Abstract: Abstract The image recognition and identification procedures are comparatively new in the scope of ISAR (Inverse Synthetic Aperture Radar) applications and based on specific defects in ISAR images, e.g., missing pixels and parts of the image induced by target’s aspect angles require preliminary image processing before identification. The present paper deals with ISAR image enhancement algorithms and neural network architecture for image recognition and target identification. First, stages of the image processing algorithms intended for image improving and contour line extraction are discussed. Second, an algorithm for target recognition is developed based on neural network architecture. Two Learning Vector Quantization (LVQ) neural networks are constructed in Matlab program environment. A training algorithm by teacher is applied. Final identification decision strategy is developed. Results of numerical experiments are presented.

8 citations


Cites methods from "Automatic Recognition of ISAR Image..."

  • ...Automatic target recognition of ISAR images of multiple targets and ATR results, and algorithms for automatic ISAR images recognition and classification are presented in [4-6]....

    [...]


Journal ArticleDOI
15 Nov 2018-Remote Sensing
TL;DR: A new method for phase error correction on the basis of the phase retrieval algorithm and the Gabor wavelet transform (GWT) is proposed, which does not need to estimate any motion parameters of the maneuvering target and is very effective for an ideal range–Doppler processing.
Abstract: The imaging issue of a rotating maneuvering target with a large angle and a high translational speed has been a challenging problem in the area of inverse synthetic aperture radar (ISAR) autofocus imaging, in particular when the target has both radial and angular accelerations. In this paper, on the basis of the phase retrieval algorithm and the Gabor wavelet transform (GWT), we propose a new method for phase error correction. The approach first performs the range compression on ISAR raw data to obtain range profiles, and then carries out the GWT transform as the time-frequency analysis tool for the rotational motion compensation (RMC) requirement. The time-varying terms, caused by rotational motion in the Doppler frequency shift, are able to be eliminated at the selected time frame. Furthermore, the processed backscattered signal is transformed to the one in the frequency domain while applying the phase retrieval to run the translational motion compensation (TMC). Phase retrieval plays an important role in range tracking, because the ISAR echo module is not affected by both radial velocity and the acceleration of the target. Finally, after the removal of both the rotational and translational motion errors, the time-invariant Doppler shift is generated, and radar returned signals from the same scatterer are always kept in the same range cell. Therefore, the unwanted motion effects can be removed by applying this approach to have an autofocused ISAR image of the maneuvering target. Furthermore, the method does not need to estimate any motion parameters of the maneuvering target, which has proven to be very effective for an ideal range–Doppler processing. Experimental and simulation results verify the feasibility of this approach.

5 citations


Proceedings ArticleDOI
01 Jan 2019-
TL;DR: This case study explores the integration of automatic classification technology with human judgement for maritime surveillance using inverse synthetic aperture radar (ISAR) and considers how both human and machine classification capabilities are best integrated to result in more accurate performance while maintaining human control for decisions.
Abstract: Critical to the development of airborne human-machine surveillance systems is understanding how they perform individually and how to optimally integrate them. We present a case study that explores the integration of automatic classification technology with human judgement for maritime surveillance using inverse synthetic aperture radar (ISAR). We have conducted this study by combining methods from human factors research and constructive simulation. Accordingly, we first performed a forced choice reaction time experiment to objectively quantify human ISAR classification performance which revealed important differences in performance as a function of vessel type. One of the results we present is the asymmetrical confusion matrices that describe the accuracy of observer classification performance.By reviewing current automatic ISAR vessel classification performance, we used algorithms to produce machine classification confusion matrices and subsequently compared them with human performance. Machine performance generally outperformed humans on the common ISAR images used in both studies. Finally, we considered how both human and machine classification capabilities (as measured via their corresponding confusion matrices) are best integrated to result in more accurate performance while maintaining human control for decisions. We investigated two models for combining human and machine classification and tested them using a discrete event simulation experiment. Our findings suggest that the optimal combination depends on the search priorities, for timeliness it is when automatic machine classification is used to pre-filter targets by their Perceptual Class, leaving the Gross class to be classified using human judgement. If accuracy is the dominant priority then automation should be applied for the Gross Naval classification.

