2.5. Hyperspectral data-cube analysis
2.5.1. Modeling
For each mango slice, the mean relative reflectance spectrum
was extracted by averaging the spectral signals of all pixels. All
mean spectra without any spectral pretreatments were used for
subsequent analysis. The relationship between the mean spectra
obtained and the reference moisture contents were established
by partial least squares (PLS) regression. PLS is a linear regression
approach widely used in multivariate analysis. The outstanding
property of PLS is to relate one or several responses Y to a large
number of dependent variables X that can be correlated (Liu,
Sun, & Zeng, 2013). In the current study, PLS models performing
in full-range wavebands and selected wavebands were developed.
The robustness of the developed PLS models for calibration,
cross-validation and prediction was characterized by R2 and RMSE