A neural-network based approach to cargo inspections using photon spectroscopy

Abstract

Abstract Bremsstrahlung based dual-energy attenuation is the most widely used methodology for high-energy cargo inspection systems and Z e f f mapping. Currently, the inspection systems do not exploit the spectroscopic component of the transmitted spectrum, relying mostly on energy integrated attenuation signals. In this work, we propose a neural-network based approach to determine Z e f f from energy resolved transmitted spectra using single high-energy attenuations. The neural-network approach based on a regression algorithm was benchmarked against the dual-energy attenuation method. The evaluation is made using Geant4 simulated data covering the full range of Z values and indicates potential improvements in terms of imaging speed and Z e f f discriminating power for the neural-network based approach.