<p>The multiple nuclides identification algorithm with low consumption and strong robustness is crucial for rapid radioactive source searching. This study investigates the design of a low-consumption multiple nuclides identification algorithm for portable gamma spectrometers. First, the gamma spectra of 12 target nuclides (including the background case) were measured to create training datasets. The characteristic energies, obtained through energy calibration and full-energy peak addresses, are utilized as input features for a neural network. A large number of single- and multiple-nuclide training datasets are generated using random combinations and small-range drifting. Subsequently, a multi-label classification neural network based on a binary cross-entropy loss function is applied to export the existence probability of certain nuclides. The designed algorithm effectively reduces the computation time and storage space required by the neural network and has been successfully implemented in a portable gamma spectrometer with a running time of <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="41365_2025_1701_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="53" /> </InlineMediaObject> <EquationSource Format="TEX">$t_\text {r}<{2\,\textrm{s}}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msub> <mi>t</mi> <mtext>r</mtext> </msub> <mo><</mo> <mrow> <mn>2</mn> <mspace width="0.166667em" /> <mtext>s</mtext> </mrow> </mrow> </math></EquationSource> </InlineEquation>. Results show that, in both validation and actual tests, the identification accuracy of the designed algorithm reaches 94.8%, for gamma spectra with a dose rate of <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="41365_2025_1701_Article_IEq2.gif" Format="GIF" Height="19" Rendition="HTML" Resolution="72" Type="Linedraw" Width="102" /> </InlineMediaObject> <EquationSource Format="TEX">$d\approx {0.5\,\mathrm{\upmu Sv/h}}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>d</mi> <mo>≈</mo> <mrow> <mn>0.5</mn> <mspace width="0.166667em" /> <mrow> <mi mathvariant="normal">μ</mi> <mi mathvariant="normal">Sv</mi> <mo stretchy="false">/</mo> <mi mathvariant="normal">h</mi> </mrow> </mrow> </mrow> </math></EquationSource> </InlineEquation> and a measurement time <InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="41365_2025_1701_Article_IEq3.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="65" /> </InlineMediaObject> <EquationSource Format="TEX">$t_\text {m}={60\,\textrm{s}}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msub> <mi>t</mi> <mtext>m</mtext> </msub> <mo>=</mo> <mrow> <mn>60</mn> <mspace width="0.166667em" /> <mtext>s</mtext> </mrow> </mrow> </math></EquationSource> </InlineEquation>. This improves the ability to perform rapid on-site nuclide identification at important sites.</p>

A low-consumption multiple nuclides identification algorithm for portable gamma spectrometer

期刊论文
发表日期：

2025年5月9日

查看全文

Xi Huang,
Yong-Gang Yuan,
Yu-Xuan Zhu,
Hua Chen,
Jin-Hui Qu,
Zhao-Yi Tan

摘要

The multiple nuclides identification algorithm with low consumption and strong robustness is crucial for rapid radioactive source searching. This study investigates the design of a low-consumption multiple nuclides identification algorithm for portable gamma spectrometers. First, the gamma spectra of 12 target nuclides (including the background case) were measured to create training datasets. The characteristic energies, obtained through energy calibration and full-energy peak addresses, are utilized as input features for a neural network. A large number of single- and multiple-nuclide training datasets are generated using random combinations and small-range drifting. Subsequently, a multi-label classification neural network based on a binary cross-entropy loss function is applied to export the existence probability of certain nuclides. The designed algorithm effectively reduces the computation time and storage space required by the neural network and has been successfully implemented in a portable gamma spectrometer with a running time of $t_\text {r}<{2\,\textrm{s}}$ $t_{r} < 2 s$ . Results show that, in both validation and actual tests, the identification accuracy of the designed algorithm reaches 94.8%, for gamma spectra with a dose rate of $d\approx {0.5\,\mathrm{\upmu Sv/h}}$ $d \approx 0.5 μ Sv / h$ and a measurement time $t_\text {m}={60\,\textrm{s}}$ $t_{m} = 60 s$ . This improves the ability to perform rapid on-site nuclide identification at important sites.