Oliveira, Hermeson
Figure 5. Segmented image
On the image classification step, two distinct processes were used, as mentioned earlier. On the first process,
non-supervised classification, the algorithm has as criteria the Mahalanobis distance (INPE, 1996), that is the
same as minimum distance, except by the use of the covariance matrix. It was a cluster algorithm. From the
segmented image of the earlier step, it tries to find regions that are similar. Using this algorithm 24 distinct
regions or classes whose pixels showed similar spectral characteristics were found, as shown in figure 6.
: M ero. ? ig E
Figure 6. Classified image using cluster algorithm
The second process, supervised classification, used samples of the result obtained in the earlier classification
process as training fields. The criteria used was the Bhattacharya distance (INPE, 1996). The results are
shown in figure 7.
Figure 7: Classified image using supervised classification
By the analysis of the results we find out that the thematic class *urban area" showed the lower percentage of
efficiency. That's because the urban area of Palotina City is full of trees higher than 10 meters. The asphaltic
surface is cover by the shadows of the trees most of the time. This peculiar characteristic contributes to the
lower result on classifying this class and also interfered, with less intensity, in the classification of other
classes.
1070
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.