-
Notifications
You must be signed in to change notification settings - Fork 0
/
2024-02-04.html
executable file
·315 lines (260 loc) · 22.6 KB
/
2024-02-04.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta http-equiv="X-UA-Compatible" content="IE=Edge">
<meta name="description" content="">
<meta name="keywords" content="">
<meta name="author" content="">
<meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=1">
<title>The linear sampling method for small random scatterers</title>
<!--
Template 2085 Neuron
http://www.tooplate.com/view/2085-neuron
-->
<link rel="stylesheet" href="css/bootstrap.min.css">
<link rel="stylesheet" href="css/all.css">
<link rel="stylesheet" href="css/magnific-popup.css">
<!-- Main css -->
<link rel="stylesheet" href="css/style.css">
<link href="https://fonts.googleapis.com/css?family=Lora|Merriweather:300,400" rel="stylesheet">
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=UA-108246943-1"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'UA-108246943-1');
</script>
<!-- Google AdSense -->
<script async src="https://pagead2.googlesyndication.com/pagead/js/adsbygoogle.js?client=ca-pub-8299683413193339" crossorigin="anonymous"></script>
<!--<script async src="https://fundingchoicesmessages.google.com/i/pub-8299683413193339?ers=1" nonce="AuY4vKpDElA9N7YwmFfMjg"></script>
<script nonce="AuY4vKpDElA9N7YwmFfMjg">(function() {function signalGooglefcPresent() {if (!window.frames['googlefcPresent']) {if (document.body) {const iframe = document.createElement('iframe'); iframe.style = 'width: 0; height: 0; border: none; z-index: -1000; left: -1000px; top: -1000px;'; iframe.style.display = 'none'; iframe.name = 'googlefcPresent'; document.body.appendChild(iframe);} else {setTimeout(signalGooglefcPresent, 0);}}}signalGooglefcPresent();})();</script>-->
</head>
<body>
<!-- PRE LOADER -->
<div class="preloader">
<div class="sk-spinner sk-spinner-wordpress">
<span class="sk-inner-circle"></span>
</div>
</div>
<!-- Navigation section -->
<div class="navbar navbar-default navbar-static-top" role="navigation">
<div class="container">
<div class="navbar-header">
<button class="navbar-toggle" data-toggle="collapse" data-target=".navbar-collapse">
<span class="icon icon-bar"></span>
<span class="icon icon-bar"></span>
<span class="icon icon-bar"></span>
</button>
<a href="index.html" class="navbar-brand">H. Montanelli</a>
</div>
<div class="collapse navbar-collapse">
<ul class="nav navbar-nav navbar-right">
<li><a href="about.html">About</a></li>
<li><a href="blog.html">Blog</a></li>
<li><a href="publications.html">Publications</a></li>
<li><a href="research.html">Research</a></li>
<li><a href="students.html">Students</a></li>
<li><a href="talks.html">Talks</a></li>
<li><a href="teaching.html">Teaching</a></li>
<li><a href="https://www.paypal.com/donate/?hosted_button_id=UCLCSJLFL433E"><b>Donate</b></a></li>
</ul>
</div>
</div>
</div>
<!-- Home Section -->
<section id="home" class="main-about parallax-section">
<div class="overlay"></div>
<div class="container">
<div class="row">
<div class="col-md-12 col-sm-12">
<h1>The linear sampling method for small random scatterers</h1>
</div>
</div>
</div>
</section>
<!-- Blog Single Post Section -->
<section id="about">
<div class="container">
<div class="row">
<div class="col-md-offset-1 col-md-10 col-sm-12">
<p><i>February 4, 2024 — Support my next blog post, <a href="https://www.paypal.com/donate/?hosted_button_id=UCLCSJLFL433E">buy me a coffee</a> ☕.</i></p>
<p>In this post, I talk about an extension of the linear sampling method (LSM) for solving the sound-soft <a href="https://en.wikipedia.org/wiki/Inverse_scattering_problem">inverse acoustic scattering</a> problem with data generated by randomly distributed small scatterers. For details, check out my <a href="https://arxiv.org/pdf/2403.19482.pdf">paper</a> with <a href="https://perso.ensta-paris.fr/~haddar/">Houssem Haddar</a> and <a href="https://josselin-garnier.org/">Josselin Garnier</a>!</p>
<h2>Introduction</h2>
