The content presents a novel approach to solving the three-dimensional inverse obstacle scattering problem, which aims to recover the shape of an obstacle from sparse and noisy far-field measurements.
The key aspects are:
The method uses a trained latent representation of surfaces, called DeepSDF, as the generative prior. This latent representation enjoys excellent expressivity within the given class of shapes, while the latent dimensionality is low, which greatly facilitates the computation.
The admissible manifold of surfaces is realistic, and the resulting optimization problem is less ill-posed compared to traditional approaches.
The shape derivative is employed to evolve the latent surface representation by minimizing the loss function. A local convergence analysis of a gradient descent type algorithm to a stationary point of the loss is provided.
Numerical examples, including backscattered and phaseless data, are presented to showcase the effectiveness of the proposed algorithm. The method is observed to be highly efficient and robust to data noise, converging within tens of iterations and yielding reasonable approximations even with up to 40% noise in the data.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Junqing Chen... at arxiv.org 04-18-2024
https://arxiv.org/pdf/2311.07187.pdfDeeper Inquiries