232
9: Chintan A. Shah
Table 9.4. Error matrix for the classification of data, preprocessed by SPCA. a ICAMM
classification algorithm and b K-means classification algorithm
a
ICAMM (SPCA)
Reference map
Total
C 1
C2
C3
C4
Cl
350
60
52
463
Classification
C2
35
521
0
390
946
Map
C3
154
18
382
99
653
C4
55
41
0
1042
1138
Total
594
640
383
1583
3200
Producer's Accuracy (%)
58.9
81.4
99.7
65.8
User's Accuracy (%)
75.6
55.1
58.5
91.6
Overall Accuracy (%) = 71. 7
C 1 - Background, C 2 - Corn-notill, C 3 - Grass/Trees, C 4 - Soybeans-min
b
K -means (SPCA)
Reference map
Total
C 1
C2
C3
C4
Cl
42
0
0
288
330
Classification
C2
129
162
10
271
572
Map
C3
311
1
373
55
740
C4
112
477
0
969
1558
Total
594
640
383
1583
3200
Producer's Accuracy (%)
7.1
25.3
97.4
61.2
User's Accuracy (%)
12.7
28.3
50.4
62.2
Overall Accuracy (%) = 48.3
C 1 - Background, C 2 - Corn-notill, C 3 - Grass/Trees, C 4 - Soybeans-min
error matrix (Table 9.3b), where 485 pixels out of the total 640 pixels belonging
to class corn-notill have been classified as soybeans-min. This misclassification caused by the K -means algorithm is compensated by significantly higher
accuracies (both producer's and user's) for class soybeans-min. The lowest
producer'S and user's accuracies for the K-means algorithm are obtained for
the class background (7.2% and 1l.5% respectively). In case of the K -means
algorithm applied on the data preprocessed by the SPCA (Table 9.4b), similar
conclusions can be drawn about the accuracies of the class background. These
accuracies are as low as 7.1 % (producer's) and 12.7% (user's). However, these
accuracies obtained by the K -means algorithm for the class background are relatively higher for the asp and PP preprocessed data. For the SPCA preprocessed
data, ICAMM exhibits very high accuracy (99.7%) in classifying grass/trees.
Thus, based on the classification accuracies (Table 9.2), the classification
maps (Fig. 9.4) and the corresponding error matrices (Table 9.3 to Table 9.6),