DiffusionEMIS: Diffusion Model for 3D Electromagnetic Inverse Scattering

We propose DiffusionEMIS, a new paradigm that formulates the 3D electromagnetic (EM) inverse scattering (EMIS) as a denoising diffusion process from a 4D noise distribution to the specific 3D scatterer point cloud with reasonable electrical parameters at each point under the scattering field as condition to enrich the research of 3D EMIS community, while avoiding the computational burden of 3D meshes. During training process, scatterer point clouds diffuse from ground truth shapes to random distributions, and the model learns to reverse this noising process. During inference process, the model progressively refines the randomly generated shape to the output result. We design a certain 3D scattering system to construct two datasets: 3DEMIS-MNIST, where the samples are non-uniform 3D handwritten digits, and 3DEMIS-SHAPENET, where the samples are uniform 3D objects. The extensive evaluations on 3DEMIS-MNIST and 3DEMIS-SHAPENET, show that our DiffusionEMIS achieves favorable performance and great stability which means 3D EMIS can be solved by a generative way. Our model outperforms previous common EMIS methods such as the Born Iterative Method (BIM) by a large margin and has extremely robust noise resistance. To evaluate the generalization performance of diffusionEMIS, we construct the dataset 3DEMIS-EMNIST consisting of non-uniform 3D handwritten letters for testing the model trained on 3DEMIS-MNIST, the test results indicate a great consistency with the groundtruths. We also conducted tests on real data to further demonstrate the effectiveness of our method.