<p>Inverse scattering problems arise across a multitude of fields, ranging from medical imaging and <a href="https://en.wikipedia.org/wiki/Nondestructive_testing">non-destructive testing</a> to <a href="https://en.wikipedia.org/wiki/Radar">radar</a> technology and <a href="https://en.wikipedia.org/wiki/Seismology">seismology</a>. At their core, these problems involve the task of deducing the characteristics of an object or medium from the scattered signals it generates. These problems involve several challenges, including nonlinearity, which is particularly pronounced near <a href="https://en.wikipedia.org/wiki/Resonance">resonance</a>, making linearization inapplicable; these are also severely <a href="https://en.wikipedia.org/wiki/Well-posed_problem">ill-posed</a>, raising questions about uniqueness and stability, and necessitating the inclusion of regularization. Additionally, for <a href="https://en.wikipedia.org/wiki/Iterative_method">iterative</a> procedures, reconstruction time can be lengthy and prior knowledge is required for initiation. In this context, fast, data-driven algorithms, such as the LSM, can be useful in reducing computational costs, and initializing more sophisticated algorithms.</p>
<p>Two acquisition configurations are commonly used in inverse scattering problems—active and passive. Active imaging involves sending waves through <i>controlled</i> sources and recording the medium's response through <i>controlled</i> sensors. Passive imaging, on the other hand, employs controlled sensors but relies on <i>random</i>, <i>uncontrolled</i> sources (such as <a href="https://en.wikipedia.org/wiki/Microseism">microseisms</a> and <a href="https://en.wikipedia.org/wiki/Swell_(ocean)">ocean swells</a> in seismology). In this setup, it is the <a href="https://en.wikipedia.org/wiki/Cross-correlation">cross-correlations</a> between the recorded signals that convey information about the medium. Passive imaging is a rapidly growing research topic because it enables the imaging of areas where the use of active sources is not possible due to, e.g., safety or environmental reasons. Additionally, even in scenarios where active sources could be employed, utilizing passive sources helps reduce operational costs and enhances <a href="https://en.wikipedia.org/wiki/Stealth_technology">stealth</a> in defense applications.</p>
<p>In a previous paper, we introduced an extension of the LSM to address the sound-soft inverse scattering problem involving random sources. This current study builds upon that foundation, focusing on a scenario where a single <i>controlled</i> source, operating at a specified wavelength \(\lambda\), illuminates a small <i>random</i> scatterer and an object of comparable size to \(\lambda\)—the goal is to reconstruct the shape of the latter. The reflection of the wave field transmitted by the point source on the small scatterer serves as a random source, aligning with the context explored in our earlier work. </p>
<p>The motivation for this research stems from the necessity, when imaging an unknown obstacle \(D\), to illuminate it with a diverse set of incoming waves. For deterministic, controlled point sources, one considers a family of sources located at points \(\boldsymbol{z}_m\) generating the incident fields
$$
\begin{align}
\phi(\boldsymbol{x},\boldsymbol{z}_m) = \frac{i}{4}H_0^{(1)}(k\vert\boldsymbol{x}-\boldsymbol{z}_m\vert),
\end{align}
$$
and measures the resulting scattered fields \(u^s(\boldsymbol{x}_j,\boldsymbol{z}_m)\) at points \(\boldsymbol{x}_j\) to populate the near-field matrix \(N\) with entries \(N_{jm} = u^s(\boldsymbol{x}_j,\boldsymbol{z}_m)\). In the case of random, uncontrolled sources at positions \(\boldsymbol{z}_\ell\), as demonstrated in our prior work, the relevant matrix is the cross-correlation matrix \(C\). Its entries are given by:
$$
\begin{align}
C_{jm} = \frac{2ik\vert\Sigma\vert}{L}\sum_{\ell=1}^L\overline{u(\boldsymbol{x}_j,\boldsymbol{z}_\ell)}u(\boldsymbol{x}_m,\boldsymbol{z}_\ell) - \left[\phi(\boldsymbol{x}_j,\boldsymbol{x}_m) - \overline{\phi(\boldsymbol{x}_j,\boldsymbol{x}_m)}\right],
