基于滤波器的动植物油光谱信号预处理方法比较及识别分类  

Comparison and recognition of spectral signal pretreatment methods for animal and vegetable oils based on filters

作　　者：邱薇纶[1] 丁圣 QIU Weilun;DING Sheng(School of Forensic Science,Hunan Police Academy,Changsha 410138,China;Network Security Brigade,Tianxin Branch of Changsha Public Security Bureau,Changsha 410004,China)

机构地区：[1]湖南警察学院刑事科学技术系,湖南长沙410138 [2]湖南省长沙市公安局天心分局网安大队,湖南长沙410004

出　　处：《食品与发酵工业》2023年第8期281-288,共8页Food and Fermentation Industries

摘　　要：为实现对动植物油的快速无损检验,探究滤波器在提高光谱分析模型区分能力方面的可行性,该研究借助衰减全反射-表面增强红外吸收光谱技术,采集了动物油(159份)和植物油(188份)共计347份样本的光谱信息数据,构建了Fisher判别分析、支持向量机和决策树3种分类模型。比较了希尔伯特变换、有限长单位脉冲响应滤波器、无限长冲激响应滤波器、快速傅里叶变换和小波变换5种滤波器对3种分类模型精度的影响,同时考察了滤波器窗函数(矩形窗、汉宁窗、海明窗、布莱克曼窗)、小波基函数(Morlet、Dgauss、Mexhat、Haar、Daubechies、Biorthogonal)、滤波方式(低通、高通、带通、带阻)在动植物油样本区分效果方面的差异性。结果发现,滤波器能显著提升光谱分析模型的准确性,低通和带阻滤波方式,矩形窗和布莱克曼窗函数能有效提升模型对样本的区分能力,相较于其他2种算法,支持向量机对各样本的识别区分能力最好。基于FIR低通/带阻滤波器处理后构建的SVM模型(RBF核函数)可作为动植物油样本识别的最佳模型,其对347份样本实现了100%的“类别-品牌”的两级准确区分。综上,滤波器可有效提升光谱分析模型的准确性,结合ATR-SEIRAS光谱信息数据,可准确区分不同的动植物油样本,这为包括动植物油在内的诸多样本的快速无损分析提供了一定参考,为滤波器在提升光谱分析模型方面的应用提供了一定借鉴。To achieve the rapid and non-destructive determination of animal and vegetable oils and explore the feasibility of the filter in improving the discrimination ability of the spectral analysis model,the study collected the spectral information data of 347 samples of animal oil(159 samples)and vegetable oil(188 samples)with the help of attenuated total reflection-surface-enhanced infrared absorption spectroscopy.Three discrimination models,including Fisher discriminant analysis,support vector machine,and decision tree were constructed.Among them,five filters(Hilbert transform,finite length unit impulse response filter,infinite length impulse response filter,fast Fourier transform,and wavelet transform)were considered to discuss the improvement of the model accuracy.Besides,the discrimination differences of filter window functions(rectangular window,Hanning window,Hamming window,and Blackman window),wavelet basis functions(Morlet,Dgauss,Mexhat,Haar,Daubechies,and Biorthogonal)and filtering methods(low-pass,high-pass,band-pass,and band-stop)were investigated and explored.Results showed that filters could significantly improve the models′accuracy.Low-pass/band-stop filtering modes,rectangular window,and Blackman window functions were superior and satisfactory.The support vector machine was optimal in distinguishing all samples.The SVM model(RBF kernel function)based on FIR low-pass/band-stop filter could be considered as the favorable model.It achieved 100%accurate differentiation.Filters and ATR-SEIRAS could effectively distinguish animal and vegetable oils.It could provide the application of enhancing the model performance.

关 键 词：动植物油 衰减全反射-表面增强红外吸收光谱 滤波器 FISHER判别分析 支持向量机 决策树 

分 类 号：O657.3[理学—分析化学] D918.9[理学—化学]