A Robust and Efficient Implicit Neural Scatterer Representation for Inverse Scattering Problems

This letter introduces Neural Scatterer Representation (NeSR), a novel implicit neural representation (INR) framework that resolves the robustness and efficiency issues in the state-of-the-art (SOTA) implicit inverse scattering method. We reveal that the current method suffers from slow convergence and optimization failures in specific scenarios. NeSR overcomes the problems through a more robust and efficient architecture with shared spatial features and total electric field representation. Comprehensive experiments demonstrate that NeSR achieves a 10${\times }$efficiency gain (0.84 min vs. 8.09 min) and a zero failure rate under fair comparison with same iteration steps. Furthermore, we validate practical applicability of NeSR on experimental data and show its enhanced robustness via a dynamic multi-frequency fusion strategy.