3.3. PLS models at full wavelength range
The PLS regression models for moisture content prediction were
developed at two spectral regions of 400–1000 nm and 880–
1720 nm, respectively. Model performances are displayed in
Fig. 3. In a PLS model, the residual variance is an important factor
in measuring the variations of y values on the linear regression
line. Generally, the plot of y-variance decreases continuously with
the increase in the number of principal components (PCs). An optimal
model could be characterized by using the minimum numbers
of PCs to explain the maximum amount of the variance. In
Fig. 3(a) and (c), the value of y-variance at PC-0 meant the total
variance in the calibration samples. For the PLS model developed
on full-wavelength range of 400–1000 nm, 94.1% of the total variance
could be explained by the first five PCs. After the five PCs, negligible
increase in explained variance could be observed when
using more PCs, that is to say, the model performance could not
be improved significantly by employing more PCs. Similarly, the
first four PCs were selected in the PLS model established on a spectral
range of 880–1720 nm, which explained 97.8% of the total
variance.
The level of moisture content in mango slices during MVD was
predicted by models of FW-PLS-1 (PLS model in full wavelength
(FW) range of 400–1000 nm) and FW-PLS-2 (PLS model in FW
range of 880–1720 nm). The calibration models were validated
by full-cross-validation and the prediction performances are
shown in Fig. 3(b) and (d), respectively. It was observed that the
performance of the FW-PLS-2 model was better than that of the
FW-PLS-1 model, giving a much higher Rc
2 of 0.985 and Rcv
2 of