203 
THE ROAD EXTRACTION IN THE AREA COVERED WITH HIGH VEGETATION 
USING THE FUSION IMAGE OF SAR AND TM 
Shen Jin-li, Yu Wu-yi, Qi Xiao-ping, Zhang Yi-min 
The research institute of petroleum exploration and development 
Shenjili Jc@ 163 .com 
KEYWORDS: Synthetic aperture radar, road detection, fusion method, interpretation 
ABSTRACT: 
Due to SAR images features of all day obtained and its penetration ability, we apply fusion method of radar image and 
multi-spectrum images, in order to improve the accuracy of road interpretation in the area of densely covered by high vegetation. 
Two methods for fusion SAR and TM multi-spectral images are presented in this paper. The experimental results show that the 
method of Principle Component Analysis (PCA) conversion fusion is more effective than the method of HIS fusion in the study area. 
1. INTRODUCTIONS 
Synthetic aperture radar (SAR) images not only have the 
characteristics of all-day, all-weather, but also provide object 
information which is different from visible and infrared sensors. 
SAR images are widely applied in many fields, such as 
agriculture, forestry, geology, hydrology etc. With high 
resolution and no spectral information, it requires professional 
experiences to interpret SAR images. Though optical images 
have abundant spectral information, it may disturb us to extract 
tiny objects. Remotely sensed image fusion has been proved to 
be a good method in feature extraction to avoid the faults of 
SAR images more speckles and fewer bands. 
Adopting a compound framework, image fusion enhancement 
integrates data of various sensors to get high quality 
information. Data fusion can take advantage of the use of 
complementary information to obtain a better overall accuracy 
than using single data source only. On the basis of previous 
work, we can reduce the uncertainty; improve the accuracy and 
reliability of the experimental results. There are several 
methods of image fusion such as HIS conversion fusion and 
principle conversion fusion, which are based on pixel-level 
with aim at improving visual interpretation results or 
classification results. There are two main pixel-level methods 
to be discussed in this paper, based on Principle Component 
Analysis (PCA) and HIS conversion. At last, the maximum 
likelihood classifier was used for both fusion images. 
It is difficult to extract road from moderate resolution images in 
high vegetation region. Synthetic Aperture Radar (SAR) is a 
kind of high resolution imaging radar. It can generate images 
independent of time and weather condition. It also has the 
ability of penetrating through some depth of the earth or 
vegetation. We experiment to identify more exact global 
information by radar images-ENVISAT and LANDSAT TM 
fusion. 
2. DATA AND METHODS 
The data sets used in this research includes: 1) LANDSAT TM 
acquired on October 11, 2006; with 30m resolution(Figl). 
ENVISAT SAR image acquired on October 16, 2006; C wave 
band, HH polarization. We applied relief map of 1:50000 and 
GPS ground control points surveyed in the field for image 
correction and registration. The error is controlled within one 
pixel. We denoise the SAR image by Lee filter. This method 
not only can reduce speckle noise, but also can remain the edge 
features. The size of image is 400X401 pixels. The study area 
is in plain with high vegetation. In order to select a more 
suitable method applied in the study area, we experiment two 
fusion ways: Principle Component Analysis (PCA) conversion 
and HIS conversion. 
Figl: TM multi-spectral image 
Principle Component Analysis-''(PCA) conversion is used to 
fuse the TM and SAR images. First, because of the largest 
quantity of information in the PC-1 component, we replaced 
the PC-5 component by SAR image, different from the usual 
processing. Then the inverse transform is followed (Fig2). 
Another fusion method is HIS fusion. We transform TM image 
from RGB to HIS space, then replace the I-component by SAR 
image, finally inverse to RGB space with H- and S-component 
(Fig3). At last, the maximum likelihood classifier was used for 
both two fusion images and the TM multi-spectral image. 
Fig2: image of PCA fusion method