Physically Unrolling Network Under Contraction Integral Equation for Limited-Aperture Inverse Scattering Problem

Xu, Kuiwen; Qian, Zemin; Song, Rencheng; Ye, Xiuzhu; Xu, Ning; Pan, Xiao-Min; Zhao, Peng; Chen, Shichang; Wang, Gaofeng; Li, Wenjun

doi:10.1109/tap.2023.3305834

Cited by 2 publications

(5 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As such, it is particularly suitable for the reconstruction of complex scenarios. Moreover, it is different from [36] and [38], as it only focuses on the contrast function and considers at each block only a neural network without involving traditional optimization methods.…”

Section: Introductionmentioning

confidence: 99%

“…In the last decade, by virtue of the rapid progress in the field of artificial intelligence, learning-based approaches have also been recently proposed to solve ISPs, (see for example [23][24][25][26][27][28][29][30][31][32][33][34][35][36][37][38]). Learning strategies could provide a good way to overcome the drawbacks related to the aforementioned optimization approaches, making the reconstruction process as reliable as possible and speeding up the convergence time.…”

Section: Introductionmentioning

confidence: 99%

“…Another possibility of improving generalizability and reliability is represented by the class of 'unrolled' or 'unrolling' methods, which combine traditional iterative methods and DL [35][36][37][38][39]. Paper [29] was probably the first to discuss the connection between conventional iterative inverse-scattering algorithms and deep neural networks.…”

Section: Introductionmentioning

confidence: 99%

“…Then, the 'unrolled' idea was developed and proposed in [35], which is a general and theoretical work about the advantages and disadvantages of the unrolling algorithms [37] proposes an unrolled DL method for magnetic resonance imaging, while [39] considers 2D photoacoustic tomography. As far as [36] and [38] are concerned, they are both applied to the ISP and consider physics within the neural network. In [36] an iterative neural network structure is proposed that combines the contrast source inversion (CSI) method and residual network.…”

Section: Introductionmentioning

confidence: 99%

“…It well incorporates domain knowledge into the learning process by embedding the CSI iterations into the deep learning model. On the other hand, in [38] several sub-modules were proposed, constructed with a small CNN and an operator of the least square method, which are used to update the induced current and the modified contrast function, respectively.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

MiPhDUO: microwave imaging via physics-informed deep unrolled optimization

Zumbo,

Mandija,

Isernia

et al. 2024

Inverse Problems

View full text Add to dashboard Cite

Microwave imaging (MWI) is a non-invasive technique that can identify unknown scattererobjects’ features while offering advantages such as low cost and portable devices with respectto other imaging methods. However, MWI faces challenges in solving the underlying inversescattering problem (ISP), which involves recovering target properties from its scattered fields.Existing methods include linearized and non-linear optimization approaches, but they havelimitations respectively in terms of range of validity and computational complexity (in view ofthe possible occurrence of ‘false solutions’). In recent years, learning-based approaches haveemerged as they can allow real-time imaging but usually lack generalizability and a directconnection to the underlying physics. This paper proposes a physics-informed approach thatcombines convolutional neural networks (CNNs) with physics-based calculations. It is basedon a few cascaded operations, making use of the gradient of the relevant cost function, andsuccessively improving the estimation of the unknown target. The proposed approach isassessed using simulated as well as experimental Fresnel data. The results show that theintegration of physics with deep learning can contribute to improve reconstruction accuracy,generalizability, and computational efficiency in MWI.

show abstract

Section: Introductionmentioning

confidence: 99%