963
Beijing 2008
rea
i
d in Section 2,
vONOS were
¡cause of the
teps,including
-classification
i3.5.Firstly,we
el-level SVM
i we adopted
>ixel to obtain
ification
s represent
fectiveness of
: (1) attribute
tnge type,(2)
;e accuracy is
lie geometric
;a and shape,
dices derived
Dmpare these
iracy is the
s (objects or
The detection
¡ted changed
a of omitted
he proportion
dysis (Cohen,
appa statistic,
based on the
ice agreement
arch we took
methods and
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
compared the performance of object-level post classification
comparison(OL-PCC) method and pixel-level post classification
comparison (PL-PCC)approach.
Change detection approach
OL-PCC
PL-PCC
Detection accuracy (%)
67.3
52.6
Overall accuracy (%)
85.4
68.9
Omission error (%)
32.7
48.4
Commission error (%)
15.7
25.2
Overall kappa
0.74
0.54
Table 3. Change detection accuracy of OL-PCC and PL-PCC
5. RESULT DISCUSSION
As outlined above, the proposed OL-PCC method combinng
MS&RG multiscale image segmentation, SVM and OOA was
proved to have advantages against PL-PCC
methodology.Besides the low detection precision,it is worthing
noting that the phenomenon of “ Salt and pepper ” is severe in
the change map based on PL-PCC.
Through the OLCD, the initial multiscale image segmentation
process insures the quality of the multispectral data to be
submitted to SVM classification. Indeed, the object delineation
combined the spectral, spatial and contextual information to
create consistent units of interest. The segmentation is also less
sensitive to misregistration errors than traditional pixel-level
analysis methods(Makela & Pekkarinen, 2001)between
multidate images and reduces the change detection processing
time given that there are much fewer objects than pixels. Based
on these objects,OLCD method breakes the constraint of sensor
characteristics and spatial resolution in multisource satellite
images and the change detection performances are
increased.Moreover, the object boundaries derived directly from
the segmented images are more convenient to update GIS
database in land surface monitoring and map updating.
6. SUMMARY AND RECOMMENDATIONS
The object-level change detection method proposed here proved
to be very efficient to identify land use and land cover changes
of HR satellite images.A detection accuracy higher than 85%
and an overall kappa higher than 0.7 were achieved using a
SPOT5 and IKONOS multitemporal data set covering a 4-years
time span. This technique can be considered scene-independent
in the sense that OOA determines whether change or not
according to class attributes of each object ,instead of
predefined change threshold of the multidate image.
Whereas this research focused mainly on the whole technique
workflow and final detection accuaracy,igoring the influence of
segmentation precison.If the aim is to obtain accurate and
quantitative assessments about the change area,this
approach needs further theoretical developments.
References
Atkinson,P.M.,&Lewis,P,2000.Geostatistical classification for
remote sensing:An introduction.Computers and
Geoscience,26,pp.361 -371.
Burroughs, P.P. & McDonnel, R.A. 1998, Principles of GIS,
Oxford University Press, pp. 162 -166.
Baudouin Desclee, Patrick Bogaert, Pierre Defoumy,
2006.Forest change detection by statistical object-based
method.Remote Sensing of Environment 102, pp. 1-11.
Cohen, J. 1960. A coefficient of agreement for nominal scales.
Educational and Psychological Measurement, 20, 37-46.
D. Lu et al.,2004,Change detection techniques, INT J REMOTE
SENSINQ25 (12): 2365-2407.
Flanders, D., Hall-Beyer, M., & Pereverzoff, J.2003.
Preliminary evaluation of eCognition object-based software for
cut block delineation and feature extraction. Canadian Journal
of Remote Sensing, 29(4), 441-452.
Hunter, G.J. , 1998.Boolean Operations, 451-620 Lecture Notes,
The University of Melbourne.
Lobo, A. 1997. Image segmentation and discriminant analysis
for the identification of land cover units in ecology. IEEE
Transactions on Geoscience and Remote Sensing, 35,
1136-1145.
Massonnet D,Rossi M,Carmona C,et al, 1993. The displacement
field of the Landers earthquake mapped by radar
interfereometry[J], Nature,(364), pp. 138-142.
Makela, H., & Pekkarinen, A. 2001. Estimation of timber
volume at the sample plot level by means of image
segmentation and Landsat TM imagery.Remote Sensing of
Environment, 77, 66-75.
Mu H. Wang , Hai T. Li, Ji.X Zhang,Jing H.Yang,2007.Fusion
SAR and optical images to detect object-specific
changes.’Mapping without the Sun’-ISPRS.
Rosenfield, G. H., & Fitzpatrick-Lins, A. 1986. A coefficient of
agreement as a measure of thematic classification accuracy.
Photogrammetric Engineering and Remote Sensing, 52,
223-227.
Shackelford A.K.,Davis,C.H.A combined fuzzy pixel-based and
object-based approach for classification of high-resolution
multispectral data over urban area.IEEE Transactions on
Geoscience and Remote Sensing,2003,41(10):2354-2363.
Townshend,J.R.G.,Huang,C.,Kalluri,S.N.et al,2000.Beware of
per-pixel characterization of land cover.Intemational Journal of
Remote Sensing,21,839-843.
Wang Jianmei,Li Deren,Qin WenZhong,2005.A combined
segmentation and pixel-based classification approach for
QuickBird imagery for land cover mapping.MIPPRTmage
Analysis Techniques. 6044:U1-U9.
Zhan, Q., Wang, J., Peng, X., Gong, P., & Shi, P. 2002. Urban
built-up land change detection with road density and spectral
information from multitemporal Landsat TM data. International
Journal of Remote Sensing, 23,3057-3078.
ACKNOWLEDGEMENTS
This work was supported by National Key Basic Research and
Development Program(Grant No. 2006CB701303).