摘要
核素识别作为当前核安全研究方向的焦点,其检测的准确性对于安全防护具有重要意义.为了解决放射性核素识别速度慢,准确率不高的问题,提出一种基于特征融合Swin-Tiny Transformer轻量级模型的核素识别方法.通过NaI探测器测量^(133)Ba、^(60)Co、^(152)Eu、^(137)Cs等4种单个放射性核素的能谱数据,制作训练数据集;预处理阶段,采用格拉姆角场法、希尔伯特曲线法与Chirplet变换法将γ能谱信息转化二维图像,并运用特征融合方法更加凸显γ能谱信息特征;设计残差分组卷积模块,通过两个分支分别提取图像的局部和全局特征,并使用残差连接将分支信息进行有效聚合;使用利用NaI探测器采集以上4种放射性核素的混合γ能谱作为测试数据集进行识别验证.实验结果表明,模型的平均识别准确率达到了99.87%,F_(1)分数为99.88%,与其他算法相比,该算法不仅有效提高了放射源的安全防护,避免了辐射威胁,同时在保证识别速度的前提下,进一步提升了识别的准确性.
The nuclide identification,as the focus of the current nuclear security research direction,was considered of great significance for security protection.In order to solve the problem of slow speed and low accuracy of radionuclide identification,a nuclide identification method based on feature fusion Swin-Tiny Transformer lightweight model was proposed.The energy spectrum data of four individual radionuclides,such as^(133)Ba,^(60)Co,^(152)Eu,and 137 Cs,were measured by the NaI detector to produce the training dataset;in the preprocessing stage,Gram s angle field method,Hilbert s curve method,and Chirplet transform method were used to transform theγspectrum information into a two-dimensional image,and the feature fusion method was applied to highlight the characteristics of theγspectrum information more.The design of the residual grouping convolution module was designed to extract the local and global features of the image through two branches,and the branch information was effectively aggregated using the residual connection.The mixedγenergy spectra of the above four radionuclides were collected using the NaI detector as a test dataset for the identification and verification.The experimental results showed that the average recognition accuracy of the model reached 99.87%,and the F_(1)score was 99.88%.Meanwhile,compared with other algorithms,the comparison results showed that the algorithm not only effectively improved the security of the radioactive sources and avoided the radiation threat,but also further improved the recognition accuracy while ensuring the recognition speed.
作者
顾威
孟献才
洪兵
GU Wei;MENG Xiancai;HONG Bing(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan 232000,China;Institute of Energy,Hefei Comprehensive National Science Center(Anhui Energy Laboratory),Hefei 230000,China)
出处
《哈尔滨商业大学学报(自然科学版)》
2025年第2期161-168,共8页
Journal of Harbin University of Commerce:Natural Sciences Edition
基金
国家自然科学基金(12105135)
国家自然科学基金青年项目(12305200)
合肥综合性国家科学中心能源研究院(安徽省能源实验室)项目(21KZS202、21KZS208)
安徽省住房城乡建设科学技术计划项目(2023-RK043)
高校协同创新项目(GXXT-2022-003)。