3 citations


References
More filters

Book
01 Jan 2002-
TL;DR: This work presents a meta-analysis of radar electronic backscattering radar signals radar ambiguity function and matched filter synthetic aperture radar imaging, which highlights the importance of time-frequency transforms for radar applications.
Abstract: Introduction: radar electronic backscattering radar signals radar ambiguity function and matched filter synthetic aperture radar imaging Time-frequency transforms: linear transforms bilinear transforms Detection and extraction of signal noise: time-varying frequency filtering SNR improvement in time-frequency domain CFAR detection in joint time-frequency domain signal extraction in joint time-frequency domain Time-frequency analysis of radar range profiles: electromagnetic phenomenology embedded in backscattered data time-frequency representation of range profiles application of high-resolution time-frequency techniques to scattering data extraction of dispersive scattering features from radar imagery using time-frequency processing Time-frequency based image formation: radar imaging of moving targets motion compensation time-frequency based image formation radar imaging of manoeuvring targets radar imaging of multiple targets Motion compensation in inverse synthetic aperture radar imaging using time-frequency techniques: motion compensation algorithms time-frequency based motion compensation motion compensation examples of simulated and measured data presence of three-dimensional target motion Synthetic aperture radar imaging of moving target: radar signatures of moving targets effect of target motion on SAR imaging detection and imaging of moving targets SAR imaging of moving targets using time-frequency transforms Time-frequency transform for analysis of micro-doppler phenomena: vibration induced micro-doppler rotation induced micro-doppler examples Trends in time-frequency transforms for radar applications

779 citations


"Automatic Recognition of ISAR Image..." refers background in this paper

  • ...Efficient algorithms such as range-Doppler [10] and time frequency analysis [11] were proposed to image a single target, but when multiple targets occur in a single radar beam, imaging each target separately is a difficult task....

    [...]


Journal ArticleDOI
Abstract: Imaging from ground-based (stationary) radars of moving targets is often possible by utilizing a "synthetic aperture" developed from the target motion itself. The theory and experimental results associated with such processing are addressed. An aircraft is imaged from both a straight flight and a turn with recognizable results. Analysis shows that two-phase components exist in the radar return, one being gross velocity induced, the other being interscatterer interference within the target itself. The former phase must be removed prior to imaging and techniques are developed for this task. Preprocessing, range curvature, range alignment, motion compensation, and presumming are all addressed prior to presenting the experimental results. Coherence processing intervals, range collapsing, and range realignment are all examined during the processing aspects of the paper.

678 citations


"Automatic Recognition of ISAR Image..." refers background in this paper

  • ...Efficient algorithms such as range-Doppler [10] and time frequency analysis [11] were proposed to image a single target, but when multiple targets occur in a single radar beam, imaging each target separately is a difficult task....

    [...]


Journal ArticleDOI
01 Jan 1980-

456 citations


Journal ArticleDOI
TL;DR: The nature of ISAR imaging of ships, and single-frame and multiple-frame techniques for segmentation, feature extraction, and classification are described and results are shown which illustrate a capability for automatic recognition of ISar ship imagery.
Abstract: Inverse synthetic aperture radar (ISAR) produces images of ships at sea which human operators can be trained to recognize. Because ISAR uses the ship's own varying angular motions (roll, pitch, and yaw) for cross-range resolution, the viewing aspect and cross-range scale factor are continually changing on time scales of a few seconds. This and other characteristics of ISAR imaging make the problem of automatic recognition of ISAR images quite distinct from the recognition of optical images. The nature of ISAR imaging of ships, and single-frame and multiple-frame techniques for segmentation, feature extraction, and classification are described. Results are shown which illustrate a capability for automatic recognition of ISAR ship imagery.

166 citations


"Automatic Recognition of ISAR Image..." refers background in this paper

  • ...Due to its applicability regardless of weather and illumination conditions, it is used in many military applications such as classification of non-cooperative aircraft, and battlefield awareness [8, 9]....

    [...]


Journal ArticleDOI
Kyung-Tae Kim1, Dong-Kyu Seo2, Hyo-Tae Kim2Institutions (2)
TL;DR: This approach can provide efficient features for classification by the combined use of a polar mapping procedure and a well-designed classifier and the resulting feature vectors are able to meet the requirements that efficient features should have: invariance with respect to rotation and scale, small dimensionality, as well as highly discriminative information.
Abstract: In this paper, we propose a method to classify inverse synthetic aperture radar images from different targets. Our approach can provide efficient features for classification by the combined use of a polar mapping procedure and a well-designed classifier. The resulting feature vectors are able to meet the requirements that efficient features should have: invariance with respect to rotation and scale, small dimensionality, as well as highly discriminative information. Typical experimental examples of the proposed method are provided and discussed.

111 citations


"Automatic Recognition of ISAR Image..." refers methods in this paper

  • ...We use the polar mapping classifier (PMC) which is invariant to variation in the scale and the translation of the ISAR image [16]....

    [...]


Performance
Metrics
No. of citations received by the Paper in previous years
YearCitations
20191
20181
20172
20162