# American Institute of Mathematical Sciences

April  2017, 11(2): 373-401. doi: 10.3934/ipi.2017018

## Multiwave tomography with reflectors: Landweber's iteration

 Department of Mathematics, Purdue University, West Lafayette, IN 47907, USA

*First author partly supported by NSF Grant DMS-1301646

Received  April 2016 Revised  November 2016 Published  March 2017

We use the Landweber method for numerical simulations for the multiwave tomography problem with a reflecting boundary and compare it with the averaged time reversal method. We also analyze the rate of convergence and the dependence on the step size for the Landweber iterations on a Hilbert space.

Citation: Plamen Stefanov, Yang Yang. Multiwave tomography with reflectors: Landweber's iteration. Inverse Problems & Imaging, 2017, 11 (2) : 373-401. doi: 10.3934/ipi.2017018
##### References:

show all references

##### References:
The unstable case, reconstructions with the Landweber method with noise. Left: unfiltered data. Right: filtered data.
$\log_{10}$ error at each step vs. $\gamma$ with exact data. This is the graph of $\log_{10}\| K\|$ as a function of $\gamma$ for $\mu^2=1$, $\|\mathcal{L}\|^2=20$. The optimal $\gamma^*$ is $\gamma^*=2/21\approx0.0952$. The error increases fast on the right of $\gamma^*$.
Error vs. $\gamma$ for the Landweber iterations in the stable example. Plotted are errors after $10$ iterations (boxes), after $30$ iterations (diamonds) and after $50$ ones (dots). Right: Error vs. the number of the iterations. The bottom curve with the circles are the ATR errors. The other curves correspond to $\gamma$ ranging from $0.1$ to $0.06$, as on the left, starting form the top to the bottom in the top left corner.
The functions $g_N(\lambda)$ with $\gamma=1$ and $N=5, 20, 40, 80$. As the number of iterations $N$ increases, the maximum increases as $C_1\sqrt N$ and its location shifts to the left to $\lambda_N=C_2/\sqrt N$.
A stable case. The smooth increasing curve represents the eigenvalues of $\mathcal{L}^*\mathcal{L}$ in the discrete realization on an $101\!\times\!101$ grid restricted to a $94\!\times\!94$ grid. There are small imaginary parts along the horizontal line in the middle. $\mathcal{L}^*$ is computed as the adjoint to the matrix representation of $\mathcal{L}$. The rough curve represents the squares of the Fourier coefficients of the SL phantom.
The eigenfunctions of $L^*L$ corresponding to $\lambda_1$, $\lambda_{1,000}$ and $\lambda_{7,500}$
Error plots in the stable case with data not in the range. Approximate convergence.
Reconstruction and error plots in the stable case with noise with $200$ iterations; $\gamma=0.01,0.02,0.03,0.04$. A slow divergence. For $\gamma=0.05$ we get a fast divergence.
Speed $1$ with $T=1.8$ with invisible singularities on the top. From left to right: (a) ATR with 50 steps, a cut near the border, 34\% error. (b) Landweber with a cut near the borders, error $31\%$; (c) Landweber without a cut near the borders, error $36\%$.
Error vs. $\gamma$ in the unstable case. Left: errors after $10$ iterations (boxes), after $30$ iterations (diamonds) and after $50$ ones (dots). The vertical axis is on a $\log_{10}$ scale and the horizontal axis represents $\gamma$. The lower curve, corresponding to 50 iterations, is flatter than in the stable case. On the right, the curve with the diamond marks is the error with the ATR method.
The unstable case. The smooth increasing curve represents the eigenvalues of $\mathcal{L}^*\mathcal{L}$ in the discrete realization on an $101\!\times\!101$ grid restricted to a $94\!\times\!94$ grid. $\mathcal{L}^*$ is computed as the adjoint matrix rather than with the wave equation solver. The rough curve represents the squares of the Fourier coefficients of the SL phantom. The arrows indicate the effective lower bound of the power spectrum of the SL phantom, and the first positive eigenvalue modulo small errors, respectively.
A unstable case with Gaussian noise. Left: error curves with $\gamma$ ranging from $0.03$ to $0.17$ (the diverging curve). The critical value of $\gamma$ looks close to that in the zero noise case in Figure 9. The iterations for $\gamma\le 0.16$ diverge slowly in contrast with the noise free case in Figure 9. Right: Filtered data, $\gamma\le 0.15$. The errors look like in Figure 9.
Reconstructions with a discontinuous speed, plotted on the left. The original is black and white disks on a gray background. Data on the marked part of the boundary, $T=4$. Center: ATR reconstruction, error $1\%$. Right: Landweber reconstruction, error $1.5\%$.
