基于显微共聚焦拉曼光谱结合机器学习方法对粉底液的分类研究  

Classification of Liquid Foundation Based on Microconfocal Raman Spectroscopy Combined with Machine Learning

作　　者：倪昕蕾 李春宇[1] 孔维刚[2] Ni Xinei;Li Chunyu;Kong Weigang(School of Investigation,People's Public Security University of China,Beijing,China;Institute of Criminal Science and Technology,Zhengzhou Public Security Bureau,Zhengzhou,China)

机构地区：[1]中国人民公安大学侦查学院,北京 [2]郑州市公安局刑事科学技术研究所,河南郑州

出　　处：《科学技术创新》2024年第5期223-228,共6页Scientific and Technological Innovation

基　　金：中国人民公安大学刑事科学技术双一流创新研究专项(2023SYL06)。

摘　　要：为建立一种快速检验粉底液样品的方法,运用显微共聚焦拉曼光谱和机器学习对不同品牌、色号、价格的粉底液进行分类研究。首先将收集到的50个粉底液样品的拉曼光谱数据利用Savitzky-Golay平滑和归一化算法进行预处理;其次通过主成分分析法进行特征值提取,提取前2个主成分用于后续研究;采用K-Means聚类法将50个样品分成5类,系统聚类法验证分类结果;最后以40个样品为训练集,10个样品为测试集搭建支持向量机(SVM)分类模型。结果表明在Linear核函数下的SVM模型训练集和测试集的准确率可达90%。说明该方法能够实现区分粉底液品牌和价格自动化,为公安机关物证检验、定罪处罚提供新思路。In order to establish a rapid test method for foundation samples,confocal Raman spectroscopy and machine learning were used to classify different brands,colors and prices of foundation.Firstly,the Raman spectral data of 50 foundation samples were preprocessed by Savitzky-Golay smoothing and normalization algorithm.Secondly,eigenvalues were extracted by principal component analysis,and the first two principal components were extracted for subsequent research.The 50 samples were divided into 5 categories by K-Means clustering method,and the classification results were verified by systematic clustering method.Finally,a support vector machine(SVM)classification model was built with 40 samples as training set and 10 samples as test set.The results show that the accuracy of SVM model training set and test set under Linear kernel function can reach 90%.It shows that this method can realize the automation of distinguishing the brand and price of foundation liquid,and provide a new idea for the public security organs'material evidence inspection,conviction and punishment.

关 键 词：显微共聚焦拉曼光谱 粉底液 分类识别 支持向量机 

分 类 号：O436[机械工程—光学工程]