Speaker
Description
Neutron radiography plays an important role in non-destructive evaluation, including aerospace inspection and battery analysis. Recent advances in compact accelerator-driven neutron sources and neutron flat-panel detectors (nFPDs) have improved the accessibility and performance of neutron imaging, enabling practical three-dimensional neutron computed tomography (CT). However, reducing acquisition time by decreasing projection numbers or exposure time, inevitably degrades reconstruction quality due to limited neutron statistics. This leads to a trade-off between measurement efficiency and image accuracy.
To address this issue, we proposed the Dykstra-like splitting with proximal minimization trick (DLPM) algorithm, which provides high reconstruction accuracy, fast convergence, and does not require empirically tuned parameters. Simulation results under low-dose CT conditions demonstrate that DLPM effectively suppresses noise compared to the conventional filtered back projection (FBP) method.
Furthermore, the proposed method is applied to industrial neutron CT using experimental data acquired at the compact neutron source AISTANS (AIST, Japan). The results demonstrate that DLPM is a practical reconstruction approach with potential for improving throughput in industrial neutron CT applications, particularly under time-constrained conditions.