摘要
以过氧化值为参考,用偏最小二乘判别分析(PLSDA)和最小二乘支持向量机(LSSVM)两种方法对食用植物油的近红外光谱进行建模和预测,初步鉴别食用植物油的质量。分析了数据中心化、方差比例、正交信号校正这三种不同的前处理方法对PLSDA预测结果的影响。对同种划分训练集和验证集方法的PLSDA,数据中心化的预测结果都较好。同时分析了Kennard-stone(KS)法及SPXY法这两种不同的划分训练集和验证集方法对PLSDA和LSSVM的预测结果的影响,结果表明训练集和验证集的划分方法对两者影响均较小。实验比较了PLSDA与LSSVM两种方法的判别能力,结果显示在以过氧化值为参考的基础上,两种方法判别能力相近,最高判别总正确率均达到94.3%。
On the basis of peroxide value,partial least squares discriminant analysis（PLSDA）and least squares support vector machines（LSSVM）were used to model and predict the near-infrared（NIR）spectra of edible vegetable oils;then the quality of edible vegetable oils was primarily identified.Firstly,three different pretreatment methods（mean center,variance scaling,and orthogonal signal correction）were applied to PLSDA,and the effect of pretreatment was studied.For the same selection method of PLSDA,the prediction results of mean center were all preferable.Moreover,the predicted results by PLSDA and LSSVM were analyzed for two different selection of calibration set methods,Kennard-stone and SPXY（Sample Set Partitioning Based On Joint X-Y Distance）selection,and it showed that the influence on both PLSDA and LSSVM were all mild.Finally,PLSDA and LSSVM were compared,and the results showed that the discriminant abilities of both were similar on the basis of the peroxide value as the reference method and the best overall accuracy rate of both PLSDA and LSSVM was 94.3%.
出处
《分析科学学报》
CAS
CSCD
北大核心
2015年第6期763-768,共6页
Journal of Analytical Science
基金
广东省教育厅产学研结合项目(No.2007A090302100)
关键词
近红外光谱
偏最小二乘判别分析
最小二乘支持向量机
食用植物油
过氧化值
Near-infrared spectroscopy
Least squares support vector machines
Partial least squares discriminant analysis
Edible vegetable oil
Peroxide value
