Figure 5-18. Classification training data of 35 landscape classification features

Figure 5-17. Landsat 7 ETM image of central Australia

rsmnl
Page Navigation
105

EM 1110-2-2907

1 October 2003

COLOR TRAINING CLASSIFIED % TOTAL % DATA

12 juncus-low-density

13 juncus-high-density

13 juncus-panicum-mix

14 juncus-mixed-clumps-field

16 g1=hd-scol+background+w

17 g2=md-scol+background

18 g4=md-scol+spartina+mud

19 g3=md-scol+spartina+background Purple1

20 g5=ld-scol+mud

21 g1=md-spal+w

22 g2=hd-spal+w

23 g3=hd-spal+w+sppa

24 g4=md-spal+w+sppa

25 g5=hd-spal+mud

26 g6=mixed-spal

26 g7=md-spal+lit+mud

28 g8=md-mixed-spal

29 g1=hd-sppa+mix

30 g2=hd-sppa+mud

31 g3=mhd-sppa+spal+background Cyan1

32 g4=lmd-sppa+mix+background Magenta1

33 g9-ld-sppa+spal+mud

34 g10=ld-sppa+mix+background

35 g11=ld-sppa+w+mix

Figure 5-18. Classification training data of 35 landscape classification features. "Training" pro-

vides the pixel count after training selection; classification provides the image pixel count after a

classification algorithm is performed. This data set accompanies Figure 5-16, the classified image.

(Campbell, 2003).

(3) Classification Algorithms. Image pixels are extracted into the designated

classes by a computed discriminant analysis. The three types of discriminant analysis

algorithms are: minimum mean distance, maximum likelihood, and parallelepiped. All

use brightness plots to establish the relationship between individual pixels and the

training class (or training site).

(a) Minimum Mean Distance. Minimum distance to the mean is a simple com-

putation that classifies pixels based on their distance from the mean of the training class.

5-32

Western Governors University

Privacy Statement
Press Release
Contact

Integrated Publishing, Inc. - A (SDVOSB) Service Disabled Veteran Owned Small Business