青稞特征营养成分含量快速检测模型的建立及优化

Development and Optimization of a Rapid Detection Model for Characteristic Nutrient Content in Highland Barley

本文以76份青稞为研究对象,利用近红外光谱仪采集青稞4000~10000 cm-1波段光谱,并联合其水分、β-葡聚糖、直链淀粉、蛋白质实测含量数值,构建了基于近红外光谱技术的青稞特征营养成分含量快速检测模型。结果显示,SG卷积平滑(Savitzky Golay,SG)是水分、直链淀粉、β-葡聚糖含量的偏最小二乘法(Partial Least Squares,PLS)预测模型的最优光谱预处理方法,而SG卷积平滑+多元散射校正(Multiplicative Scatter Correction,MSC)是蛋白质含量的偏最小二乘法(PLS)预测模型的最优光谱预处理方法。为进一步提高青稞各成分含量预测模型的准确性,考察了竞争性自适应重加权法(Competitive Adaptive Reweighted Sampling,CARS)、连续投影算法(Successive Projections Algorithm,SPA)和变量组合集群分析混合迭代保留信息变量法(Variables Combination Population Analysis and Iterative Retained Information Variable,VCPA-IRIV)特征波长选择算法对模型预测结果的影响。结果表明,VCPA-IRIV处理可有效提高水分、直链淀粉、蛋白质含量预测模型的预测决定系数,降低预测均方根误差;CARS对β-葡聚糖含量预测模型优化效果显著。基于上述最优方法建立的青稞水分、β-葡聚糖、直链淀粉、蛋白质实测含量预测模型,其预测相关系数分别为0.9868、0.9808、0.9701、0.9879;预测均方根误差分别为0.2042、0.1846、0.8135、0.2095。综上,本研究建立的基于近红外光谱的青稞特征营养成分含量快速检测模型具有较高的准确性,对加工企业快速了解原料品质及高效筛选合格原料有一定指导意义。

In this paper,the highland barley moisture,β-glucan,amylopectin and protein content rapid detection model was established based on the 4000~10000 cm-1 near-infrared spectrum and actual contents of 76 highland barleys.The results showed that SG convolution smoothing was the optimal spectral preprocessing method for the partial least squares(PLS)prediction model of moisture,amylopectin andβ-glucan contents,while SG convolution smoothing+multiplicative scatter correction(MSC)was the optimal spectral preprocessing method for the PLS prediction model of protein content.In order to further improve the accuracy of the prediction model,the different characteristic wavelength selection algorithms including competitive adaptive reweighted sampling(CARS),successive projections algorithm(SPA)and variables combination population analysis and iterative retained information variable(VCPA-IRIV)were used on the prediction results of the model.The results showed that VCPA-IRIV treatment could effectively improve the determination coefficient of prediction of moisture,amylose and protein content prediction model,and reduce the root mean square error of prediction.The treatment of CARS had a remarkable effect on the prediction accuracy for theβ-glucan content.Ultimately,the established prediction models of moisture,β-glucan,amylopectin and protein content for highland barley had good prediction accuracy with the appropriate Rp(0.9868,0.9808,0.9701 and 0.9879)and RMSEP(0.2042,0.1846,0.8135 and 0.2095)value,respectively.In conclusion,the rapid detection model of highland barley characteristic nutrient content based on near infrared spectroscopy established in this study had high accuracy,which would have certain guiding significance for processing enterprises to quickly understand the quality of raw materials and efficiently screen qualified raw materials.