- By:
- Yang, Diyu; Chowdhury, Mohammad Samin Nur; Tang, Shimin ; Singanallur Vaidyanathan, Venkatakrishnan ; Bilheux, Hassina ; Buzzard, Gregery; Bouman, Charles
- Page Number:
- 664-668
- Book Title:
- 2024 58th Asilomar Conference on Signals, Systems, and Computers
- Publication Date:
- September 21, 2026
- Publisher Location:
- IEEE, New Jersey, United States of America
- Conference Name:
- Asilomar Conference on Signals, Systems, and Computers (ACSSC)
- Conference Location:
- Pacific Grove, California, United States of America
- Conference Sponsor:
- IEEE
- View DOI Listing:
- https://doi.org/10.1109/IEEECONF60004.2024.10942621
Abstract
Hyperspectral neutron computed tomography (HSnCT) is an effective technique for characterizing poly crystalline material samples. A typical scan involves making multiple HS projection measurements by rotating the sample about a single axis and using a standard algorithm for tomographic reconstruction. Recently, an autonomous polycrystalline material decomposition (APMD) algorithm was proposed to obtain accurate 3D reconstruction of the different materials or crystallographic phases in the object. However, for objects with complex compositions and shapes, using data from a single rotation axis may result in reconstructions with significant noise and inaccuracies in the material decomposition. In this paper, we present a multi-pose reconstruction algorithm to produce a single reconstruction from HSnCT data corresponding to multiple poses of the object. Our algorithm extends previous APMD work to incorporate hyperspectral neutron measurements from multiple poses, utilizing the Multi-Agent Consensus Equilibrium (MACE) framework to integrate projections from different rotation axes into a single reconstruction. We apply our method to simulated data and demonstrate that multi-pose APMD achieves significantly improved material decomposition accuracy compared to single-pose APMD.
