(d) Result: Post-classification
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
(e) Result: Stacked-features
(f) Result: Kernel-composition
Figure 2: Landsat ETM- image scenes, groundtruth and results
Table 1: Approaches and overall accuracies
[
| Approach | Composition | Overall accuracy |
Post-classification - 86.7%
Stacked-features - 87.596
Kernel-Composition | Direct. Sum. 88.876
Kernel-Composition | Weighted. Sum. 88.1%
Kernel-Composition Cross-Info. 87.9%
Kernel-Composition Image Diff. 88.6%
by the direct summation approach on the two classes is 88.8%.
As can be seen, all kernel-composition approaches yield slightly
higher accuracy values than the other approaches. Although specif-
ically designed for this task, the image difference kernel does not
yield the highest accuracy value. However, the performance dif-
ference to the direct summation kernel is only 0.2 percent points
— a value which should not be over-interpreted. It should be noted
that more simple kernels yield better results than more complex
ones. A result which is in agreement with the findings of the
inventors of the framework (Camps-Valls et al., 2006b), (Camps-
Valls et al., 2008), (Tuia and Camps- Valls, 2009). A visual result
after setting other classes to black is given in Fig.2(f). There are
much less false positives than in the other approaches. The only
exception is a large barren field where open Loess soil mixed
with limestone rocks is found (south-east corner of the image).
The spectral characteristics of this field are very similar to the
limestone pit thus making the classifier susceptible for confusion
with the limestone pit.
4.4 McNemar’s Test
The advantages in overall accuracy of e.g. kernel-composition
over post-classification may seem only a slight gain. Therefore,
they were tested for significance using McNemar's test (Foody,
2004). McNemar's test is based on x? statistics and can be em-
ployed to test the significance of differences between two nom-
inal labellings. The advantage is considered as significant if the
resulting test value |z| 7 1.96. Testing the advantage of kernel-
composition over post-classification yielded |z| = 13.95 indicat-
ing a significant advantage. A test of kernel-composition against
the stacked-features approach yielded |z| z 6.50 which also is
significant. The stacked-features approach yielded |z| z 10.16
over post-classification. Thus, all advantages described are sig-
nificant.
5 DISCUSSION
We present a comparison of change detection based on kernel-
composition with two traditional methods, post-classification change
detection and a stacked-features approach. As Fig.2(d) reveals,
post-classification change detection produces many false alerts.
In fact, the limestone pit is only located in the very south east.
The overall accuracy yielded is 86.7%. Stacking features be-
fore classification (i.e. combining the two 8 channel datasets
to a new 16 channel dataset) and assigning two limestone pit
classes (LS-Pit present in 2001 and LS-Pit new between 2001
and 2005) produces better results. However, there are still too
many false alerts Fig.2(e). The overall accuracy yielded is 87.5%.
Approaches based on kernel-composition produce the most ac-
curate results. However, there are slight differences when us-
ing kernel-composition that depend on the type of composition.
In agreement with other authors, more simple composition ap-
proaches tend to produce better results than more complex ones.
In our case, the direct summation and the image difference kernel
produced the best results. The highest overall accuracy yielded
is 88.8% by a direct summation kernel. There are almost no
false alerts and the change between 2001 and 2005 becomes most
