Convolutional neural networks (CNN) have been successfully employed to tackle several remote sensing tasks such as image classification and show better performance than previous techniques. For the radar imaging community, a natural question is: Can CNN be introduced to radar imaging and enhance its performance? The presented letter gives an affirmative answer to this question. We firstly propose a processing framework by which a complex-valued CNN (CV-CNN) is used to enhance radar imaging. Then we introduce two modifications to the CV-CNN to adapt it to radar imaging tasks. Subsequently, the method to generate training data is shown and some implementation details are presented. Finally, simulations and experiments are carried out, and both results show the superiority of the proposed method on imaging quality and computational efficiency.
Deep Dive into Enhanced Radar Imaging Using a Complex-valued Convolutional Neural Network.
Convolutional neural networks (CNN) have been successfully employed to tackle several remote sensing tasks such as image classification and show better performance than previous techniques. For the radar imaging community, a natural question is: Can CNN be introduced to radar imaging and enhance its performance? The presented letter gives an affirmative answer to this question. We firstly propose a processing framework by which a complex-valued CNN (CV-CNN) is used to enhance radar imaging. Then we introduce two modifications to the CV-CNN to adapt it to radar imaging tasks. Subsequently, the method to generate training data is shown and some implementation details are presented. Finally, simulations and experiments are carried out, and both results show the superiority of the proposed method on imaging quality and computational efficiency.
1
Abstract—Convolutional neural networks (CNN) have been
successfully employed to tackle several remote sensing tasks such
as image classification and show better performance than
previous techniques. For the radar imaging community, a natural
question is: Can CNN be introduced to radar imaging and
enhance its performance? The presented letter gives an
affirmative answer to this question. We firstly propose a
processing framework by which a complex-valued CNN
(CV-CNN) is used to enhance radar imaging. Then we introduce
two modifications to the CV-CNN to adapt it to radar imaging
tasks. Subsequently, the method to generate training data is
shown and some implementation details are presented. Finally,
simulations and experiments are carried out, and both results
show the superiority of the proposed method on imaging quality
and computational efficiency.
Index
Terms—Convolutional
neural
network
(CNN),
Complex-valued convolutional neural network (CV-CNN), Radar
imaging, Super resolution, Side-lobe reduction.
I. INTRODUCTION
MPROVING radar imaging quality using signal processing
techniques under given hardware platforms is an appealing
topic and has attracted much attention [1],[2]. In the recent
decade, compressive sensing [3] has greatly inspired the
research on sparsity-driven radar imaging techniques [4],[5].
Typically, the following linear model is adopted
1
1
1
,
,
,
,
M
M
N
M N
y
Ax
n y
n
x
A
(1)
where y is the echo signal, n is the additive noise, x
represents the image to be estimated and A is called the
sensing matrix or the dictionary. According to (1), conventional
imaging can be expressed as
H
ˆ
x
A y , which can be achieved
by back projection algorithm (BPA) or by Fast Fourier
Transform (FFT). However, either methods suffer from limited
resolution, high side-lobes and strong speckle. Sparsity-driven
methods impose prior constrains to x and the imaging
problem is transformed into an optimization problem
2
2
2
ˆ
argmin
s.t.
x
x
x
y
Ax
(2)
where
x
contains the prior knowledge on x ,
2
represents the energy of n which is always unknown. The most
widely used prior is the 1l -norm constraint in the image domain,
i.e.
1
x
x . The constraint is also usually imposed to the
transformed domain, i.e.
1
x
Tx
, where T can be the
wavelet transform, the derivative operator and so on. More
complicated
x s which take the inner coherence of x into
consideration are also proposed [6]. It can be expected that one
can obtain better ˆx by designing more sophisticated
x
and developing more effective optimization algorithms.
Although sparsity-driven methods can improve imaging
quality remarkably, it faces great challenges. Firstly, they are
too time-consuming to achieve real time imaging. Different
from conventional linear imaging process, the problem in (2) is
nonlinear. Usually, a large amount of iterations are needed to
converge to a reasonable solution. Secondly, their stability and
robustness cannot easily be guaranteed. The imaging quality
promotion obtained by solving (2) is based on accurate
modeling. If A is inaccurate or the constraint
x
is
unsuitable, the results can be greatly degraded. Although
several methods have been proposed for these problems [7], it
is at the cost of heavier computational burdens.
In fact, the model in (1) and (2) is a general problem solving
framework. In many signal processing fields, e.g. image
processing or computer vision [8], this model is also widely
used. CNN is famous mostly for its cutting-edge performance
in image classification tasks [9]. Recently, more and more
researchers have extended and applied CNN to regression-type
problems such as superresolution or denoising [10],[11]. In
[10], the authors showed the close relationship between
regression-type CNN and sparsity-based methods. In [12], the
authors pointed out the similarity between the iterative
shrinkage methods and the feedforward process of CNN.
Moreover, CNN can learn from and be adaptive to the training
data, and its structure is highly parallel and free of iterations. As
a result, CNN can outperform sparsity-based methods in both
accuracy and efficiency.
On top of the above review, a reasonable question is whether
CNN can be applied to radar imaging. In our opinion, several
issues need to be addressed when applying CNN to radar
imaging. For example, what is the overall processing
framework? What is the input and output of the CNN? How can
one obtain the training data? Is CNN more effective than recent
imaging methods? In the following sections, we will introduce
our solutions to these issues.
II. METHODS
A. Overall framework of CV-CNN-enhanced radar imaging
The proposed processing framework is shown in Fig. 1, the
dashed and solid lines show the data flow in training and
imaging/testing period respectively. The
…(Full text truncated)…
This content is AI-processed based on ArXiv data.
