Firstly, the full range of 500e700 nm was chosen as one of the
wavelength ranges. Next, as in the previous study of Usui et al.
(2005), the range of 500e600 nm was selected to detect white
shelled blood-spot eggs and the discrimination rate was very
high, so this wavelength range was also chosen in this study.
Lastly, according to the correlation between each wave band
and the categorical variable of the samples, the range of
576e579 nm and 607e609 nm were selected as another
modelling spectral range. The discrimination results of the
three models using these different wavelength ranges are
shown in Table 1. It can been seen that the result of the range
of 500e700 nm was worst, probably due to this wavelength
range containing too much uncorrelated information. And the
result of the range of 500e600nm was lower than the previous
research on white eggshell with the possible reason that the
optical properties of protoporphyrin of brown eggshell were
very similar to those of haemoglobin (Usui et al., 2005). The
best result was obtained for the wavelength ranges of
576e579 nm and 607e609 nm. As shown in Fig. 3, the prediction
values of calibration set for the normal eggs were all between
0.5 and 0.5, and only two eggs in the prediction set
were misclassified as blood-spot eggs. However, for artificial
blood-spot eggs, 11 eggs in the calibration set and four eggs in
prediction set were misclassified. According to the spectral
characteristics of the normal eggs and artificial blood-spot
eggs, as shown in Fig. 2, the range of 576e579 nm contained
the absorption peak of blood, and the spectral characteristics
in the range of 607e609 nm which was used as the reference
were very similar.