IEEE Geoscience and Remote Sensing Magazine - December 2022 - 33

compression and synthetic aperture technologies for high
resolution in the range and azimuth dimensions, respectively
[2]. In a broad view, inverse SAR (ISAR) can be included in
SAR by treating it as one typical mode. Usually, imaging techniques
based on the MF are applied due to their simplicity
and stability. In fact, MF is restricted by the precondition of
Nyquist sampling and has certain shortcomings in imaging
performance, such as resolution limitations and inevitable
artifacts of the side lobe, speckle, and so on [3].
To enhance the performance of image interpretation,
one of the most important characteristic trends is resolution
improvement, which means a demand for large wideband
signal configuration. Meanwhile, modern SAR systems usually
move up together with the requirements of wide swath
[4]-[6], multifunction [2], and multidimension (i.e., multichannel/frequency/temporal
and so forth) [7], [8], [9], [10],
[11]. As a result, the complexity of the state-of-the-art radar
system increases dramatically such that it usually requires a
high sampling rate as well as large-scale data processing and
storage. These aforementioned challenges introduce significant
limitations and significantly hinder the development of
modern imaging radar.
Over the last few decades, new radar imaging technologies
have been invented to offer considerable achievements
during SAR image formation, such as superresolution [11]
[12], [13], [14], [15], [16], [17]; feature enhancement [18],
[19], [20], [21], [22], [23]; and image reconstruction from
missing or highly corrupted data [24], [25], [26], [27], [28],
[29]. Among these emerging methods, the most famous one
is sparse image reconstruction from CS [11], [30], [31], [32],
[33], [34], [35], [36], [37], which has a certain robustness to
the data quality and quantity. The developed CS theory tells
us that it can precisely reconstruct a sparse or compressible
signal from fewer measurements beyond Nyquist sampling
using sparse techniques [38], [39], [40].
Until now, the introduction of CS into SAR imaging has
greatly influenced the paradigms of imaging theory and
methodology, obtaining new system designs and novel imaging
algorithms [11], [20], [30], [31], [41]. In the scheme of system
design, various interesting studies have been introduced,
mainly focusing on the scenarios of waveform design and
sensing geometry, which directly determine the measurement
matrix of data collection with compressive sampling. Built
upon sparse reconstruction, radar imaging incorporating a
sparse prior has produced a new formulation of SAR imaging
methodology, known as sparse SAR imaging. Profiting from
all of these efforts, CS has achieved remarkable progress and
found a wide application in the SAR community, including
SAR/ISAR [20], [26], [31], [42]; interferometric SAR/ISAR (InSAR/InISAR)
[27], [43], [44], [45]; SAR ground moving target
indication (SAR GMTI) [46], [47]; tomographic SAR (TomoSAR)
[11], [15], [16], [17], [48], [49]; array SAR/ISAR [50],
[51], [52], [53], [54]; bi-/multistatic SAR [55]; and others [56].
Figure 1 shows the number of papers and articles published
on this topic. It can be seen that there has been obvious
growth since 2006 to achieve a maximum of more than 150
DECEMBER 2022 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
papers and articles in 2016, indicating that sparse SAR imaging
is a hot topic. However, there has been no growth and
even an obvious decline in the number of publications since
2016, indicating the relatively mature and bottleneck status
of this research. In fact, although much success has been observed
in applying CS to SAR imaging, there are still several
challenges in the current status, limiting the practical application
of sparse SAR imaging. Objectively speaking, the sparse
representation of SAR images is still an open question due to
unique characteristics [57], such as random phase and speckle
nature [20], [30]. In addition, the computational complexity
of CS is usually very large because an iterative solution is
necessary to find the optimal imaging results. In this case, it is
too time-consuming for real-time imaging. Due to the current
limitations, further study is needed to establish a compelling
architecture and practical solutions for sparse SAR imaging.
Recently, ML technology [58], [59], [60], [61], [62], [63],
[64], [65], including DL, which can be treated as a continuation
and extension of CS, has been introduced to sparse SAR
imaging. There are several emerging established algorithms
that show tremendous potential. Typically, the low-rank constraint
methods [66], [67], [68], [69], [70], [71], [72] have
shown some superiorities in sparse SAR imaging. In contrast
to sparsity enforced by CS, the low-rank property is adopted
in a data-adaptive manner, which is effective in preserving the
data structure and capturing latent components. In particular,
the low-rank method takes advantage of handling multidimensional
data due to the feasibility of dimension reduction.
Moreover, inspired by DL, researchers have investigated
deep neural networks (DNNs) [57], [58], [59], [61], [62],
[64], [73], [74], [75], [76], [77], [78], which are capable of
learning image priors from training data, for sparse SAR imaging.
Due to the parallel structure and free iteration, DL can
provide more efficient and stable results than traditional CS.
As with the developments of CS, the ML/DL-based methods
have shown significant potential and can be considered as
good candidates for a new generation of sparse SAR imaging.
Figures 2 and 3 display the publication statistics for lowrank
and DL methods, respectively. There has been obvious
research growth for these two methods during recent years,
indicating that sparse SAR imaging is taking off.
180
160
140
120
100
80
60
40
20
155
129
119
75 80
78
30
21
3 3 4 6
151
137
122
FIGURE 1. The number of publications on CS-based methods.
33
Total Number of Publications
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
IEEE Geoscience and Remote Sensing Magazine - December 2022

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - December 2022
