> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) <
1
Abstract—Electrical resistance tomography (ERT) is a sensing
technique for process monitoring and measurement, the image
reconstruction algorithms are required fast and accurate. The
OMP class algorithm has the property of reconstructing a higher
quality signal by fewer iteration steps (shorter elapsed time) and
faster convergence speed compared with other iterative
algorithms. But the classical OMP is not self-adapted and may fall
in to a local optimal solution. In this paper, a modified orthogonal
matching pursuit (MOMP) was presented by adding continuous
constraint to prevent local optimal solution and self-adaption
process on classical OMP algorithm which finds the optimal
number of iterations based on the information entropy of
measurement to make it suitable for ERT inverse problem. The
algorithm was experimentally validated versus other algorithms
through a 16-electrode ERT system. Comparing with other
algorithms, MOMP algorithm has a lower mean square error
than non-iterative algorithms and a shorter time cost than
iterative algorithms. Dynamic experiments showed that the
MOMP algorithm is promising for industrial process monitoring.
Index Terms—Electrical resistance tomography; Image
reconstruction; Compressed sensing; OMP algorithm
I. I
NTRODUCTION
LECTRICAL resistance tomography (ERT) is a novel
sensing technique [1-3], which attracted much attention in
medical and industrial imaging [4, 5]. Electrical resistance
tomography offers a complementary and low-cost solution to
improve the efficiency of the process and chemical industrial
equipment
.
The inverse problem of ERT is an ill-posed problem. So far,
the inverse problem of ERT was solved by many algorithms
which were divided into two main categories: iterative
algorithms and non-iterative algorithms. As an example of
non-iterative algorithms, filtered back projection (FBP) was
firstly presented and has been widely applied in X-ray
tomography, magnetic resonance imaging (MRI), electrical
magnetic tomography (EMT) [6]. However the measurement of
This work was supported by the National Natural Science Foundation of
China (No. 61571321 and No. 61473206) and The Natural Science Foundation
of Tianjin (No. 17JCZDJC38400).
The authors are with the Tianjin Key Laboratory of Process Measurement
and Control, School of Electrical and Information Engineering, Tianjin
University, Tianjin 300072, China (e-mail: tanchao@tju.edu.cn).
ERT was spatially sparse. The measurement data is also a
sparse signal which is proved by RIP criteria. Due to the sparse
character in 16-electrodes’ set-up and measurement data, the
FBP cannot deliver accurate results. Simplified from FBP
algorithm, linear back projection (LBP) method had been
widely used in many industrial applications till now since it
showed good anti-noise performance and robustness. Cheney et
al presented Newton’s One-Step Error Reconstructor method
(NOSER) [7]; Wang et al presented a one-step Landweber
method. They achieved off-line pre-iteration and on-line
one-step reconstruction. The reconstruction speed is markedly
improved [8]. However, when the real distribution was not
close to reference distribution, the non-iterative algorithms
suffered bad imaging precision which hindered the application
in many areas. Non-iterative algorithms are suitable for quick
measurement in industry [9-11].
Conventional iterative algorithms are regularization
algorithms, the regularization added additional constraints on
the inverse problem in order to solve an ill-posed problem or to
prevent overfitting. These iterative algorithms have higher
resolution than non-iterative algorithms but are
time-consuming which hindered the application in real industry.
Landweber iteration algorithm was introduced to solve the
inverse problem [12-14]. Another widely used iterative
algorithm is Tikhonov regularization algorithm [15, 16].
Some extended regularization algorithms emerged: hybrid
regularization [17]; regularization-homotopy algorithm [18];
adaptive regularization parameter choice method [29].
The speed of regularization is slow for most of the industrial
applications. Tikhonov regularization is 2-norm regularization
method. However, as a smooth and convex regularization, the
drawback of the Tikhonov regularization is that the image
edges cannot be preserved in the restoration process.
Recently, sparse reconstruction method was presented to
solve electrical tomography inverse problem. Ye et al added
sparse constraint on electrical capacitance tomography (ECT)
to improve the image quality [20]. Zhao et al studied Lp-Lq
norm regularization algorithm which applied on electrical
resistance tomography, when p is small, a large penalty is
implemented, and when p is large, a lesser penalty is
implemented [21]. However, the change in norm is not always
suitable to various distributions. How to choose a proper norm
p is a problem for norm regularization method [22, 23].
Compressed sensing algorithm is presented by Candes, Tao
Electrical Resistance Tomography Image
Reconstruction based on Modified OMP
Algorithm
Wei Zhang, Chao Tan*, Senior Member, IEEE, Yanbin Xu, Member, IEEE, and Feng Dong, Senior
Member, IEEE
E