为了提高货物或人体放射性筛查的可靠性，提出了一种基于主成分分析和Mahalanobis距离的异常γ能谱识别方法。该方法首先对大量不含异常放射性的测量对象产生的正常γ能谱进行主成分分析，提取出其所有主成分，并按从大到小的顺序，选取前若干主成分构成子空间；将正常及待识别γ能谱在此子空间上投影，得到它们的Mahalanobis距离，通过比较这些距离的相对大小实现对异常γ能谱的识别。Monte Carlo模拟实验和实际测试实验表明，在子空间信息量占原始信息比例大于99%时该方法可靠有效。
A principal component analysis (PCA) and Mahalanobis distance (MD) based anomaly gamma spectra detection method is developed to improve the reliability of radiation scanning of goods and human bodies. This method first extracts all the principal components (PCs) of large numbers of benign gamma spectra by PCA and selects several largest PCs to form a subspace. The algorithm then projects the benign, unknown gamma spectra to this subspace, calculates their MDs, and completes the anomaly detection by comparing these MDs. Monte Carlo simulations and actual tests show that the method is reliable and effective when the subspace has more than 99% of the original information.
[1] Canberra Industries. Genie 2000 Customization Tools Manual[R]. Meriden, CT:Canberra Industries, 2006:300-310. [2] Advanced Measurement Technology. GammaVision V6 Users Manual[R]. Oak Ridge, TN:Advanced Measurement Technology, 2006:180-185. [3] Boardman D, Reinhard M, Flynn A. Principal component analysis of gamma-ray spectra for radiation portal monitors[J]. IEEE Transactions on Nuclear Science, 2012, 59(1):154-161. [4] Runkle R, Tardiff F, Anderson K, et al. Analysis of spectroscopic radiation portal monitor data using principal components analysis[J]. IEEE Transactions on Nuclear Science, 2006, 53(3):1418-1425. [5] Dragovic S, Onjia A. Classification of soil samples according to their geographic origin using gamma-ray spectrometry and principal component analysis[J]. Journal of Environmental Radioactivity, 2006, 89:150-158. [6] Heggemanna T, Welpa G, Amelung W, et al. Proximal gamma-ray spectrometry for site-independent in situ prediction of soil texture on ten heterogeneous fields in Germany using support vector machines[J]. Soil & Tillage Research, 2017, 168:99-109. [7] Varley A, Tyler A, Smith L, et al. Mapping the spatial distribution and activity of <sup>226</sup>Ra at legacy sites through machine learning interpretation of gamma-ray spectrometry data[J]. Science of the Total Environment, 2016, 545:654-661. [8] Wang C L, Funk L, Riedel R, et al. Improved neutron-gamma discrimination for a <sup>3</sup>He neutron detector using subspace learning method[J]. Nuclear Instruments and Methods in Physics Research A, 2017, 853:27-35. [9] Hata H, Yokoyama K, Ishimori Y, et al. Application of support vector machine to rapid classification of uranium waste drums using low-resolution γ-ray spectra[J]. Applied Radiation and Isotopes, 2015, 104:143-146. [10] 杨杨. 基于模糊聚类的地层岩性测井识别研究与配套软件研制[D]. 成都:西南石油大学, 2014.YANG Yang. Research and Development of Formation Lithology Logging Identification and Matching Software Based on Fuzzy Clustering[D]. Chengdu:Southwest Petroleum University, 2014. (in Chinese) [11] 周志华. 机器学习[M]. 北京:清华大学出版社, 2016.ZHOU Zhihua. Machine Learning[M]. Beijing:Tsinghua University Press, 2016. (in Chinese) [12] Han J, Kamber M, Pei J. Data Mining:Concepts Techniques[M]. Singapore:Elsevier (Singapore) Pte Ltd, 2012.
