Scholarlink

Purpose <p>Accurately identifying radioactive nuclides is essential for environmental monitoring and nuclear security.This study aims to develop a deep learning-based methodology within tomographic gamma scanning (TGS) to distinguish <sup>137</sup>Cs <sup>110m</sup>andAg, as their misidentifi cation poses signifi cant safety risks in nuclear waste management.</p> Methods <p>A bidirectional long short-term memory (Bi-LSTM) network was employed as the baseline classifi er, withhyperparameter optimization performed to enhance discrimination capability. A Monte Carlo model was constructedaccording to actual nuclear waste drum dimensions, incorporating detector response and environmental radioactivebackground contributions. The Bi-LSTM model was trained and evaluated using simulated spectroscopic data undervarying <sup>137</sup>Cs/ <sup>110m</sup> ⁰ᵐAg activity ratios.</p> Results <p>The optimized Bi-LSTM model achieved an average identifi cation accuracy exceeding 97% across all testedactivity ratio scenarios, demonstrating robust and reliable separation between the two nuclides.</p> Conclusion <p>The proposed methodology signifi cantly improves the accuracy and effi ciency of radioactive nuclideidentifi cation in low- and intermediate-level radioactive waste materials, providing a reliable foundation for activityestimation and safer nuclear waste management practices.</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

The identification of radioactive nuclides 137Cs/110mAg with deep learning in TGS

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Yun Wang,
  • Xiaoshen Kang,
  • Shaobo Wang,
  • Xuan Wang,
  • Xiaosheng Qiu,
  • Lei Wang,
  • Jianxiong Shao,
  • Lina Li,
  • Panfeng Xu,
  • Junyi Pei,
  • Dingyan Jiang,
  • Huiru Yu,
  • Jiashuo Zhang,
  • Wei Guan,
  • Jie Qiu

摘要

Purpose

Accurately identifying radioactive nuclides is essential for environmental monitoring and nuclear security.This study aims to develop a deep learning-based methodology within tomographic gamma scanning (TGS) to distinguish 137Cs 110mandAg, as their misidentifi cation poses signifi cant safety risks in nuclear waste management.

Methods

A bidirectional long short-term memory (Bi-LSTM) network was employed as the baseline classifi er, withhyperparameter optimization performed to enhance discrimination capability. A Monte Carlo model was constructedaccording to actual nuclear waste drum dimensions, incorporating detector response and environmental radioactivebackground contributions. The Bi-LSTM model was trained and evaluated using simulated spectroscopic data undervarying 137Cs/ 110m ⁰ᵐAg activity ratios.

Results

The optimized Bi-LSTM model achieved an average identifi cation accuracy exceeding 97% across all testedactivity ratio scenarios, demonstrating robust and reliable separation between the two nuclides.

Conclusion

The proposed methodology signifi cantly improves the accuracy and effi ciency of radioactive nuclideidentifi cation in low- and intermediate-level radioactive waste materials, providing a reliable foundation for activityestimation and safer nuclear waste management practices.