基于最小二乘支持向量机的番茄汁糖酸度分析研究

NIR Spectroscopy Based on Least Square Support Vector Machines for Quality Prediction of Tomato Juice

近红外光谱应用于农产品内部品质无损检测的方法引起人们的广泛关注,在分析过程中建立一个稳定可靠的模型用于处理非线性数据集是十分重要的,也是有一定难度的。目前常用的偏最小二乘(PLS)、主成分回归(PCR)以及逐步多元线性回归(SMLR)等方法还不能解决这类问题。文章提出了将基于统计学原理的最小二乘支持向量机(LS-SVM)回归方法用于番茄汁的近红外(NIR)光谱分析,预测番茄汁品质(糖度和有效酸度)。运用LS-SVM方法以67个番茄汁样本建模,采用高斯径向基函数(RBF)为核函数,对33个样本进行糖酸度预测,糖度的相关系数为0.99025,均方根标准预测误差为0.0056°Brix;有效酸度的相关系数为0.9675,均方根标准预测误差为0.0245。结果表明,LS-SVM方法要优于PLS和PCR建模方法,是一种快速、准确的近红外光谱分析方法。

The application of least square support vector machines （LS-SVM） regression method based on statistics study theory to the analysis with near infrared （NIR） spectra of tomato juice was introduced in the present paper. In this method, LS-SVM was used for establishing model of spectral analysis, and was applied to predict the sugar contents （SC） and available acid （VA） in tomato juice samples. NIR transmission spectra of tomato juice were measured in the spectral range of 800-2 500 nm using In-GaAs detector. The radial basis function （RBF） was adopted as a kernel function of LS-SVM. Sixty seven tomato juice samples were used as calibration set, and thirty three samples were used as validation set. The results of the method for sugar contents （SC） and available acid （VA） prediction were： a high correlation coefficient of 0. 990 3 and 0.967 5, and a low root mean square error of prediction （RMSEP） of 0. 005 6° Brix and 0. 024 5, respectively. And compared to PLS and PCR methods, the perform- ance of the LS-SVM method was better. The results indicated that it was possible to built statistic models to quantify some common components in tomato juice using near-infrared （NIR） spectroscopy and least square support vector machines （LS-SVM） regression method as a nonlinear multivariate calibration procedure, and LS-SVM could be a rapid and accurate method for juice components determination based on NIR spectra.