Research on an Improved Reconstruction Approach for Local Transmission Imaging in TGS

Tomographic gamma scanning is a nondestructive γ-ray measurement technique that utilizes computed tomography–based reconstruction methods to analyze the internal distribution of radioactive waste materials. To address the insufficient reconstruction accuracy of local regions under sparse-angle transmission measurement conditions, a geometry-guided global-to-local adaptive reconstruction method is proposed in this study.A surface layer voxel model of a radioactive waste container was established according to the practical configuration of low- and intermediate-level radioactive solid waste steel containers. The surface region was divided into 25 square voxel grids, with each a voxel size of 12 × 12 cm. Concrete filling material was placed at the 16th voxel position, and two models were constructed, in which the concrete occupied one-ninth and one-half of the initial voxel volume, respectively.Based on the source-detector scanning geometry, a geometric relationship X = H-2W was introduced to determine the minimum voxel partition length and the required number of subdivisions in the local region of interest. The Alternating Direction Method of Multipliers (ADMM)–Orthogonal Matching Pursuit (OMP) reconstruction framework was employed to reconstruct sparse-angle transmission projection data and improve the transmission image quality. Monte Carlo simulations were performed to validate the proposed reconstruction framework.The results demonstrated that under sparse-angle sampling conditions, the OMP method produces relatively poor reconstruction performance, with peak signal-to-noise ratio (PSNR) values of approximately 30 and structural similarity index measure (SSIM) values of around 0.6. In contrast, the ADMM-OMP framework effectively improved the reconstruction results, achieving PSNR values close to 40 and SSIM values close to 1, thereby improving the quality of the reconstructed transmission images.Finally, the global reconstructed image and the locally refined reconstruction results were fused, effectively alleviating the degradation of local reconstruction quality caused by the filling material occupying only a small fraction of the initial voxel in the region of interest.