基于可见/近红外光谱的南丰蜜桔可溶性固形物预测模型优化研究

Optimization of prediction model for detecting SSC of Nanfeng mandarin based on the visible/near-infrared spectroscopy

利用可见/近红外光谱技术对蜜桔可溶性固形物含量进行检测研究。采用QualitySpec型光谱仪采集蜜桔的可见/近红外透射光谱,结合多种预处理方法、三种变量优选方法以及三种建模方法进行对比研究。经比较发现,采用CARS(competitive adaptive reweighted sampling)结合多元线性回归(MLR)建立的蜜桔SSC模型效果有很大提升,变量数从401减少到27,预测相关系数由0.949上升为0.973,预测均方根误差下降了0.135,因子数下降到1。结果表明,可见/近红外漫透射光谱结合CARS方法联合MLR建模能有效优化蜜桔SSC预测模型。

Abstract：The content of mandarin soluble solids was detected and studied through using visible/near-infrared spectroscopy technology. The visible/near infrared transmission spectroscopy of mandarin taken by QualitySpec type Spectrometer was compared and analysed through combining with a variety of pre-treatment method, three ways of select variables and three kinds of modeling methods. After comparison,the effect of mandarin SSC model built by competitive adaptive reweighted sampling（CARS） combined with multiple linear regression （MLR） was greatly improved. The number of variables was reduced from 401 to 27,predicted correlation coefficient rised from 0.949 to 0.973,forecast the root mean square error decreased 0.135,factor number droped to 1. The result indicated that mandarin prediction model based on the visible/near-infrared diffuse transmittance spectroscopy combined with CARS jointed MLR modeling could be effectively optimized.