An object-oriented pattern recognition approach for urban classification - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 130

TABLE 9.3 Overall accuracy, producer's accuracy, user's accuracy, and Kappa coefficient produced by the rule set 1.

Referece Data Total Producer's User's

Non-Pool Pool Pixels Accuracy Accuracy

Non-Pool 172705 339 173044 99.78% 99.80%

Pool 384 1634 2018 82.82% 80.97%

Total Pixels 173089 1973 175062

Overall Accuracy = 99.59%

Kappa Coeficient = 0.82%

TABLE 9.4 Overall accuracy, producer's accuracy, user's accuracy, and Kappa coefficient produced by the rule set 2.

Referece Data Total Producer's User's

Non-Pool Pool Pixels Accuracy Accuracy

Non-Pool 173002 225 173227 99.95% 99.87%

Pool 87 1748 1835 88.60% 95.26%

Total Pixels 173089 1973 175062

Overall Accuracy = 99.82%
Kappa Coeficient = 0.92%

approaches to the entire dataset and evaluate the outputs visually
and quantitatively.

two lowest user's accuracies were associated with Impervious
(57.38%) and Trees (66.67%). This implies that there is some
signature confusion among grass, trees, and impervious surface.
The higher overall accuracy (87.50%) resultedwith the first set
of data (mean values of the original bands andPCAbands 1, 2, and
3) (Table 9.8). This is a very high accuracy for an urban LULC
map derived semi-automatically from fine spatial resolution
data. The two lowest producer's accuracies were associated with
shrubs (75.86%) and impervious (80.00%). The two lowest user's
accuracies were associated with pool (68.18%) and soil (81.25%).
This implies that greater signature confusion occurs among
shrubs, impervious, pool, and soil classes. Low producer's and
user's accuracies were not consistent between the two different

9.5.2 Nearest neighbor classifier to
extract urban land covers

The second application example and corresponding data set
(mean values of the original bands, brightness band, and
maximum difference band) yielded an overall accuracy of
73.50% (Table 9.7). The two lowest producer's accuracies
were associated with Grass (47.83%) and Trees (50.00%). The

TABLE 9.5 Overall accuracy, producer's accuracy, user's accuracy, and Kappa coefficient produced by the rule set 3.

Referece Data Total Producer's User's

Non-Pool Pool Pixels Accuracy Accuracy

Non-Pool 172641 91 172732 99.74% 99.95%

Pool 448 1882 2330 95.39% 80.77%

Total Pixels 173089 1973 175062

Overall Accuracy 99.69%

Kappa Coeficient = 0.87%

=

TABLE 9.6 Overall accuracy, producer's accuracy, user's accuracy, and Kappa coefficient produced by the rule set 4.

Referece Data Total Producer's User's

Non-Pool Pool Pixels Accuracy Accuracy

Non-Pool 172585 109 172694 99.71% 99.94%

Pool 504 1864 2368 94.48% 78.72%

Total Pixels 173089 1973 175062

Overall Accuracy = 99.65%
Kappa Coeficient = 0.86%