The results of this study provide us with an effective strategy
for measuring bruise-damage in strawberries from their nearinfrared
hyperspectral images. The developed method
enables estimation of the ratio of bruised to unbruised areas
by classifying hyperpixels in the edible portion of a strawberry.
A fully autonomous multi-band segmentation algorithm
extracts the edible area of the fruit by generating pixel
masks from each band of the hypercube. A decision-fusion
strategy is used to exploit the information from different
spectral bands and the method does not require a priori
information to select these specific bands for generating the
mask. Our results demonstrate that the performance of the
decision-fusion strategy is statistically better than the best
uni-band classifier and is also better than the multi-band
multivariate classifier for bruise detection in strawberries.
In conclusion, the decision-fusionstrategydevelopedin this
paper offers an effective method for exploiting information
from multiple hyperspectral bands. Although this paper
focused on strawberry inspection, it should be noted that the
formulation of the overall inspection strategy is quite general;
and therefore, is applicable to inspecting other biological entities.
Moreover, it is not restricted to dichotomous inspection