基于红外光谱多元分析的记号笔种类鉴别研究

Research on identification of marker types based on multivariate analysis

借助衰减全反射红外光谱结合K近邻算法和层次聚类,建立记号笔墨水种类鉴别的分类模型,为记号笔的种类鉴别提供有效的新方法。研究基于记号笔墨水的外光谱矩阵数据,通过建立主成分和判别分析分析模型和K近邻算法分类模型,实现对模型性能的比较和对模型分类结果的验证。实验结果表明,以水性和油性作为分类标准,模型对样本的区分能力好,其正确率为100%。借助红外谱图进一步分析水性油性样本时,其最强峰与其对应溶剂相符合。借助K近邻算法进行验证性分析,按重要性加权特征给不同的样本施加不同的权重,运用训练样本即为测试样本的方法交互验证,选取K值为1,训练集∶保持集=3∶1,建立分类模型,模型总分类准确率达100%,区分效果良好,不同品牌的油性、水性样本能被聚类为一组。综上,衰减全反射红外光谱结合K近邻算法和层次聚类可作为记号笔墨水种类鉴别的一种快速准确的分析方法。

A classification model for discriminating the types of marker ink is established with the use of attenuated total reflection infrared spectroscopycombined with the K-Nearest Neighbor Algorithm and hierarchical clustering,which provides an effective new method for identifying the types of marker.Based on the external spectrum matrix data of marker ink,the principal component,discriminant analysis model and the K-Nearest Neighbor Algorithm classification model,we achieveda comparison of model performance and verification of model classification results.The experimental results show that the model has good ability to distinguish samples and achieves 100%accuracy using waterborne and oiliness as classification criteria.When using infrared spectroscopy to further analyze aqueous or oily samples,their peaks coincide with corresponding solvent.Then K-Nearest Neighbor Algorithm is used for confirmatory analysis.Different samples are weighted by importance weighting feature,and the training samplesare used to verify the test samples interactively.Value K is selected as 1,the training set:keep set=3:1.A classification model is established,which has good discrimination and reaches 100%total accuracy of classification withaqueous and oily samples of different brands be clustered into one group.In summary,attenuated total reflectance infrared spectroscopy combined with K Nearest Neighbor Algorithm and hierarchical clustering can be used as a fast and accurate method for identifying the types of marker ink.