Scholarlink

<p>Muon scattering tomography (MST) harnesses naturally occurring cosmic-ray muons to non-invasively probe dense or thick materials, making it particularly effective for identifying high-Z objects in applications such as cargo inspection. In this study, we propose a novel pipeline that integrates points of closest approach (PoCA)-based 3D voxel reconstruction with a 3D convolutional neural network (3D-CNN) denoiser. Our framework is tested on a dataset comprising six objects-two dense materials (iron and flour), each in three size variants-scanned using muon tomography. The PoCA-based method first transforms raw muon scattering measurements into high-resolution 3D voxel grids, capturing the spatial distribution of the objects within the scanned volume. Subsequently, our 3D-CNN-based VoxelDenoiser refines these voxel grids by suppressing noise while preserving essential features. Together, the PoCA reconstruction and deep-learning denoising significantly enhance visualization and facilitate accurate identification of targets within cluttered or complex volumes. This synergistic approach extends the capabilities of traditional MST, offering a robust solution for security screening, nuclear safeguards, and broader material characterization tasks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

3D reconstruction and denoising of high-Z materials from muon tomography using 3D CNN

  • 期刊论文

  • 发表日期：

    2025年3月8日

查看全文
  • Prasoon Kumar Vinodkumar,
  • Egils Avots,
  • Cagri Ozcinar,
  • Gholamreza Anbarjafari

摘要

Muon scattering tomography (MST) harnesses naturally occurring cosmic-ray muons to non-invasively probe dense or thick materials, making it particularly effective for identifying high-Z objects in applications such as cargo inspection. In this study, we propose a novel pipeline that integrates points of closest approach (PoCA)-based 3D voxel reconstruction with a 3D convolutional neural network (3D-CNN) denoiser. Our framework is tested on a dataset comprising six objects-two dense materials (iron and flour), each in three size variants-scanned using muon tomography. The PoCA-based method first transforms raw muon scattering measurements into high-resolution 3D voxel grids, capturing the spatial distribution of the objects within the scanned volume. Subsequently, our 3D-CNN-based VoxelDenoiser refines these voxel grids by suppressing noise while preserving essential features. Together, the PoCA reconstruction and deep-learning denoising significantly enhance visualization and facilitate accurate identification of targets within cluttered or complex volumes. This synergistic approach extends the capabilities of traditional MST, offering a robust solution for security screening, nuclear safeguards, and broader material characterization tasks.