×

You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Title
Mapping without the sun
Author
Zhang, Jixian

4. EXPERIMENT AND RESULT ANALYSIS
Figure 10. Texture feature image of contrast
Figure 11. Texture feature image of correlation
4.1 Spectrum-feature-based classification
The spectrum feature is the basic feature of the remote sensing
image. The traditional image classification algorithms are
based on the spectrum feature. The spectrum-feature-based
maximum likelihood classification algorithm in the supervised
classification is used in the paper. Classification result and
precision statistics are as follows:
Figure 14. The result of using the maximum likelihood
classification method base on spectrum feature
Class
Prod.Acc.
(Percent)
User.Acc.
(Percent)
inhabited area
86.31
84.10
Paddy field
78.31
85.20
Terraced field
83.45
81.31
Forest
84.41
82.21
Bare land
87.81
85.81
lakes
88.33
80.27
Table 1. Precision statistics of the maximum likelihood
classification
precision=84.0789%, Kappa= 0.7828
4.2 Texture feature assistance classification
Figure 12. Texture feature image of entropy
Figure 13. Texture feature image of second moment
In figure 6~9 images are based on the 5x5 sliding window. In
figure 10-13 images are based on the 7x7 sliding window.Seen
from the charts, brightness of contrast image’s is bigger as well
as the inhabited area appears quite obviously in this figure. But
other terrain features are not very prominent depending on the
visual observation in contrast image. The entropy image
gradually becomes bigger with the window and texture features
turns more abundant.
For supervisor classification algorithm of texture features
assistance classification, we have still used the maximum
likelihood classification of supervised classification to compare
conveniently.
The image with 4 bands is formed by 3 bands spectrum feature
image by adding a texture feature image. The texture feature
images are obtained by 5x5 sliding window computing.
We obtain four supervisor classification images using texture
feature assistance spectrum feature classification. These are
Figure 15. The result of classification using correlation
texture feature image assistant spectrum feature
103