ough
. For
fuses
iy Lj
'k on
06b),
27)
(2)
2)
(3)
sf)
(4)
z
form
| ker-
ue
es À
TOSS-
e the
rma-
ween
dint).
et al.,
ition
ange
land-
rnel-
time
only
time.
. For
xj)
le in-
ixels
d for
spe-
ction
> dif-
(5)
| ker-
mage
n the
iven.
it has
ieral.
e-art
nsive
llan-
esent
Xu,
)pha-
"Teen
ews
focused on applications of change detection of satellite images
in forestry. A comparable contribution for landscape monitor-
ing is given by (Kennedy et al., 2009). (Almutairi and Warner,
2010) present important considerations on accuracy assessment
and the influence of accuracy to the final change detection result.
(Van Oort, 2007) presents an insightful contribution on the impor-
tance of the error matrix of multitemporal classification. Some
state-of-the-art papers are given e.g. by (Bruzzone and Serpico,
1997) and (Bruzzone et al., 2004) which present an iterative ap-
proach for change detection. (Coops et al., 2010) apply Land-
sat time series for assessing forest fragmentation. (Dianat and
Kasaei, 2010) use a polynomial regression technique for change
detection which considers neighborhoods. Application schemes
based on SVMs are presented by (Nemmour and Chibani, 2006),
(He and Laptev, 2009) and (Bovolo et al., 2008). (Mota et al.,
2007) and (Feitosa et al., 2009) present fuzzy approaches based
on modeling the class transitional probabilities.
3.2 Kernel-composition
Kernel-composition as a method of data fusion has originally
been proposed by (Camps- Valls et al., 2006b). An example for
multispectral data fusion is given by (Camps-Valls and Bruzzone,
2009). Kernel-composition is widely applied for spectral-spatial
classification fusing hyperspectral data with wavelets (Tan and
Du, 2011), (Velasco-Forero and Manian, 2009), spatial correla-
tion data (Chen et al., 2011), morphological profiles (Tan and Du,
2010), (Tuia and Camps-Valls, 2009), empirical mode decompo-
sition (Demir and Erturk, 2010), median filters (Marconcini et
al., 2009) or self-complementary filters (Fauvel et al., 2008). An
approach fusing hyperspectral and laserscanning data is given
by (Braun et al, 2011). Another application domain is semi-
supervised learning, where kernel-composition is used to incor-
porate the spectral information of unlabeled pixels (Tuia and Camps-
Valls, 2011), (Tuia and Camps-Valls, 2009), (Marconcini et al.,
2009), (Camps- Valls et al., 2007), (Bandos et al., 2006). (Camps-
Valls et al., 2008), (Camps- Valls et al., 20062) provide the frame-
work for multitemporal classification used herein. The frame-
work has so far been evaluated e.g. by (Bovolo et al., 2010) in
a support vector domain description approach on Landsat images
and by (Volpi et al., 2011) in an unsupervised approach on mul-
titemporal VHR data. To the knowledge of the authors, an eval-
uation on a multitemporal scenario with medium resolution data
based on supervised classification has not been published so far.
4 RESULTS
Within this section, results for three change detection approaches
will be presented. The main objective is to monitor the growth
of a limestone pit close to the village Mauer (near Heidelberg) in
the Upper Rhine Graben, Germany, Latitude: 49^ 19'50"N, Lon-
gitude: 8?48' I" E (cf. Fig. 1). Two 122 x 135 pixel (~14.8km?)
subsets of Landsat ETM+ images are used. The first is from 02-
11-2001 (Fig.2(a)), the second is from 17-4-2005 (Fig.2(b)). For-
tunately, the village is located in the very center of both images,
so the failure of the Landsat ETM+ scan line corrector does not
affect the work at all. Between the two points in time, the lime-
stone pit has considerably grown. Although more landuse classes
are assigned for classification in the first place (e.g. meadows,
forests, settlements), only two classes of interest will be consid-
ered in the final result: LS-Pit present in 2001 (yellow) and LS-
Pit new between 2001 and 2005 (red). Complete groundtruth has
been made available by a digitization in the field and is shown
in Fig.2(c). Except the limestone pit, all other landuse classes
will not be considered and set to black. At first, change detec-
tion based on a post-classification approach will be employed.
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
Secondly, features will be stacked to represent the change of pix-
els intensities between the two points in time. Lastly, kernel-
composition approaches based on (Camps- Valls et al., 2008), (Camps-
Valls et al., 2006a) and (Camps- Valls et al., 2006b) will be em-
ployed. These approaches also incorporate changes in pixels in-
tensities. All change detection approaches are based on image
classification. Each classification was done using an SVM with a
Gaussian RBF kernel. Kernel parameters were tuned using a 5-
fold grid search in the ranges y € [2 1?, 2?] and C € [27,277].
The LibSVM 3.11 library was utilized (Chang et al., 2001).
4.1 Post-classification approach
Two classifications are performed using the same landuse classes
in each dataset. The limestone pit is represented by a single class.
Since classification of the two points in time is performed sepa-
rately, it is not possible to assign the class LS-Pit new between
2001 and 2005 in the 2005 dataset. No features are available
which indicate whether or not a pixel has belonged to the lime-
stone pit in 2001 when classifying the 2005 image. From there,
one class LS-Pit has been assigned in both datasets and the two
classes of interest have been determined by overlaying the results
afterwards. The overall accuracy on the two classes is 86.7%.
A visual result after setting other classes (like meadows, forest,
settlements) to black is given in Fig.2(d). Note the high amount
of pixels falsely assigned to the class LS-Pit new between 2001
and 2005. Since the entire image scene is made up of limestone,
many places of bare soils have been confused with the limestone
pit.
4.2 Stacked-features approach
In order to provide implicit information on the changes of pixels
intensities, a stacked-features approach was performed. The data
matrices of both 8 channel Landsat datasets were concatenated to
build a 16 channel data matrix. Within this feature space, the two
classes LS-Pit present in 2001 and LS-Pit new between 2001 and
2005 can be distinguished in a single classification step. While
the class LS-Pit present in 2001 shows grey color in both images,
LS-Pit new between 2001 and 2005 would change from e.g. green
to grey. This indicates that a change from e.g. meadows to lime-
stone pit has taken place between the two points in time. From
there, only one classification needs to be performed. The overall
accuracy on the two classes is 87.5%. A visual result after setting
other classes to black is given in Fig.2(e). Note that the amount
of false positive pixels is considerably reduced. False positives
are now assigned to the class LS-Pit present in 2001. Much less
pixels are found in LS-Pit new between 2001 and 2005. Since the
latter class is characterized by a change in color from green to
grey, it is less confused with bare soils which would stay grey in
both points in time.
4.3 Kernel-composition approach
The last approach performed is similar to the stacked-features ap-
proach. In order to incorporate information on e.g. color changes,
both datasets are combined to a new dataset. However, it is aimed
to perform the fusion not in the feature space, but in the RKHS.
Therefore, a composed kernel matrix was build, e.g. by Kchange (xf, x5) =
Koone a0) + Kaoos (ar, mi The following kernel-
composition approaches were followed: direct summation (Eq.2),
weighted summation (Eq.3), cross-information (Eq.4) and image
differencing (Eq.5). The overall accuracy values for each ap-
proach can be seen in Tab.1. The best overall accuracy yielded
LH Y — ezp( vo; + z5||)?). implemented in LibSVM
3.11
