摘要
采用太赫兹时域光谱系统(THz-TDS),研究了4种食用油(黑芝麻油、芝麻油、小磨香油和花生油)在0.2~1.6 THz波段的延时特性和折射率特性。使用主成分分析法(PCA),根据累计贡献率的大小提取光谱的特征数据。提取了4个主成分(累计贡献率大于95%)作为一个支持向量机(SVM)模型的输入用于识别食用油的种类。结果表明:结合主成分分析法,通过选择合适的支持向量机核函数及其参数,食用油种类识别的正确率可达到93%;通过与主成分回归(PCR)、偏最小二乘回归(PLS)和后向(BP)神经网络方法的比较,支持向量机结合主成分分析(PCA-SVM)方法具有更突出的分类性能,同时也说明了采用太赫兹时域光谱,结合化学计量学方法精准鉴别食用油种类的可行性。
Delay characteristics and refractive index characteristics of four kinds of edible oils （black sesame oil, traditional sesame oil, sesame oil, peanut oil） in the range of 0.2-1.6 THz were investigated by terahertz time-domain spectroscopy（THz-TDS）.Principal component analysis （PCA） was employed to extract feature data according to the accumulative contribution rates. The top four principal components （accumulative contribution rate above 95%） were selected, and then a support vector machine （SVM） method was applied. The results showed that by choosing the appropriate kernel function and its parameters of SVM, the samples were identified with an accuracy of 93%.Furthermore, compared with principal component regression, partial least squares regression, and back-propagation neural networks, PCA-SVM had a more prominent classification performance and also indicated that the THz-TDS technology combined with PCA-SVM was efficient and feasible for identifying different kinds of edible oils.
出处
《中国油脂》
CAS
CSCD
北大核心
2017年第7期69-73,共5页
China Oils and Fats
基金
国家863计划项目(2012A101608)
河南省基础与前沿计划项目(152300410079)
关键词
食用油
太赫兹时域光谱系统
主成分分析
支持向量机
预测模型
edible oil
terahertz time - domain spectroscopy
principal component analysis
support vector machine
prediction model
