摘要
为了实现对案件现场常见食品包装纸的快速分类及认定,提出一种基于X射线荧光光谱(XRF)结合深度学习算法的食品包装纸可视化检验方法。首先,采用XRF检验44个不同来源的食品包装纸样本中的无机元素,并根据主要构成元素的含量,对其进行人工分类和系统聚类分析。其次,分别使用主成分分析和t分布随机邻域嵌入两种降维算法处理数据以检验聚类效果,并实现数据分类可视化。最后,随机选取80%的样本作为训练集构建人工神经网络,并进行相关实验。实验结果表明,所提方法在测试集上的分类正确率为88.9%,可以为未来公安业务实际应用提供参考。
To quickly classify and identify common food packaging paper at the scene of the case, a visual inspection method of food packaging paper based on X-ray fluorescence spectroscopy(XRF) and deep learning algorithm is proposed. First, the inorganic elements in 44 samples of food packaging paper from different sources were detected via XRF, and artificial classification and cluster analysis were performed based on the content of the main constituent elements. Second, to test the clustering effect and visualize data classification, two-dimensionality reduction algorithms,principal component analysis, and t-distribution random neighborhood embedding are used. Finally, 80% of the samples are randomly selected as the training set to construct the artificial neural network, and relevant experiments are carried out. The experimental results show that classification accuracy of the proposed method on the test set is 88. 9%, which can be used as a reference for future practical applications of public security business.
作者
郭琦
姜红
杨金颉
吴克难
满吉
Guo Qi;Jiang Hong;Yang Jinjie;Wu Kenan;Man Ji(Institute of Criminal Investigation,People’s Public Security University of China,Beijing 100038,China;Institute of Computer Science and Technology,Wuhan University of Technology,Wuhan,Hubei 430070,China;Beijing Huayi Honrizon Technology Co.,Ltd.,Beijing 100123,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2022年第4期458-464,共7页
Laser & Optoelectronics Progress
基金
国家重点研发计划(2017YFC0822004)
中国人民公安大学2019年度基科费重点项目(2021JKF212)。
关键词
X射线光学
X射线荧光光谱
系统聚类
主成分分析
t分布随机邻域嵌入
多层前馈神经网络
X-ray optics
X-ray fluorescence spectroscopy
hierarchical clustering
principal component analysis
tdistribution random neighborhood embedding
multilayer feedforward neural network