\end{align}
$$
where \(u(\boldsymbol{x},\boldsymbol{z})=\phi(\boldsymbol{x},\boldsymbol{z})+u^s(\boldsymbol{x},\boldsymbol{z})\) is the total field for the incident wave \(\phi(\boldsymbol{x},\boldsymbol{z})\), and \(\vert\Sigma\vert\) is the area of the surface \(\Sigma\) on which the random sources are distributed.</p>
<p>Suppose now that we only have a single <i>controlled</i> source located at some given point \(\boldsymbol{z}\), as opposed to several positions \(\boldsymbol{z}_m\). This is insufficient for reconstructing the obstacle's shape with the LSM (the near-field matrix \(N\) would have rank one). To address this limitation, we propose introducing a random medium between the source and the obstacle. We illustrate this concept, in acoustic scattering, by considering a single small random scatterer between the source and the obstacle—a seemingly simple yet powerful model of a random medium that allows us to apply the LSM in a novel manner, with a modified version of \(C\).</p>
<h2>Modified Helmholtz–Kirchoff identity</h2>
<p>Consider the incident field \(\phi(\boldsymbol{x},\boldsymbol{z}_\epsilon)\) generated by a point source located at \(\boldsymbol{z}_\epsilon=\lambda\epsilon^{-q}e^{i\theta_z}\) for some scalars \(\epsilon>0\), \(q>0\), and \(\theta_z\in[0,2\pi]\). Here, \(k>0\) is the wavenumber and \(\lambda=2\pi/k\) is the wavelength. Let \(D\) be an obstacle of size proportional to \(\lambda\) and independent on \(\epsilon\). Without loss of generality, we assume that \(D\) is centered at the origin. We also consider a small disk \(D_\epsilon=D_\epsilon(\boldsymbol{y}_\epsilon)\) of radius \(\rho(D_\epsilon)=\lambda\epsilon\) centered at \(\boldsymbol{y}_\epsilon = \lambda\epsilon^{-p}e^{i\theta_y}\) for some scalars \(0 < p < q\) and \(\theta_y\in[0,2\pi]\). Finally, let \(B\subset\mathbb{R}^2\setminus\overline{D\cup D_\epsilon}\) be a compact set whose size and distance to \(D\) are proportional to \(\lambda\) and independent on \(\epsilon\). Measurements will be taken inside the volume \(B\), which is consistent with our previous work.</p>
<p>We examine the scattering of the incident field \(\phi(\cdot,\boldsymbol{z}_\epsilon)\) by \(D\) and \(D_\epsilon\), which generates the scattered field \(w^s_\epsilon\). More precisely, let \(w^s_\epsilon(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon)\) be the solution to the sound-soft scattering problem
$$
\begin{align}\label{eq:w^s}
\left\{
\begin{array}{ll}
\Delta w^s_\epsilon(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) + k^2 w^s_\epsilon(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) = 0 \quad \text{in $\mathbb{R}^2\setminus\{\overline{D\cup D_\epsilon}\}$}, \\[0.4em]
w^s_\epsilon(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) = -\phi(\cdot,\boldsymbol{z}_\epsilon) \quad \text{on $\partial D\cup\partial D_\epsilon$}, \\[0.4em]
\text{$w^s_\epsilon(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon)$ is radiating}.
\end{array}
\right.
\end{align}
$$
We showed that \(w^s_\epsilon(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon)\) may be approximated as the sum of three terms,
$$
\begin{align}
w_\epsilon^s(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) \approx u^s(\cdot,\boldsymbol{z}_\epsilon) + v_\epsilon^i(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) + v_\epsilon^s(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon)
\end{align}
$$
with an error \(\mathcal{O}(\vert\log\epsilon\vert^{-1}\epsilon^p\epsilon^{q/2})\) in the \(H^1(B)\)-norm; the scattered fields in the expansion above are the solutions to the sound-soft scattering problems
$$
\begin{align}\label{eq:u^s}
\left\{
\begin{array}{ll}
\Delta u^s(\cdot,\boldsymbol{z}_\epsilon) + k^2 u^s(\cdot,\boldsymbol{z}_\epsilon) = 0 \quad \text{in $\mathbb{R}^2\setminus\overline{D}$}, \\[0.4em]
u^s(\cdot,\boldsymbol{z}_\epsilon) = -\phi(\cdot,\boldsymbol{z}_\epsilon) \quad \text{on $\partial D$}, \\[0.4em]
\text{$u^s(\cdot,\boldsymbol{z}_\epsilon)$ is radiating},
\end{array}
\right.
