何欣龙,王继芬,张倩,唐敏力,何亚.化学通报,2019,82(2):169-174.
基于多分类模型的记号笔墨水红外光谱分析
The Research on Marker pens Ink Infrared Spectroscopy based on Multi-classification model
投稿时间:2018-07-25  修订日期:2018-08-15
DOI:
中文关键词:  红外光谱  记号笔墨水  判别分析  径向基函数神经网络  K近邻
英文关键词:IR, Marker  pens ink, Discriminant  analysis, Radial  basis function  neural network, K-nearest  neighbor Serum  protein
基金项目:广东省化学危害应急监测技术重点实验室开放基金(KF2018002)
作者单位E-mail
何欣龙 中国人民公安大学刑事科学技术学院 1078683050@qq.com 
王继芬* 中国人民公安大学刑事科学技术学院 wangjifen58@126.com 
张倩 中国人民公安大学刑事科学技术学院  
唐敏力 四川省攀枝花市公安局东区分局  
何亚 中国人民公安大学刑事科学技术学院  
摘要点击次数: 1935
全文下载次数: 0
中文摘要:
      记号笔墨水的区分鉴别在相关案件的侦破和诉讼中具有重要意义。本实验采用红外光谱法(FTIR-ATR)获取记号笔的原始光谱,并对原始光谱分别进行自动基线校正、Savitzky-Golay平滑、峰面积归一化和小波阈值去噪四种预处理消除噪声等干扰因素并确定特征波长,同时结合判别分析、径向基函数神经网络和K近邻算法构建分类模型。结果表明,三种模型对黑色笔的分类最准确,均实现了100%的识别,对红蓝色笔区分能力次之,相比较DA和RBF,KNN模型分类精度最高。以K近邻算法(KNN)为主,判别分析(DA)和径向基函数网络(RBF)为辅,对各色水性和油性笔展开进一步区分,黑蓝色水性与油性笔的三种模型均实现了100%的准确归类,红色水性与油性笔的DA模型、RBF模型和KNN模型(训练和测试样本)的分类准确率分别为94.70%、100%、100%和75%,在KNN交互式图中,黑色水性与油性笔分布最为集中,聚敛程度最好,红色笔次之,蓝色笔较为分散,结合KNN模型,得到黑色笔的详细分类结果,分类结果理想。综上,采用FTIR-ATR结合DA-RBF-KNN法能为记号笔的类型准确检测提供新的分析手段,且模型检测精度高,方法具有普适性和一定的借鉴意义。
英文摘要:
      It’s especially important for identifying the marker pens ink in detection and litigation some cases, the paper acquired the original spectrum of the marker pens ink by FTIR-ATR. In order to eliminate noise and other interference factors and determine the characteristic wavelength range, it used automatic baseline correction, Savitzky-Golay smoothing, peak area normalization and wavelet threshold denoising to Pretreat the original spectrum. Simultaneously, combined with Discriminant Analysis, Radial Basis Function Neural Network and K Nearest Neighbor Algorithm to Construct Multi-classification Model. The results showed that the three models have the most accurate classification of Black pens and achieve 100% recognition, the Red and Blue pens are following, compared with DA and RBF, KNN model has the highest classification accuracy. Mainly based on KNN, supplemented by DA and RBF, the paper further differentiates water and oil pens of them. The three models of Black and Blue water and oily pens all achieved 100% accuracy. In the Red, DA, RBF and KNN model accuracy (training and testing samples) respectively were 94.70%, 100%, 100% and 75%. In the KNN interactive graph, the Black water and oily pens are the most concentrated, the following is the Red, the Blue is more scattered. In summary, using FTIR-ATR combined with DA-RBF-KNN can provide a new analysis method for more accurate detection of marker pens ink, and the accuracy of model detection is higher. The method has universality and certain reference significance.
查看全文  查看/发表评论  下载PDF阅读器
关闭