Oliveira, Hermeson
The results obtained by the use of this methodology were compared with the pixel-per-pixel technique of
Maximum Likelihood (table 4). The percentage of concordance of correct classified areas between the
proposed cluster methodology and pixel-per-pixel classification was 70% (table 5). This reinforces the
validity of the proposed methodology.
Forest1 {soil agricuture |agriculture | Agriculture | Agriculture |forest2 | urban Rej
1 2 3 4
Forest1 94,12% 0 0 0 0 5,88% 0 0 0
Soil 0 10096 0 0 0 0 0 0 0
Agriculture 0 0 98,13% 0 0 0 0 1,87% 1-0
1
Agriculture 0 0 0,32% 90,35% 6,43% 0 0,320012, 572] 0
2
Agriculture 0 0 0,20% 1,39% 97,43% 0 0 0,99% | 9
3
Agriculture | 10,62% 0 0 0 0 87,50% 1,88% 0 0
4
Forest2 2,12% 0 0,47% 1,18% 0 1,41% 84,94 |9,88% | 0
%
Urban 0 0 6,75% 11,5596 2,4096 0,2296 14,38 |64,71 0
9b 96
Table 4: Errors matrix of MaxVer method
Concordância(%)
Forest1 71,27
Soil 87,97
Agriculture 1 69,94
Agriculture? 57,82
Agriculture 3 75,26
Agriculture4 68,01
Forest2 58,37
Urban area 44,39
Table 5: Concordance percentage
6 CONCLUSION
In this work, a methodology of segmentation and classification of Landsat-TM image were presented. First
the image is segmented and later, using non-supervised and supervised classification techniques, the image is
classified. Another objective of this work was a comparative analysis between the pixel-per-pixel
classification and the region classification.
The method presented needs less effort by the user to obtain the concordance levels showed on table 6. That’s
because there isn’t necessary a great number of samples, saving lots of time.
The areas obtained with the segmentation technique using region-growing algorithm, showed better results
than the pixel classification. There aren’t isolated and non-classified pixels and the limits of the segmented
area have few faults.
Concluding, we can consider the general results showed by using region classification were satisfactory, The
methodology shows good potential to be used in similarly activities. It could be used as basis to map
generation in a way faster than conventional classification.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1071