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
other fallow : ; buildin
classes corn garden woodland meadow sand mountain saline desert water
corps land
a 538 321
Training samples 5 1615 495 679 1642 1483 7 4810 4690 3612 587 805
lidati 280 224
ha 1056 2295: 335 1180 — 1176 3202 2146 2484 327 479
samples 4 6
882 546
sum s 2671 724 1014 2822 2659 3 8012 6836 | 6096 914 1284
Table. 1 Training and Validation Samples
other fallow - : buildin
classes corn garden woodland meadow sand mountain saline desert water
corps land g
Classification with
: 86.2% 58.9% 62.1% 60.3% 83.1% 79.2% 85.4% 73.1% 77.7% 83.6% 872% 91.2%
Single TM
Classification by
95290 732% 782% 68.8% 91.2% 91.7% 932% 88.9% 92.7% 88.5% 96.8% 98.6%
presented method
Table. 2 Statistical Errors for the Two Algorithms
4. CONCLUSION
A new classification model for active and passive remote
sensing data is developed in this paper. In the model, a
classifier based on the Bayesian theory and MRF is set up,
ASAR in VV, VH polarization and 7 bands of TM are taken as
the input of the classifier. The validation by field measurements
shows that:
1) The classification model based on Bayesian and MRF in this
paper not only need not to assume the conditional probability
density function in advance, but also joining the spatial
correlation of the class labels, the model can get a better
classification accuracy of 89.4%.
2) Comparing with the Classification with single TM, the total
precision of classification by active and passive remote sensing
increase 11.5%, it shows the integration of TM and ASAR data
can increase the information of the surface objects, make them
easier to distinguish, and finally reach a better classification
precision.
The study area is a typical ‘oasis-desert’ dual ecological
environment in the paper, and terrain of the oasis is relatively
flat. These are conductive to identify and classify the objects in
ASAR image. But when the study area is selected a densely
populated plains or urban areas, the accuracy of classification
by active and passive remote sensing data needs to be further
verified.
REFERENCES
[1] Jia, Y., Li, D. R., 1995. Multisource classification of
remotely sensed data based on Bayesian data fusion
method, Journal of Wuhan Technical University of
surveying and Mapping, 22(3), pp. 248-251.
[2] Solberg, A. H., 1994. Multisource Classification of
Remotely Sensed Data: Fusion of Landsat TM and SAR
Images. IEEE Trans. Geosci. Remote Sens., 32(1), pp. 766-
778.
[3] Storvik, G., Roger, F., Solberg, A. H., 2005. A Bayesian
approach to classification of multi-resolution remote
sensing data. /EEE Trans. Geosci. Remote Sens., 43(3), pp.
539-547.
[4] Chellappa, R., Chatterjee, S., 1985. Classification of
Textures Using Gaussian Markov Random Fields. IEEE
Trans. Acous. Speech. Signal Process., 33(2), pp. 959—
963.
[5] Chellappa, R., Hu, T., 1983. On two-dimensional Markov
spectral estimation. IEEE Trans. Acous. Speech. Signal
Process, 31(4), pp. 836-841.
[6] Julian, H., 1986. On the statistical analysis of dirty pictures.
J. Roy. Statist. Soc, 48(1), pp. 259-302.
[7] Yonhong, J., Philip, H., 1996. Bayesian Contextual
Classification Based on Modified M-Estimates and
Markov Random Fields. IEEE Trans. Geosci. Remote
Sens, 34(1), pp. 67-75.
[8] Geman, S., 1984. Markov random field image models and
their applications to computer vision. /EEE Trans. Pattern
Anal. Mach intell., 26(2), pp. 721-743,
ACKNOWLEDGMENT
This work is supported by the National Science Foundation for
Young Scientists of China (Grant number: 41101321) and the
Key Projects in the National Science & Technology Pillar
Program (2009BAG18B01and 2012BAH28B03).
