Search papers, labs, and topics across Lattice.
This paper introduces the Coordinate-Residual Physics-Driven Neural Network (CRPDNN) for tackling the challenges of 3-D electromagnetic inverse scattering, a problem characterized by nonlinearity and ill-posedness. By eliminating the need for preliminary reconstruction-based region selection, CRPDNN directly reconstructs the unknown contrast distribution using normalized spatial coordinates and a residual convolutional network, achieving an average relative error of 2.10% in noise-free synthetic cases. Additionally, it demonstrates significant computational efficiency, providing speedups of approximately 5.5 and 12.1 times over existing methods while maintaining stability under noisy conditions.
Achieving a 2.10% reconstruction error with a method that eliminates preliminary region selection could revolutionize 3-D electromagnetic imaging.
Electromagnetic inverse scattering is a nonlinear and ill-posed 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. The proposed solver directly reconstructs the unknown contrast distribution using normalized spatial coordinates and a residual convolutional network, without requiring a preliminary reconstruction. 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. Supplementary 2-D comparisons further confirm 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. The related code is available at https://github.com/Physics-driven-methods.
