近红外光谱法的甜菜糖度快速测定

为了实现甜菜依据含糖量定等分级，甜菜收购环节的按质论价，促进甜菜制糖行业的良好健康发展，应用近红外光谱技术对甜菜糖度的快速检测进行了系统研究，确定了一种快速、无损、准确的测量甜菜糖度的方法。采集具有代表性的28个甜菜品种，820个甜菜样品作为校正集，70个样品作为预测集，扫描得到甜菜校正集样品的近红外原始光谱，选择合适的光谱预处理方法，采用偏最小二乘法建立甜菜糖度的定量预测数学模型，以校正模型的内部交互验证均方根误差(RMSECV)、决定系数(R2)和外部预测标准误差(SEP)为指标对模型的性能进行评价，并对模型的预测效果进行了比较。采用一阶导数和标准正态变量变换对光谱进行预处理并结合偏最小二乘法所建立的定量预测数学模型的预测能力较好。甜菜糖度定量校正数学模型的模型决定系数为0.908 3，内部交互验证预测均方根误差为0.376 7。用此数学模型对预测集70个样品进行预测，预测值与实测值的相关系数达到0.921 4，预测标准误差为0.439，预测值和实测值之间不存在显著性差异(p＞0.05)。结果表明：近红外光谱法作为一种简单、快速、无损、环保的检测方法，能够良好的评价甜菜的糖度。建立的模型具有很高的精确性，可以满足甜菜糖含量测定的需要，该方法可以实现甜菜收购环节的定等分级和按质论价。

Rapid Determination of Beet Sugar Content Using Near Infrared Spectroscopy

In order to classify and set different prices on basis of difference of beet sugar content in the acquisition process and promote the development of beet sugar industry healthily, a fast, nondestructive, accurate method to detect sugar content of beet was determined by applying near infrared spectroscopy technology. Eight hundred twenty samples from 28 representative varieties of beet were collected as calibration set and 70 samples were chosen as prediction set. Then near infrared spectra of calibration set samples were collected by scanning, effective information was extracted from NIR spectroscopy, and the original spectroscopy data was optimized by data preprocessing methods appropriately. Then partial least square(PLS)regression was used to establish beet sugar quantitative prediction mathematical model. The performances of the models were evaluated by the root mean square of cross-validation (RMSECV), the coefficient of determination (R2) of the calibration model and the standard error of prediction (SEP), and the predicted results of these models were compared. Results show that the established mathematical model by using first derivative(FD) and standard normal variate transformation(SNV) coupled with partial least squares has good predictive ability. The R2 of calibration models of sugar content of beet is 0.908 3, and the RMSECV is 0.376 7. Using this model to forecast the prediction set including 70 samples, the correlation coefficient is 0.921 4 between predicted values and measured values, and the standard error of prediction (SEP) is 0.439, without significant difference (p＞0.05) between predicted values and measured values. These results demonstrated that NIRS can take advantage of simple, rapid, nondestructive and environmental detection method and could be applied to predict beet sugar content. This model owned high accuracy and can meet the precision need of determination of beet sugar content. This detection method could be used to classify and set different prices on basis of difference of beet sugar content in the acquisition process.