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
Wi anctassifiec
M artificid
UU barstand
#8 cropland
Figure 5. The classification result based on C5.0
3.2 Accuracy assessment
One test samples were used to test the C5.0 and the
MLC classification results, and got an assessment
result: the overall accuracy of C5.0 classification
result is 78.8716%, and kappa coefficient is 0.7538.
While the overall accuracy of MLC classification
result is 74.6884%, Kappa Coefficient is 0.7080. Table
I and table II show the confusion matrix for the two
classification results separately.
TABLE I. CONFUSION MATRIX FOR C5.0
CLASSIFICATION RESULT
Class Os; artificial | bzreland | cropland Total
Unclassified | 045 1.82 ces $38
artificial. 853: $290 273 259
bareland EXT! 233 3.88 344
cropland 1.36 273 $279 1406
forest 227 ces 18 .$8
grass 3.18 1.82 242 10.43
shout cs 13) 88 382
water 273 9.00 9035 19.61
watland 200 1909 992
Total 100 ice oo
TABLE II. CONFUSION MATRIX MLC
CLASSIFICATION RESULT
baseisnt tes ET 142
apis 4.7 358 346i
fant Tag Si; iit
xat 4.78
m sn lis
wa S2 B p
wea ice 34
Tetas i 138 Hol ime i
To compare the accuracy of the two classification
result, we can conclude that the C5.0 method is more
excellent than MLC. We can see from table I and II,
artificial, cropland, forest and water can get a higher
precision, while the precisions of bareland, grass,
shrub are much lower. And compared the two tables
above, we can see, the misclassification phenomenon
in MLC result is very serious. For example, the
misclassification between artificial and forest, 7.4% of
the pixels of forest were classified to artificial.
4 CONCLUSIONS
Study shows that using C5.0 classification method
can get a higher precision than MLC classification
method. At the same time, in the process of
classification, it needn’t to select training samples on
every image when using C5.0 method, while we
should select training samples on every image to be
classified, and it will waste too much time. Compared
with MLC method, C5.0 classification method
increased image features information, which increased
discrimination between categories, so we can get
better result. Furthermore, in the analysis of data with
high dimensionality such as multi temporal
LANDSAT data, the computational speed of the
maximum likelihood classifier is reduced because the
classification time increases as the square of the
number of bands.
In all, we can conclude that decision tree based on
C5.0 classification method is suitable for large area
land cover classification for its automation,
high-speed, and high precision.
Acknowledgement
This work was funded by National High
Technology Research and Development Program of
China (2009AA122003) and the National Key
Technology R&D Program (2012BAH28B01). We
thank the anonymous reviewers for their helpful
comments. We also would like to acknowledge every
member of the GLC project team at CASM.
References
Liang ZHAI, Wenhan XIE, Huiyong SANG , Jinping
SUN. Land cover mapping with Landsat data: The
Tasmania case study. The 2011 International
Symposium on Image and Data Fusion, 9-11 August
2011, Tengchong, Yunnan, China.
S. M. JOY, A non-parametric, supervised
classification of vegetation types on the Kaibab
National Forest using decision trees. International
Journal of Remote Sensing, 2003, vol24, NO.9,
1835-1852.
Perera, K. and Tsuchiya, K., 2009. Experiment for
mapping land cover and it's change in southeastern
Sri Lanka utilizing 250m resolution MODIS imageries.
Advances in Space Research, 43 (9). pp. 1349-1355.
Heinl, M., Walde, J., Tappeiner, G., and Tappeiner U.,
2009. Classifiers vs. input variables—The drivers in
image classification for land cover mapping.
International Journal of Applied Earth Observation
and Geoinformation, 11(6). pp. 423-430.
Foody, G. M. 2002. Status of land cover
classification accuracy assessment. Remote Sensing of
Environment, 80(1). pp. 185-201.
Lu D. and Weng Q., 2007. A survey of image
classification methods and techniques for improving
classification performance. International Journal of
Remote Sensing, 28(5). pp. 823-870.
Herold, M., Mayaux, P., Woodcock, C.E., Baccini, A.,
and Schmullius, C., 2008. Some challenges in global
land cover mapping: An assessment of agreement and
accuracy in existing 1 km datasets. Remote Sensing of
Environment, 112(5). pp. 2538-2556.
Loveland, T. R., Reed, B. C., Brown, J. F., Ohlen, D.
Q. Zhu, Z., Yang, L. and Merchant, J. W.,
(2000). Development of a global land cover
