Coordinate-Residual Physics-Driven Neural Network for Inverse Scattering Imaging

Abstract

Electromagnetic inverse scattering is a nonlinear and ill-posed computational imaging problem, where accurate reconstruction is challenging due to measurement limitations, noise, and high computational costs, especially for 3-D imaging. Although physics-driven neural networks (PDNNs) reduce the dependence on labeled training data, existing accelerated PDNN frameworks often rely on preliminary reconstruction-based region selection, which may introduce instability when the selected region is inaccurate. In this paper, a coordinate-residual physics-driven neural network (CRPDNN) is proposed for 3-D electromagnetic inverse scattering. CRPDNN represents the unknown complex contrast distribution using normalized spatial coordinates and a residual convolutional network, whose parameters are optimized by enforcing consistency between the measured and model-predicted scattered fields. Unlike existing subregion-accelerated PDNN approaches, CRPDNN does not require a preliminary reconstruction, thereby avoiding dependence on its accuracy. For the reported noise-free 3-D synthetic cases, CRPDNN achieves an average relative error of 2.10\%, compared with 7.97\% for CSI and 3.99\% for L_2/3-FBE-WCIE, while providing approximately 5.5- and 12.1-fold speedups over the two baselines, respectively. Additional 2-D comparisons further demonstrate its stability and computational efficiency relative to existing PDNN frameworks. CRPDNN also maintains reliable reconstruction performance under noisy measurements, and the 3-D Fresnel experiments further indicate its potential for practical imaging applications.

Hard-Constrained Physics--Informed Neural Network with Adaptive Regional Residual Balancing for the Generalized Falkner--Skan Problem
Mehari Fentahun Endalew, Xiaoming John Zhang
arXiv20 Sep 2026PreprintCC BY 4.0Full textFluid dynamicsarXiv
Sparse defect localization - A Periodic Empirical Operator Framework for Inverse Scattering and Structural Health Monitoring
Richard Wüest, Johannes Aichele, Jonas Mueller +1
HAL9 Oct 2026PreprintLicence not statedGeneral physicsHAL
Inverse scattering method for nonlinear negative first order coupled Klein-Gordon equation
Cihan Sabaz, Ilmar Gahramanov, Mansur I. Ismailov
arXiv2 Sep 2026PreprintCC BY 4.0Full textAnalysis & PDEs Mathematical physicsarXiv
Microfacet-based physics-driven framework for single-photon LiDAR rough-surface scattering and detection
Hao Liao, Yanhui Li, Pengfei Yang +3
Optics Express2 Oct 2026Peer-reviewedPublisher termsApplied physics & instrumentationCrossref
Thermodynamic Bethe Ansatz for the Toda chain from inverse scattering transform
Alberto Brollo
arXiv2 Nov 2026PreprintarXiv licenceFull textMathematical physics Statistical & nonlinear physicsarXiv

Cite

BibTeX RIS

BibTeX

@misc{du2026coordinateresidual,
  title = {{Coordinate-Residual Physics-Driven Neural Network for Inverse Scattering Imaging}},
  author = {Yutong Du and Zicheng Liu and Bo Qi and Yali Zong and Peixian Han},
  howpublished = {arXiv},
  year = {2026},
  month = sep,
  doi = {10.48550/arxiv.2608.09382},
  eprint = {2608.09382},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2608.09382},
}

RIS

TY  - UNPB
TI  - Coordinate-Residual Physics-Driven Neural Network for Inverse Scattering Imaging
AU  - Yutong Du
AU  - Zicheng Liu
AU  - Bo Qi
AU  - Yali Zong
AU  - Peixian Han
PY  - 2026
DA  - 2026/09/01
PB  - arXiv
DO  - 10.48550/arxiv.2608.09382
UR  - https://arxiv.org/abs/2608.09382
AB  - Electromagnetic inverse scattering is a nonlinear and ill-posed computational imaging problem, where accurate reconstruction is challenging due to measurement limitations, noise, and high computational costs, especially for 3-D imaging. Although physics-driven neural networks (PDNNs) reduce the dependence on labeled training data, existing accelerated PDNN frameworks often rely on preliminary reconstruction-based region selection, which may introduce instability when the selected region is inaccurate. In this paper, a coordinate-residual physics-driven neural network (CRPDNN) is proposed for 3-D electromagnetic inverse scattering. CRPDNN represents the unknown complex contrast distribution using normalized spatial coordinates and a residual convolutional network, whose parameters are optimized by enforcing consistency between the measured and model-predicted scattered fields. Unlike existing subregion-accelerated PDNN approaches, CRPDNN does not require a preliminary reconstruction, thereby avoiding dependence on its accuracy. For the reported noise-free 3-D synthetic cases, CRPDNN achieves an average relative error of 2.10\%, compared with 7.97\% for CSI and 3.99\% for L_2/3-FBE-WCIE, while providing approximately 5.5- and 12.1-fold speedups over the two baselines, respectively. Additional 2-D comparisons further demonstrate its stability and computational efficiency relative to existing PDNN frameworks. CRPDNN also maintains reliable reconstruction performance under noisy measurements, and the 3-D Fresnel experiments further indicate its potential for practical imaging applications.
ER  -

Abstract

Related

Cite