Rapid determination of maturity in apple using outlier detection and calibration model optimization.

摘要

A non-destructive method to predict the maturity (ripeness) and quality of intact apples is described using Fourier transform NIR spectroscopy at 814-1100 nm. Mathematical models for calibration and prediction of sugar content (SC) and titratable acidity (TA), indices of maturity, were developed using partial least squares regression. Modelling procedures were systematically studied with a focus on outlier detection and calibration model optimization. 321 optimum sample sets were successfully chosen from an original sample set of 333 using 2 outlier detection techniques. Optimized regression models for maturity were obtained at 814-1100 nm with correlation coeff. of 0.95 and 0.74 and s.e. of prediction of 0.54 and 0.04 for SC and TA, respectively. Results obtained using this method were comparable with those obtained using reference methods (no significant difference at the 0.05 level). Outlier detection methods are useful for optimizing FT-NIR calibration models and should improve the accuracy of prediction models.