基于MLP-BDA融合模型的塑钢窗显微共聚焦拉曼光谱分类  

Classification of the Plastic Steel Window Based on Micro-confocal Raman Spectroscopy and MLP-BDA Fusion Model

作　　者：张震 王继芬[1] 鲁朋武 付赵奎 白海涛 ZHANG Zhen;WANG Jifen;LU Pengwu;FU Zhaokui;BAI Haitao(School of Investigation,Renmin University Public Security University of China,Beijing 100038,China;School of Public Security Management of People s Public Security University of China,Beijing 100038,China;Criminal Investigation Detachment of Taiyuan Municipal Public Security Bureau,Shanxi Province,Taiyuan 030000,China)

机构地区：[1]中国人民公安大学侦查学院,北京100038 [2]中国人民公安大学公安管理学院,北京100038 [3]山西省太原市公安局刑侦支队,山西太原030000

出　　处：《塑料工业》2022年第3期121-125,174,共6页China Plastics Industry

基　　金：中央高校基本科研业务费专项资金资助(2021JKF208)。

摘　　要：为了对案件中出现的塑钢窗类检材进行无损、准确的识别与认定,通过实验建立了基于预处理并结合数学建模的分类方法。运用显微共聚焦拉曼光谱分析技术获取了“金鹏”、“瑞恒”等5个品牌共计150份的光谱谱图。基于构建的多层感知器模型,比较了Savitzky-Golay滤波、希尔伯特变换和小波变换三种预处理方式在模型识别精度方面的差异。选择最佳预处理方式,进一步构建Bayesian分类模型对各样本进行分类区分。结果表明,显微共聚焦拉曼光谱能够反映出不同样本理化信息方面存在的差异。预处理能够提升模型的识别精度,其中小波变换处理(96%)>基于希尔伯特变换处理(80%)>基于Savitzky-Golay滤波处理(76%)>未处理(72%)。小波变换没有改变波峰区及其吸光度,但它去除了光谱数据中的噪声。基于小波变换处理结合Bayesian判别分析,成功实现了对150份样本100%的准确区分。In order to identify the steel window materials without damage and accurately,a classification method based on pretreatment combined with mathematical modeling was established.With the help of confocal Raman spectroscopy,150 spectra of 5 brands such as“Jinpeng”and“Ruiheng”were obtained.Based on the constructed multi-layer perceptron model,the difference of model recognition accuracy between Savitzky-Golay filter,Hilbert transform and wavelet transform was compared.The best preprocessing method was selected and the Bayesian classification model was further constructed to distinguish each sample in terms of brand,manufacturer and place of origin.The results show that confocal Raman spectra could reflect the differences in physical and chemical information of different samples,which is the basis for establishing models and carrying out classification work.The pre-processing can improve the recognition accuracy of the model,in which the wavelet transform processing(96%)>Hilbert transform-based processing(80%)>Based on Savitzky-Golay filtering processing(76%)>no processing(72%).The wavelet transform does not change the peak region and its absorbance,but it removes the noise in the spectral data.Based on wavelet transform processing combined with Bayesian discriminant analysis,100%accurate discrimination of 150 samples is successfully achieved.

关 键 词：塑钢窗 显微共聚焦拉曼光谱 预处理 分类模型 

分 类 号：O657.3[理学—分析化学]