Left: the original SL phantom. Right: The reconstructed SL phantom with $c=1$ and $T=2$. The phantom was originally on a $201\times 201$ grid with $\Delta t=\Delta x /\sqrt{2}$. The data was computed on a grid rescaled by a factor of $5.7$ in the spatial variables and $7.41$ times in the time variable; and then rescaled to the the original one on the boundary before inversion. The ripple artifacts are due to high frequency waves in the time reversal moving slower than the speed $c=1$.
Same situation as in Figure 14 but the phantom contains Gaussians with predominately low frequency content. Left: original. Right: reconstruction with 50 iterations. The relative $L^2$ error is $1.8\%$, the $L^\infty$ error is $3.5\%$.
 [1] Daniela Calvetti, Erkki Somersalo. Microlocal sequential regularization in imaging. Inverse Problems & Imaging, 2007, 1 (1) : 1-11. doi: 10.3934/ipi.2007.1.1 [2] Linh V. Nguyen. A family of inversion formulas in thermoacoustic tomography. Inverse Problems & Imaging, 2009, 3 (4) : 649-675. doi: 10.3934/ipi.2009.3.649 [3] James W. Webber, Sean Holman. Microlocal analysis of a spindle transform. Inverse Problems & Imaging, 2019, 13 (2) : 231-261. doi: 10.3934/ipi.2019013 [4] Victor Palamodov. Remarks on the general Funk transform and thermoacoustic tomography. Inverse Problems & Imaging, 2010, 4 (4) : 693-702. doi: 10.3934/ipi.2010.4.693 [5] Leonid Kunyansky. Fast reconstruction algorithms for the thermoacoustic tomography in certain domains with cylindrical or spherical symmetries. Inverse Problems & Imaging, 2012, 6 (1) : 111-131. doi: 10.3934/ipi.2012.6.111 [6] Venkateswaran P. Krishnan, Eric Todd Quinto. Microlocal aspects of common offset synthetic aperture radar imaging. Inverse Problems & Imaging, 2011, 5 (3) : 659-674. doi: 10.3934/ipi.2011.5.659 [7] Raluca Felea, Romina Gaburro, Allan Greenleaf, Clifford Nolan. Microlocal analysis of Doppler synthetic aperture radar. Inverse Problems & Imaging, 2019, 13 (6) : 1283-1307. doi: 10.3934/ipi.2019056 [8] Ennio Fedrizzi. High frequency analysis of imaging with noise blending. Discrete & Continuous Dynamical Systems - B, 2014, 19 (4) : 979-998. doi: 10.3934/dcdsb.2014.19.979 [9] Xuzhou Chen, Xinghua Shi, Yimin Wei. The stationary iterations revisited. Numerical Algebra, Control & Optimization, 2013, 3 (2) : 261-270. doi: 10.3934/naco.2013.3.261 [10] Fanwen Meng, Kiok Liang Teow, Kelvin Wee Sheng Teo, Chee Kheong Ooi, Seow Yian Tay. Predicting 72-hour reattendance in emergency departments using discriminant analysis via mixed integer programming with electronic medical records. Journal of Industrial & Management Optimization, 2019, 15 (2) : 947-962. doi: 10.3934/jimo.2018079 [11] Caifang Wang, Tie Zhou. The order of convergence for Landweber Scheme with $\alpha,\beta$-rule. Inverse Problems & Imaging, 2012, 6 (1) : 133-146. doi: 10.3934/ipi.2012.6.133 [12] Grzegorz Graff, Piotr Nowak-Przygodzki. Fixed point indices of iterations of $C^1$ maps in $R^3$. Discrete & Continuous Dynamical Systems - A, 2006, 16 (4) : 843-856. doi: 10.3934/dcds.2006.16.843 [13] David Maxwell. Kozlov-Maz'ya iteration as a form of Landweber iteration. Inverse Problems & Imaging, 2014, 8 (2) : 537-560. doi: 10.3934/ipi.2014.8.537 [14] Dana Paquin, Doron Levy, Lei Xing. Multiscale deformable registration of noisy medical images. Mathematical Biosciences & Engineering, 2008, 5 (1) : 125-144. doi: 10.3934/mbe.2008.5.125 [15] Guillaume Bal, Olivier Pinaud, Lenya Ryzhik. On the stability of some imaging functionals. Inverse Problems & Imaging, 2016, 10 (3) : 585-616. doi: 10.3934/ipi.2016013 [16] Jan Boman, Vladimir Sharafutdinov. Stability estimates in tensor tomography. Inverse Problems & Imaging, 2018, 12 (5) : 1245-1262. doi: 10.3934/ipi.2018052 [17] Tim Kreutzmann, Andreas Rieder. Geometric reconstruction in bioluminescence tomography. Inverse Problems & Imaging, 2014, 8 (1) : 173-197. doi: 10.3934/ipi.2014.8.173 [18] Plamen Stefanov, Wenxiang Cong, Ge Wang. Modulated luminescence tomography. Inverse Problems & Imaging, 2015, 9 (2) : 579-589. doi: 10.3934/ipi.2015.9.579 [19] Benjamin Contri. Fisher-KPP equations and applications to a model in medical sciences. Networks & Heterogeneous Media, 2018, 13 (1) : 119-153. doi: 10.3934/nhm.2018006 [20] Xiangmin Zhang. User perceived learning from interactive searching on big medical literature data. Big Data & Information Analytics, 2017, 2 (5) : 1-16. doi: 10.3934/bdia.2017019

2018 Impact Factor: 1.469