Scholarlink

<p>The rapid development of Machine Learning (ML) and Deep Learning (DL) methods in recent years has made them popular among the scientific community. These methods have been applied across various fields, including radiation detection and the nuclear industry. These techniques have demonstrated the potential to enhance the performance and efficiency of automated radiation detection practices and nuclear safety and security practices. This paper will discuss incorporating deep learning techniques, Fully Connected Artificial Neural Networks (FCANN), and Convolutional Neural Networks (CNN) to detect radioactive isotopes with low-resolution gamma spectral data. Further, our work aims to enhance the performance of identification while exploring the explainability of these models. Using saliency mapping methods, we visualize the networks’ regions of interest for a better understanding of the underlying decision-making processes. Both CNN and FCANN models demonstrated high accuracies exceeding 97%, while saliency mapping highlighted the relevant spectral features for underlying decision making process of each type of model.</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

AI-driven isotope detection: applying deep learning to gamma spectral analysis

  • 期刊论文

  • 发表日期：

    2026年3月5日

查看全文
  • Yasas Patikirige,
  • Lyda Pav,
  • Shakhboz Khasanov,
  • Jagath Pitawala

摘要

The rapid development of Machine Learning (ML) and Deep Learning (DL) methods in recent years has made them popular among the scientific community. These methods have been applied across various fields, including radiation detection and the nuclear industry. These techniques have demonstrated the potential to enhance the performance and efficiency of automated radiation detection practices and nuclear safety and security practices. This paper will discuss incorporating deep learning techniques, Fully Connected Artificial Neural Networks (FCANN), and Convolutional Neural Networks (CNN) to detect radioactive isotopes with low-resolution gamma spectral data. Further, our work aims to enhance the performance of identification while exploring the explainability of these models. Using saliency mapping methods, we visualize the networks’ regions of interest for a better understanding of the underlying decision-making processes. Both CNN and FCANN models demonstrated high accuracies exceeding 97%, while saliency mapping highlighted the relevant spectral features for underlying decision making process of each type of model.