Learning a Nonlinear Solver Optimized to Solve the Problem of Electron Density and Effective Atomic Number Reconstruction

Executive Summary

We will attempt to demonstrate the feasibility of using neural networks to solve a known complex nonlinear, non-convex inverse problem related to the three-dimensional reconstruction of electron density. If successful, this approach will enable the development of dual-energy x-ray imaging to detect the presence of explosives, contraband, and nuclear materials in luggage and cargo containers, thereby reducing the threat of weapons of mass destruction by improving detection.