Scholarlink

<p>Reconstructing neutron spectra from limited measured counts is a challenging inverse problem. This paper presents a dual-domain feature-enhanced deep unfolding network (DDFEDUN), integrating the advantages of compressed sensing and deep learning. By constructing a collaborative architecture consisting of a wavelet-domain deep unfolding module (WDDUM), a convolution-domain deep unfolding module (CDDUM), and a dual-scale denoising module (DSDM), the proposed method fully leverages the structural prior information of neutron spectra across multiple transformation domains. This approach effectively enhances the model’s feature representation capability and noise robustness. Experimental results have demonstrated that DDFEDUN achieves high-precision and stable reconstruction across various typical neutron spectra.</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

A neutron spectrum unfolding method based on the dual-domain feature-enhanced deep unfolding network

  • 期刊论文

  • 发表日期：

    2025年10月24日

查看全文
  • Yanfei Liu,
  • Chenyang Wang,
  • Hao Zheng,
  • Qingshan Yin,
  • Dongdong Yang,
  • Qi Li,
  • Ning Lv

摘要

Reconstructing neutron spectra from limited measured counts is a challenging inverse problem. This paper presents a dual-domain feature-enhanced deep unfolding network (DDFEDUN), integrating the advantages of compressed sensing and deep learning. By constructing a collaborative architecture consisting of a wavelet-domain deep unfolding module (WDDUM), a convolution-domain deep unfolding module (CDDUM), and a dual-scale denoising module (DSDM), the proposed method fully leverages the structural prior information of neutron spectra across multiple transformation domains. This approach effectively enhances the model’s feature representation capability and noise robustness. Experimental results have demonstrated that DDFEDUN achieves high-precision and stable reconstruction across various typical neutron spectra.