近红外定量模型快速测定大米的营养成分

Near Infrared Spectral Quantitative Model of the Nutrient Content in Rice

大米中的蛋白质、脂肪、碳水化合物和水分是大米的重要营养成分,这些组分的检测需要依赖国标法,检测过程繁琐冗长,污染大,实现近红外快速检测具有重要的现实意义。本实验以75个大米样品为研究对象,用国标法测定其蛋白质、脂肪、碳水化合物和水分的含量,用AntaresⅡ型近红外光谱仪采集光谱信息。将样品分为校正集和验证集,其中校正集63个,验证集12个,通过6种光谱预处理方法和3种建模方法分别建立大米的蛋白质、脂肪、碳水化合物和水分的定量模型,用Workflow调用模型来实现四种组分的快速检测。结果表明,蛋白质、碳水化合物通过一阶导数与Savitzky-Golay滤波对光谱进行处理;脂肪、水分通过二阶导数和Savitzky-Golay滤波对光谱进行处理;再通过PLS回归与国标法的测定值相关联建立的大米中蛋白质、脂肪、碳水化合物及水分的定量模型具有较高的预测精度。蛋白质、脂肪、碳水化合物和水分的预测模型的内部交叉验证的相关系数R分别为0.9266、0.9333、0.9198、0.9175,RMSECV分别为0.280、0.133、0.362、0.270,内部交叉验证的相关系数R均在0.91以上,RMSECV均小于0.37,可实现对大米中蛋白质、脂肪、碳水化合物和水分的快速检测。

Protein, fat, carbohydrate and moisture in rice are important nutrients in rice. The detection of these components needs to rely on the national standard method. The detection process is tedious and has large pollution. In this experiment, 75 rice samples were studied, and their contents of protein, fat, carbohydrate and moisture were measured by national standard method. Spectral information was collected by AntaresⅡinfrared spectrometer. Samples were divided into verification set and correction set, including 63 samples in correction set and 12 samples in verification set. A quantitative model of protein, fat, carbohydrate and moisture of rice was established through six spectral pretreatment methods and three modeling methods, and the model by running workflow was used to realize rapid detection of four components. The results showed that proteins and carbohydrates followed by the first derivative and Savitzky-Golay filtering, while fat and water followed by the second derivative and Savitzky-Golay filtering. The quantitative model of protein, fat, carbohydrate and water content in rice was established by Least partial square method in combination with the measured values by national standard method and has high prediction accuracy. The internal cross-validation coefficient of the prediction model of protein, fat, carbohydrate and moisture is 0.9266, 0.9333, 0.9198, 0.9175, and RMSECV is 0.280, 0.133, 0.362 and 0.270, respectively. The internal cross-validation R value is above 0.91 and RMSECV is less than 0.35, which can realize the rapid detection of protein, fat, carbohydrate and moisture in rice.