Wang Li, Li Zeng-fang, He Yong, Liu Fei
College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310029, China.
Guang Pu Xue Yu Guang Pu Fen Xi. 2008 Aug;28(8):1810-3.
For the fast and exact detection of sugar content of fruit vinegar, near infrared (NIR) spectroscopy technique combined with least squares support vector machines (LS-SVM) algorithm was used to build the prediction model of sugar content in the present research. NIR spectroscopy is a nondestructive, fast and accurate technique for the measurement of chemical compo nents based on overtone and combination bands of specific functional groups. The pivotal step for spectroscopy technique is how to extract quantitative data from mass spectral data and eliminate spectral interferences. Principal component analysis (PCA) is a method which has been widely used in the spectroscopic analysis, and LS-SVM is a new data mining algorithm developed from the machine learning community. In the present study, they were used for the spectroscopic analysis. First, the near infrared transmittance spectra of three hundred samples were obtained, then PCA was applied for reducing the dimensionality of the original spectra, and six principal components (PCs) were selected according the accumulative reliabilities (AR). The six PCs could be used to replace the complex spectral data. The three hundred samples were randomly separated into calibration set and validation set. Least squares support vector machines (LS-SVM) algorithm was used to build prediction model of sugar content based on the calibration set, then this model was employed for the prediction of the validation set. Correlation coefficient (r) of prediction and root mean square error prediction (RMSEP) were used as the evaluation standards, and the results indicated that the r and RMSEP for the prediction of sugar content were 0.9939 and 0.363, respectively. Hence, PCA and LS-SVM model with high prediction precision could be applied to the determination of sugar content in fruit vinegar.
为了快速、准确地检测果醋中的糖分含量，本研究采用近红外（NIR）光谱技术结合最小二乘支持向量机（LS-SVM）算法建立了糖分含量预测模型。近红外光谱是一种基于特定官能团的泛音和组合带测量化学成分的无损、快速且准确的技术。光谱技术的关键步骤是如何从质谱数据中提取定量数据并消除光谱干扰。主成分分析（PCA）是一种在光谱分析中广泛使用的方法，而LS-SVM是机器学习领域发展起来的一种新的数据挖掘算法。在本研究中，它们被用于光谱分析。首先，获取了三百个样品的近红外透射光谱，然后应用PCA对原始光谱进行降维，并根据累积可靠性（AR）选择了六个主成分（PCs）。这六个PCs可用于替代复杂的光谱数据。将三百个样品随机分为校正集和验证集。基于校正集，使用最小二乘支持向量机（LS-SVM）算法建立糖分含量预测模型，然后将该模型用于验证集的预测。以预测相关系数（r）和预测均方根误差（RMSEP）作为评价标准，结果表明，糖分含量预测的r和RMSEP分别为0.9939和0.363。因此，具有较高预测精度的PCA和LS-SVM模型可应用于果醋中糖分含量的测定。
