Retention-Based Hyperspectral X-ray CT for Material Recognition and Classification

Abstract Hyperspectral X-ray computed tomography (HXCT) provides energy-resolved attenuation profiles for nondestructive material discrimination, extending conventional CT beyond a single effective attenuation value. However, the combined effects of the photon-counting detector response and tomographic reconstruction can distort reconstructed profiles and reduce the reliability of spectral features used for material identification. Here, we present an HXCT analysis framework that employs a retention-based, one-pass mapping to rectify reconstructed CT profiles toward reference profiles acquired by spectral radiography using the same photon-counting system. Experiments on ten representative materials demonstrate improved agreement between rectified CT and reference profiles while preserving profile characteristics relevant to discrimination. Across the ten materials, the mean normalized-profile RMSE decreased from 0.106 before rectification to 0.061 after retention-based rectification. In the material classification task, accuracy increased from 82.4% for uncorrected CT profiles to 91.8% after rectification, approaching the 92.3% obtained using the reference profiles. These results demonstrate the potential of retention-based spectral rectification for improving HXCT-based nondestructive material discrimination under the investigated conditions.