A Pruned Neural Network in Multi-Monostatic Inversion of Two-Dimensional Metallic Scatterers

We consider the application of Neural Networks (NNs) to the case of multi-monostatic inversion of two-dimensional metallic scatterers. The aim is twofold. First, we want to reconstruct the support of the involved objects from the scattering data. Second, we want to demonstrate a pruning process to reduce the NN complexity without a worsening of its performance. In particular, the NN result is compared to that achieved with the standard approach of reconstructing the re-flection coefficient by a Truncated Singular Value Decomposition (TSVD) scheme. The approach is validated experimentally.