for 
its 
$ in 
the 
Ires 
iral 
00. 
ion 
acy 
Ces 
tial 
ata. 
the 
ous 
ote 
lent 
pp. 
to 
tial 
A 
lass 
for 
ints 
    
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 
CLASSIFICATION OF ACTIVE MICROWAVE AND PASSIVE OPTICAL DATA BASED 
ON BAYESIAN THEORY AND MRF 
F. Yu’, H. T. Li, Y. S. Han, H. Y. Gu 
Chinese Academy of Surveying and Mapping, Beijing 100830, P. R. China — (yufan, Ihtao, hys han, guhy)(g)casm.ac.cn 
Commission VII, WG VII/4 
KEY WORDS: Active and passive remote sensing, Classification, Bayesian theory, MRF, ASAR, TM. 
ABSTRACT 
A classifier based on Bayesian theory and Markov random field (MRF) is presented to classify the active microwave and passive 
optical remote sensing data, which have demonstrated their respective advantages in inversion of surface soil moisture content. In 
the method, the VV, VH polarization of ASAR and all the 7 TM bands are taken as the input of the classifier to get the class labels 
of each pixel of the images. And the model is validated for the necessities of integration of TM and ASAR, it shows that, the total 
precision of classification in this paper is 89.4%. Comparing with the classification with single TM, the accuracy increase 11.5%, 
illustrating that synthesis of active and passive optical remote sensing data is efficient and potential in classification. 
1. INTRODUCTION 
A range of remotely sensed data from sensors differing in terms 
of their spectral, spatial, and temporal resolution is now widely 
available. Given the large size of such multisource data sets, the 
immediate problem is how to choose and apply a suitable 
classification algorithm in order to achieve a level of accuracy 
that is acceptable for the given application. 
Optic and microwave remote sensing are two common 
methods to obtain the land surface information. The optic 
remote sensing data, with rich spectral information, represents 
the surface reflective spectral or the emission spectral; 
Microwave remote sensing has the characteristics of strong 
penetration, and it is the general information of vegetation 
coverage, surface roughness, dielectric constant, structure and 
so on. When the optic images are helpless with the problems of 
‘foreign objects with the same spectrum’ and ‘identical objects 
with the different spectrum’ in the earth observation, the 
microwave images can distinguish the objects by its surface 
roughness, structure, shape, water content and so on. Therefore, 
the integration of optic and microwave remote sensing data can 
gain the features of objects in different aspects, and do well in 
the classification or feature extraction. 
Currently, integration of optical and microwave remote 
sensing in classification is attracting increasing attention, 
Reference (jia et al, 1995) used modified Bayesian Network to 
classify the Landsat TM and Aircraft SAR images, and found 
the precision of the classification by fusion TM and ASR are 
20% higher than the single TM. The decision level fusion of 
TM and SAR images was applied to classification (Solberg, 
1994), and further improved by adding the Markov random 
field; Storvik (Storvik, 2005) proposed Bayesian network to 
classify the multisource remote sensing with different spatial 
  
* . . . . . * 
Corresponding author, Ph. D, majors in classification with active microwave and passive optical remote sensing data. 
resolution and get an accuracy of 88.7%. However, the above 
reference can not handle the SAR image speckle noise, and 
discuss less of the extraction of the multi-feature of SAR, so 
they didn't set up the appropriate conditional probability 
density model. 
Consequently, we have developed a new classification model 
for multisource data based on the Markov Random Field (MRF) 
and Bayesian theory. In the model, a Bayesian classifier based 
on MRF is developed, the VV, VH polarization of ASAR and 
all the TM bands are taken as the input of the Bayesian 
classifier to get the class label of each pixels of the mutilsource 
images. At last, the model is validated by the field 
measurements. 
2. CLASSIFIER BASED ON BAYESIAN THEORY AND 
MRF 
Bayesian statistical theory has been widely used as a 
theoretically robust foundation for the classification of remotely 
sensed data. The matter of multi-source remote sensing imagery 
is : suppose multivariate image X is composed of N- 
dimensional pixels where X,,, denotes the eigenvector of X, k — 
1,2, ..., N, presents the N dimensions, and s = (i, j) denotes the 
coordinate on image X. w denotes the field which contains the 
classification of each pixel in X; points in w can take values in 
the set {1, 2, . . . , L}, where L is the number of classes. The 
multivariate image X is then classified by finding a field of 
class labels Wap such that: