油桐籽含油率近红外光谱检测模型的构建  

Modeling on determination of oil content of Vernicia fordii seeds by near-infrared spectroscopy

作　　者：李水芳[1] 李一帆 付红军[3] 李姣娟[1] 

机构地区：[1]中南林业科技大学理学院,长沙410004 [2]中南林业科技大学材料科学与工程学院,长沙410004 [3]中南林业科技大学食品科学与工程学院,长沙410004

出　　处：《林业工程学报》2017年第6期45-49,共5页Journal of Forestry Engineering

基　　金：湖南省教育厅重点项目(14A155)

摘　　要：为实现油桐籽含油率的快速检测,采用近红外光谱结合化学计量学方法对油桐籽含油率的测定进行了研究。107个样本用Kennard-Stone法划分为校正集(80个)和验证集(27个)。光谱经预处理方法优化,确定一阶导数结合均值中心化预处理最优。分别采用竞争性自适应重加权算法筛选变量及小波变换压缩变量,比较了偏最小二乘法与径向基神经网络法所建模型的预测性能,确定竞争性自适应重加权算法筛选出的8个变量用于偏最小二乘法建模,所建模型预测性能最好:验证集相关系数0.927,均方根误差2.08,相对标准偏差为3.99%。结果表明竞争性自适应重加权算法筛选变量结合偏最小二乘法建模,所建模型简单,精度较好,可用于油桐籽含油率的快速检测。Near-infrared spectroscopy（ NIR） and chemometrics methods were used for a rapid determination of oil content of Vernicia fordii seeds. There were 107 samples,including 21 V. montana Wils and 86 V. fordii Hemsley,being collected from tung oil tree germplasm in Yongshun County of Hunan Province. The near-infrared spectra of samples were collected by using scattered reflection mode through a antaris Ⅱ near-infrared spectrophotometer in the range of10 000-4 000 cm-1. The oil content was determined by Soxhlet extraction. The 107 samples were divided into a calibration set（ 80） and a validation set（ 27） by Kennard-Stone algorithm. A combination of first derivative coupled with mean centering was utilized as an optimized spectral pretreatment method. Eight key variables were selected by competitive adaptive reweighted sampling（ CARS）,and their wavenumber of correspondence were 4 019. 3,4 023. 1,4 088. 7,4 196. 7,4 917. 8,5 762. 2,5 766. 0 and 5 847. 0 cm-1. Wavelet transform（ WT） was adapted to compressed spectral data. Partial least squares（ PLS） and radial basis function neural networks（ RBFNN） were used to develop calibration models. The PLS combined with eight variables was finally used as the optimal model. The correlation coefficient（ R）,root mean square error prediction（ RMSEP） and relative standard deviation（ RSD） of validation set were 0. 927,2.08 and 3. 99%,respectively. The results showed that the accuracy of oil content prediction was improved by using NIR model combining PLS with CARS method. The method was suitable for the rapid determination of oil content of V. fordii seeds.

关 键 词：油桐籽 含油率 近红外光谱 化学计量学 

分 类 号：S794.3[农业科学—林木遗传育种] O657.3[农业科学—林学]