为实现对多样本新鲜羊肉营养含量的快速检测,该研究利用近红外光谱(near-infrared reflectance spectroscopy,NIRS)技术构建了新鲜羊肉中6种营养成分的定量分析模型。于武威市民勤县采集203份新鲜羊肉,并测定其水分(moisture,MT)、粗脂...为实现对多样本新鲜羊肉营养含量的快速检测,该研究利用近红外光谱(near-infrared reflectance spectroscopy,NIRS)技术构建了新鲜羊肉中6种营养成分的定量分析模型。于武威市民勤县采集203份新鲜羊肉,并测定其水分(moisture,MT)、粗脂肪(ether extract,EE)、粗蛋白(crude protein,CP)、葡萄糖(glucose,Glu)、粗灰分(crude ash,Ash)及总磷(phosphorus,P)的含量。使用WINISI III与Foss Calibrator定标软件分别建立羊肉6种营养成分的NIRS模型并对其结果进行比较。WINISI III软件定标结果显示,羊肉MT、EE、CP预测模型的预测决定系数(coefficient of determination for validation,RSQ)和外部验证相对分析误差(ratio of performance to deviation for vali-dation,RPD)分别为0.83与2.47、0.90与3.60、0.81与2.79;Glu、Ash预测模型的RSQ和RPD分别为0.54与3.05、0.54与1.91;P预测模型的RSQ和RPD为0.45与1.80。Foss Calibrator软件定标结果显示,MT、EE、CP的交互验证均方根误差[root mean square error of cross-verification,RMSEP(cross)]和决定系数(coefficient of determination,R^(2))分别为0.631与0.84、0.326与0.87、0.468与0.83;Glu、Ash的RMSEP(cross)和R^(2)分别为0.127与0.53、0.179与0.51;P的RMSEP(cross)和R^(2)为0.086与0.33。2种定标软件得到的结论基本一致,均表明MT、EE、CP的预测模型可在实际生产中精确预测;Glu、Ash的预测模型可在大量样品的粗略分析与筛选时应用,但还需继续优化;P的预测模型相关性较差,不能在实际生产中应用。展开更多
Near-infrared reflectance spectroscopy (NIRS) was applied to classify grape wines of different geographical origins (Changli, Huailai, and Yantai, China). Near infrared (NIR) spectra were collected in transmission mod...Near-infrared reflectance spectroscopy (NIRS) was applied to classify grape wines of different geographical origins (Changli, Huailai, and Yantai, China). Near infrared (NIR) spectra were collected in transmission mode in the wavelength range of 800-2500 nm. Wines (n=90) were randomly split into two sets, calibration set (n=54) and validation set (n=36). Discriminant analysis models were developed using BP neural network and discriminant partial least-squares discriminant analysis (PLS-DA). The prediction performance of calibration models in different wavelength range was also investigated. BP neural network models and PLS-DA models correctly classified 100% of the wines in calibration set. When used to predict wines in validation set, BP neural network models correctly classified 100%, 81.8%, and 90.9% of the wines from Changli, Huailai, and Yantai respectively, and PLS-DA models correctly classified 100% of all samples. The results demonstrated that NIRS could be used to discriminate Chinese grape wines as a rapid and reliable method.展开更多
文摘为实现对多样本新鲜羊肉营养含量的快速检测,该研究利用近红外光谱(near-infrared reflectance spectroscopy,NIRS)技术构建了新鲜羊肉中6种营养成分的定量分析模型。于武威市民勤县采集203份新鲜羊肉,并测定其水分(moisture,MT)、粗脂肪(ether extract,EE)、粗蛋白(crude protein,CP)、葡萄糖(glucose,Glu)、粗灰分(crude ash,Ash)及总磷(phosphorus,P)的含量。使用WINISI III与Foss Calibrator定标软件分别建立羊肉6种营养成分的NIRS模型并对其结果进行比较。WINISI III软件定标结果显示,羊肉MT、EE、CP预测模型的预测决定系数(coefficient of determination for validation,RSQ)和外部验证相对分析误差(ratio of performance to deviation for vali-dation,RPD)分别为0.83与2.47、0.90与3.60、0.81与2.79;Glu、Ash预测模型的RSQ和RPD分别为0.54与3.05、0.54与1.91;P预测模型的RSQ和RPD为0.45与1.80。Foss Calibrator软件定标结果显示,MT、EE、CP的交互验证均方根误差[root mean square error of cross-verification,RMSEP(cross)]和决定系数(coefficient of determination,R^(2))分别为0.631与0.84、0.326与0.87、0.468与0.83;Glu、Ash的RMSEP(cross)和R^(2)分别为0.127与0.53、0.179与0.51;P的RMSEP(cross)和R^(2)为0.086与0.33。2种定标软件得到的结论基本一致,均表明MT、EE、CP的预测模型可在实际生产中精确预测;Glu、Ash的预测模型可在大量样品的粗略分析与筛选时应用,但还需继续优化;P的预测模型相关性较差,不能在实际生产中应用。
文摘Near-infrared reflectance spectroscopy (NIRS) was applied to classify grape wines of different geographical origins (Changli, Huailai, and Yantai, China). Near infrared (NIR) spectra were collected in transmission mode in the wavelength range of 800-2500 nm. Wines (n=90) were randomly split into two sets, calibration set (n=54) and validation set (n=36). Discriminant analysis models were developed using BP neural network and discriminant partial least-squares discriminant analysis (PLS-DA). The prediction performance of calibration models in different wavelength range was also investigated. BP neural network models and PLS-DA models correctly classified 100% of the wines in calibration set. When used to predict wines in validation set, BP neural network models correctly classified 100%, 81.8%, and 90.9% of the wines from Changli, Huailai, and Yantai respectively, and PLS-DA models correctly classified 100% of all samples. The results demonstrated that NIRS could be used to discriminate Chinese grape wines as a rapid and reliable method.