\end{align}
$$
$$
\begin{align}\label{eq:v^i}
\left\{
\begin{array}{ll}
\Delta v_\epsilon^i(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) + k^2 v_\epsilon^i(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) = 0 \quad \text{in $\mathbb{R}^2\setminus\overline{D_\epsilon}$}, \\[0.4em]
v_\epsilon^i(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) = -\phi(\cdot,\boldsymbol{z}_\epsilon) \quad \text{on $\partial D_\epsilon$}, \\[0.4em]
\text{$v_\epsilon^i(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon)$ is radiating},
\end{array}
\right.
\end{align}
$$
and
$$
\begin{align}\label{eq:v^s}
\left\{
\begin{array}{ll}
\Delta v_\epsilon^s(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) + k^2 v_\epsilon^s(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) = 0 \quad \text{in $\mathbb{R}^2\setminus\overline{D}$}, \\[0.4em]
v_\epsilon^s(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) = -v^i_\epsilon(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon) \quad \text{on $\partial D$}, \\[0.4em]
\text{$v_\epsilon^s(\cdot,\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon)$ is radiating}.
\end{array}
\right.
\end{align}
$$
</p>
<p>This is how the scattered fields look like.</p>
<div class="row">
<div class="column">
<img src="/blog/asymptotics-1.jpg" class="img-responsive">
</div>
<div class="column">
<img src="/blog/asymptotics-2.jpg" class="img-responsive">
</div>
</div>
<div class="row">
<div class="column">
<img src="/blog/asymptotics-3.jpg" class="img-responsive">
</div>
<div class="column">
<img src="/blog/asymptotics-4.jpg" class="img-responsive">
</div>
</div>
</p>
<p>We proved that the scattered fields verify the following <i>modified</i> Helmholtz–Kirchhoff identity,
$$
\begin{align}
u^s(\boldsymbol{x},\boldsymbol{x}') - \overline{u^s(\boldsymbol{x},\boldsymbol{x}')} \approx 2ik\sigma_\epsilon\int_{\Sigma_\epsilon}\overline{v_\epsilon(\boldsymbol{x},\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon)} v_\epsilon(\boldsymbol{x}',\boldsymbol{y}_\epsilon,\boldsymbol{z}_\epsilon)dS(\boldsymbol{y}_\epsilon) - [\phi(\boldsymbol{x},\boldsymbol{x}') - \overline{\phi(\boldsymbol{x},\boldsymbol{x}')}],
\end{align}
$$
with scaling factor \(\sigma_\epsilon = \pi^2\vert H_0^{(1)}(2\pi\epsilon)\vert^2\epsilon^{-q}\). This motivates the introduction of the following <i>modified</i> cross-correlation matrix:
$$
\begin{align}
\widetilde{C}_{jm} = \frac{2ik\vert\Sigma_\epsilon\vert\sigma_\epsilon}{L}\sum_{\ell=1}^L\overline{v_\epsilon(\boldsymbol{x}_j,\boldsymbol{y}_\epsilon^\ell,\boldsymbol{z}_\epsilon)} v_\epsilon(\boldsymbol{x}_m,\boldsymbol{y}_\epsilon^\ell,\boldsymbol{z}_\epsilon) - \left[\phi(\boldsymbol{x}_j,\boldsymbol{x}_m) - \overline{\phi(\boldsymbol{x}_j,\boldsymbol{x}_m)}\right].
