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

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.