作者：季佳华, 卫辰洁, 王继芬, 蒋宇航, 冯源
作者单位：中国人民公安大学侦查与刑事科学技术学院, 北京 102600
Infrared spectrum identification of pigments based on multiple classification model
JI Jiahua, WEI Chenjie, WANG Jifen, JIANG Yuhang, FENG Yuan
School of Investigation and Forensic Science and Technology, People's Public Security University of China, Beijing 102600, China
Abstract: The examination and identification of pigment is an important work in judicial appraisal.In the traditional analysis, investigators often make manual comparison and analysis with the help of the infrared spectrum database, which takes a long time and has a large error, unable to meet the requirements of lossless, rapid and accurate testing of on-site pigment samples. In order to realize nondestructive, rapid and accurate testing and identification of material evidence, this experiment is to propose a testing method. The infrared spectra of 48 pigment samples from different brands were collected and analyzed. The pretreatment was carried out by using multiple scattering correction, savitzky-golay smoothing and peak area normalization, and four classification models based on k-nearest neighbor algorithm were established to realize the classification and classification of different pigments. Compared with k-nearest neighbor and Fisher's discriminant model, the classification accuracy of the multi-layer perceptrons is higher (the overall classification accuracy is 95.8%) and the classification results are better. After extracting characteristic variables through principal component analysis, the classification model can distinguish the two kinds of pigments with an accuracy rate of 100%. The classification model constructed by MLP and PCA is the best for the classification of pigment samples. For the two types of gouache, namely, ordinary gouache and Picasso gouache, the classification accuracy of the multilayer perceptrons model was 97.2%, and for the two brands of ordinary gouache samples (bebio and m&g), the classification accuracy of the multilayer perceptrons model was 100%, with satisfactory experimental results. The accurate identification and differentiation of pigment samples can be realized by using mid-infrared spectrum and multiple classification model, which is fast and non-destructive, reduces the cost of testing and identification, improves the efficiency of testing and identification, and can provide certain reference for the identification and analysis of other physical evidence.
Keywords: pigment;second derivative infrared spectrum;discriminant analysis;identification
2020, 46(6):67-71 收稿日期: 2020-01-08;收到修改稿日期: 2020-02-15
 NA N. Non-destructive and in situ identification of rice paper, seals and pigments by FT-IR and XRD spectroscopy[J]. Talanta, 2004, 64(4): 1000-1008
 WANG Z, LU D, ZHANG D, SUN, et al. Fake modern Chinese painting identification based on spectral-spatial feature fusion on hyperspectral image[J]. Multidimensional Systems and Signal Processing, 2016, 27(4): 1031-1044
 GEERT V S. Ensor’s pigments studied by means of portable and synchrotron radiation-based X-ray techniques: evolution, context and degradation[M]. Dutch: University of Antwerp Press, 2012: 94.
 KASZOWSKA Z, MALEK K, PANCZYK M, et al. A joint application of ATR-FTIR and SEM imaging with high spatial resolution: Identification and distribution of painting materials and their degradation products in paint cross sections[J]. Vibrat Spectr, 2013, 65: 1-11
 STONER J H, REBECCA R. Conservation of Easel paintings[M]. London: Routledge Press, 2012: 310.
 王继芬,高春芳,徐佰祺, 等.鞋底材料的中红外光谱可视化快速鉴别[J].中国塑料,2019,33(8):101-105.
 何欣龙, 刘文浩, 王继芬. 红外光谱结合多元统计学检验汽车前保险杠[J]. 光散射学报, 2018, 30(1): 70-76
 何欣龙, 王继芬, 韩育林, 等. 舰船甲板油漆红外指纹图谱的无损鉴别[J]. 兵器材料科学与工程, 2019, 42(6): 97-101