摘要
开展了基于中红外光谱的化学计量学模型对食用植物油掺伪的研究.通过获得市售食用油和以不同质量浓度掺伪食用油的中红外光谱,采用光谱预处理结合模式识别方法进行综合解析,使样本类间距最大化,类内距最小化,结果发现:无监督聚类包括主成分分析、非负矩阵分解均无法准确鉴别;有监督偏最小二乘判别分析的鉴别准确度较高;而二阶导数-非相关线性判别分析(SD-ULDA)对掺伪样本识别的灵敏度和特异度均达到100%.提出依据变量“负荷量”筛选特征波长(变量)新策略,考察了SD-ULDA模型识别准确率随保留波长的变化,同时准确地预测出掺伪比例,为构建高效中红外光谱食用油掺伪判别模型提供理论基础,为提高油脂品质的检测技术提供方法借鉴.
A chemometric model based on mid-infrared spectroscopy was developed to study the adulteration of edible vegetable oil.The mid-infrared spectra of commercial edible oil and adulterated edible oil with different mass concentrations were obtained and analyzed by spectral preprocessing combined with pattern recognition.The results show that unsupervised methods including principal component analysis and nonnegative matrix factorization do not perform accurately;supervised partial least squares discriminant analysis has high discrimination accuracy;the second derivative-uncorrelated linear discriminant analysis(SD-ULDA)maximizes the sample class spacing and minimizes the class spacing as much as possible,and the sensitivity and specificity for the identification of adulterated samples are 100%.A new strategy of characteristic wavelength(variables)screening according to‘load capacity’of variables is proposed,and the change of recognition accuracy of SD-ULDA model with reserved wavelength was investigated.Simultaneously,the adulteration proportion was accurately predicted.It provides a theoretical basis for constructing a high-efficiency mid-infrared spectrum edible oil adulteration discrimination model and provides a reference method for improving the detection technology of oil quality.
作者
李艳坤
许东情
LI Yankun;XU Dongqing(Key Laboratory for Collaborative Control of Multi Pollutants in Flue Gas of Coal-fired Power Plants,Department of Environment Science and Engineering,North China Electric Power University,Baoding 071003,China)
出处
《河北大学学报（自然科学版）》
CAS
北大核心
2022年第6期605-610,共6页
Journal of Hebei University(Natural Science Edition)
基金
中央高校基本科研业务费资助项目(2017MS135)。
关键词
中红外光谱
食用油
掺伪鉴别
非相关线性判别分析
波长筛选
mid-infrared spectrum
edible oil
adulteration identification
uncorrelated linear discriminant analysis
wavelength selection
