The Robustness of PLS Models for Soluble Solids Content of Mangosteen using Near Infrared Reflectance Spectroscopy

摘要

This research investigated the soluble solids content in translucent flesh which is an internal defect of mangosteen fruit (Garcinia mangostana L.). Near infrared (NIR) reflectance Spectroscopy was used for the development of partial least squares calibration models for the prediction of the soluble solids content. Samples of 213 normal mangosteens and 87 translucent mangosteens were used for this experiment. Juice samples extracted from both normal flesh and translucent flesh were measured for theirsoluble solids contents. Diffuse reflectance spectra of juice samples were acquired in a range of 1100 to 2500 nm. In this research, a NIR spectrometer was able to predict the soluble solids content of mangosteen. The best model was developed with second derivative treatment. The calibration model of normal-flesh juice obtained high accuracy for prediction with a set of normal flesh samples (standard error of prediction (SEP) = 0.655, bias = 0.047) and poor accuracy for prediction with a set of translucent flesh samples (SEP = 1.245, bias = 0.224). The results indicate that NIR technology has high potential to develop calibration models for the prediction of the difference in quality of flesh in mangosteen.