\end{align}
$$
</p>
<h2>Numerical experiments</h2>
<p>We consider a kite of size \(\lambda/2\) centered at \(2\lambda + 2\lambda i\) for the wavenumber \(k=2\pi\) (wavelength \(\lambda =1\)). We take \(\epsilon=10^{-2}\), \(p=1\), \(q=2\), and \(\theta_z=\pi\) for the asymptotic model, which yields \(\boldsymbol{z}_\epsilon=-10000\). For the LSM, we take \(J=120\) equispaced sensors on the circle of radius \(5\lambda=5\),
$$
\begin{align}
\boldsymbol{x}_j = 5e^{i\theta_j}, \quad \theta_j = \frac{2\pi}{J}(j - 1), \quad 1\leq j\leq J,
\end{align}
$$
and \(L=120\) different positions of a single small scatterer on the circle of radius \(\lambda\epsilon^{-p}=100\),
\begin{align}\label{eq:scatterer-beta}
\boldsymbol{y}_\epsilon^\ell = 100e^{i\theta_y^\ell}, \quad \theta_y^\ell = 2\pi\beta_\ell, \quad 1\leq \ell\leq L,
\end{align}
where \(\beta_\ell\sim U(0,1)\) is drawn from the <a href="https://en.wikipedia.org/wiki/Continuous_uniform_distribution">uniform distribution</a> on \([0,1]\). Finally, we add some multiplicative <a href="https://en.wikipedia.org/wiki/White_noise">white noise</a> with amplitude \(\delta=5\times 10^{-3}\) to the the near-field measurements (generating a matrix \(\widetilde{C}_\delta\)), and probe the medium on a \(100\times100\) uniform grid on \([-6\lambda,6\lambda]\times[-6\lambda,6\lambda]\).</p>
<div class="row">
<div class="column">
<img src="/blog/matrix-kite.jpg" class="img-responsive">
</div>
<div class="column">
<img src="/blog/lsm-kite.jpg" class="img-responsive">
</div>
</div>
<p>The defect is well identified by our novel sampling method, based on cross-correlations and a small random scatterer. This illustrates that the LSM can be utilized in this novel passive-imaging setup.</p>
<hr>
<h4>Blog posts about inverse scattering</h4>
<p>2024 <a href="2024-02-04.html">The linear sampling method for small random scatterers</a></p>
<p>2022 <a href="2022-11-04.html">The linear sampling method for random sources</a></p>
</div>
</div>
</div>
</section>
<!-- Footer Section -->
<footer>
<div class="container">
<div class="row">
<div class="col-md-4 col-md-offset-1 col-sm-6">
<h3>Contact</h3>
<p style="color:white"><i class="fa fa-send-o"></i>[email protected]</p><br>
<a href="https://www.inria.fr/en"><img src="/images/inria.png" class="img-responsive"></a>
</div>
<div class="col-md-4 col-md-offset-1 col-sm-6">
<h3 style="opacity:0;">Contact</h3>
<p style="color:white">Copyright © 2024 Hadrien Montanelli</p><br>
<a href="https://www.ip-paris.fr/en"><img src="/images/polytechnique.png" class="img-responsive"></a>
</div>
<div class="clearfix col-md-12 col-sm-12">
<hr style="color:white;">
</div>
<div class="col-md-12 col-sm-12">
<ul class="social-icon">
<li><a href="https://github.com/Hadrien-Montanelli" class="fa-brands fa-github" style="font-size:30px;color:white"></a></li>
<li><a href="https://scholar.google.com/citations?user=Bjmkfe8AAAAJ&hl=en&oi=sra/" class="fa-brands fa-google" style="font-size:30px;color:white"></a></li>
<li><a href="https://www.linkedin.com/in/hadrien-montanelli/" class="fa-brands fa-linkedin" style="font-size:30px;color:white"></a></li>
<li><a href="https://orcid.org/0009-0005-1742-9828" class="fa-brands fa-orcid" style="font-size:30px;color:white"></a></li>
<li><a href="https://twitter.com/drmontanelli" class="fa-brands fa-x-twitter" style="font-size:30px;color:white;"></a></li>
</ul>
</div>
</div>
</div>
</footer>
<!-- Back top -->
<a href="#back-top" class="go-top"><i class="fa fa-angle-up"></i></a>
<!-- SCRIPTS -->
<script src="js/jquery.js"></script>
<script src="js/bootstrap.min.js"></script>
<script src="js/particles.min.js"></script>
<script src="js/app.js"></script>
<script src="js/jquery.parallax.js"></script>
<script src="js/smoothscroll.js"></script>
<script src="js/custom.js"></script>
<script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
<script id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
</body>
</html>