This paper explores the enhanced detection of threat materials using a combination of dark-field x-ray imaging and deep neural networks. Dark-field imaging, sensitive to material inhomogeneities, creates characteristic textures. Combining this with conventional attenuation improves threat material discrimination. The energy dependence of these signals further refines identification. Two proof-of-concept studies demonstrate the effectiveness of deep neural networks in identifying materials from dark-field textures, significantly outperforming methods without dark-field data. While further research is needed, the results highlight the potential for this combined approach in security applications.

T. Partridge, A. Astolfo, S. S. Shankar, F. A. Vittoria, M. Endrizzi, S. Arridge, T. Riley-Smith, I. G. Haig, D. Bate, A. Olivo