and lower PA (especially for pixel-based pan-sharpened image
classifications), as compared to homogeneous ones (demonstrating
higher PA in any of the MIU and classifications considered). One
of the greatest differences in PA can be observed in the olive
orchard category. The olive orchard category showed high PA
values, greater than 92%, in all the object-based and pixel + objectbased
multispectral classifications. However, very high PA (up to
99.45%) was also obtained for MC and ML classifiers, considering
pan-sharpened image. In contrast, riverside trees and roads, which
usually exhibit lower intraclass spectral variability, showed a
higher PA for all the multispectral pixel-based classifications,
except for ML for riverside trees, where a higher PA was found for
object-based classifications. Discrimination of burnt crop stubble
land use was very successful when applying the MC method for
any MIU considered in the multispectral image with PA of 99.69%,
98.38% and 98.49%, respectively. Similarly, winter cereal stubble
discrimination was very accurate with PA of over 99.29% or even
of 100% in the multispectral image and for any MIU considered,
when applying ML or MC classifications. When comparing the
performance of object-based and pixel + object-based analyses,
there was no clear trend. For example, in ML classification, the
multispectral object-based classification showed a better PA
in the vineyard category than multispectral pixel-object-based
classification (83.87% and 69.64%, respectively). By contrast, in the
light bare soil category the PA was around 89% for both MIU, and
in the roads category the PA was higher in pixel + object-based
classification than in the object-based classification (78.01% and
45.31%, respectively).
Table 4 shows the UA for every individual land use for the different
MIU and classification methods considered. As stated before