Scholarlink

<p>Qualitative identification and analysis of radioactive nuclides in unknown environments are essential for the remote monitoring and prompt early warning of radioactive contamination. In recent years, deep learning techniques have made significant strides in automated qualitative identification. However, the quantitative analysis of radioactive nuclides still depends on traditional methods to determine peak positions and boundaries. These methods often require extensive manual expertise and parameter tuning and thus fail to meet the demands of unmanned remote monitoring. This paper presents a novel framework for automatic full-energy peak segmentation, called YOLOSpecNN. We introduce a multi-root mean square error joint optimization function and a unified regression model capable of simultaneously predicting the central position, boundaries, and confidence of full-energy peaks. To address the challenge of low recall rates due to narrow, low-intensity, and overlapping peaks, we propose a new multiscale context feature extraction module (MSNN module). This module effectively enhances the local detailed features and significantly improves the recall rates. The effectiveness of the proposed method is validated using six artificial radioactive nuclides (<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(^{241}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow /> <mn>241</mn> </mmultiscripts> </math></EquationSource> </InlineEquation>Am, <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(^{57}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow /> <mn>57</mn> </mmultiscripts> </math></EquationSource> </InlineEquation>Co, <InlineEquation ID="IEq6"> <EquationSource Format="TEX">\(^{131}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow /> <mn>131</mn> </mmultiscripts> </math></EquationSource> </InlineEquation>I, <InlineEquation ID="IEq7"> <EquationSource Format="TEX">\(^{134}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow /> <mn>134</mn> </mmultiscripts> </math></EquationSource> </InlineEquation>Cs, <InlineEquation ID="IEq8"> <EquationSource Format="TEX">\(^{137}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow /> <mn>137</mn> </mmultiscripts> </math></EquationSource> </InlineEquation>Cs, and <InlineEquation ID="IEq9"> <EquationSource Format="TEX">\(^{60}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow /> <mn>60</mn> </mmultiscripts> </math></EquationSource> </InlineEquation>Co) along with <InlineEquation ID="IEq10"> <EquationSource Format="TEX">\(^{40}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow /> <mn>40</mn> </mmultiscripts> </math></EquationSource> </InlineEquation>K, and a mixed-energy spectrum dataset is constructed for quantitative evaluation. The experimental results show that the proposed method significantly outperforms traditional approaches, achieving a precision of 0.998, a recall of 0.95, the best F1 score of 0.974@0.427, and an average precision of 0.946. Compared with traditional morphological methods, the proposed method improves the precision, recall, and best F1 score by 0.512, 0.199, and 0.391, respectively. Ablation experiments further reveal that the MSNN module notably enhances the recall by 0.067. Moreover, the proposed method performs excellently even in challenging environments with low gross counts and a low signal-to-noise ratio (SNR), achieving state-of-the-art results. Additionally, the model achieves an average real-time inference performance of 16.1941 ms on a device with a 15- W low-power budget. Overall, the proposed method demonstrates exceptional performance in the automatic search and segmentation of full-energy peaks, offering robust support for the implementation of unmanned remote radiation monitoring systems.</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

YOLOSpecNN: a novel \(\gamma \)-ray spectra full-energy peak automatic search and segmentation model inspired by YOLO

  • 期刊论文

  • 发表日期：

    2026年5月2日

查看全文
  • Cao-Lin Zhang,
  • Jiang-Mei Zhang,
  • Hao-Lin Liu,
  • Shu-Ya Qin,
  • Jia-Qi Wang

摘要

Qualitative identification and analysis of radioactive nuclides in unknown environments are essential for the remote monitoring and prompt early warning of radioactive contamination. In recent years, deep learning techniques have made significant strides in automated qualitative identification. However, the quantitative analysis of radioactive nuclides still depends on traditional methods to determine peak positions and boundaries. These methods often require extensive manual expertise and parameter tuning and thus fail to meet the demands of unmanned remote monitoring. This paper presents a novel framework for automatic full-energy peak segmentation, called YOLOSpecNN. We introduce a multi-root mean square error joint optimization function and a unified regression model capable of simultaneously predicting the central position, boundaries, and confidence of full-energy peaks. To address the challenge of low recall rates due to narrow, low-intensity, and overlapping peaks, we propose a new multiscale context feature extraction module (MSNN module). This module effectively enhances the local detailed features and significantly improves the recall rates. The effectiveness of the proposed method is validated using six artificial radioactive nuclides ( \(^{241}\) 241 Am, \(^{57}\) 57 Co, \(^{131}\) 131 I, \(^{134}\) 134 Cs, \(^{137}\) 137 Cs, and \(^{60}\) 60 Co) along with \(^{40}\) 40 K, and a mixed-energy spectrum dataset is constructed for quantitative evaluation. The experimental results show that the proposed method significantly outperforms traditional approaches, achieving a precision of 0.998, a recall of 0.95, the best F1 score of 0.974@0.427, and an average precision of 0.946. Compared with traditional morphological methods, the proposed method improves the precision, recall, and best F1 score by 0.512, 0.199, and 0.391, respectively. Ablation experiments further reveal that the MSNN module notably enhances the recall by 0.067. Moreover, the proposed method performs excellently even in challenging environments with low gross counts and a low signal-to-noise ratio (SNR), achieving state-of-the-art results. Additionally, the model achieves an average real-time inference performance of 16.1941 ms on a device with a 15- W low-power budget. Overall, the proposed method demonstrates exceptional performance in the automatic search and segmentation of full-energy peaks, offering robust support for the implementation of unmanned remote radiation monitoring systems.