1141
STUDY OF REMOTE SENSING IMAGE FUSION AND ITS APPLICATION IN IMAGE
CLASSIFICATION
Wu Wenbo,Yao Jing*,Kang Tingjun
School Of Geomatics,Liaoning Technical University, 123000, Zhonghua street,Fuxin,China - yaojingl 124@163.com
Commission VII, WG VII /6
KEY WORDS: Landsat image, data fusion, spectral Distortion, Classification accuracy, Algorithms evaluation
ABSTRACT:
Data fusion is a formal framework in which is expressed means and tools for the alliance of data originating from different sources.
It aims at obtaining information of greater quality; the exact definition of ‘greater quality’ will depend upon the application.
Satellites remote sensing image fusion has been a hot research topic of remote sensing image processing. Make Multispectral images
matching with TM panchromatic image, and the error control in 0.3 pixels within. Use Smoothing Filter-based Intensity Modulation
(SFIM), High Pass Filter (HPF), Modified Brovery, Multiplication, IHS, Principle component analysis Transform (PCA) methods
for the fusion experiment. Use some parameters to evaluate the quality of fused images. Select representative features from the fused
and original images and analysis the impact of fusion method. The result reveals that all the six methods have spectral distortion,
HPF and SFIM are the best two in retaining spectral information of original images, but the PCA is the worst. In the process of
remote sensing image data fusion, different method has different impact on the fused images. Use supervised classification and
unsupervised classification method to make image classification experiments, the study reveals that all the fused images have higher
spatial frequency information than the original images, and SFIM transform is the best method in retaining spectral information of
original image.
1. INTRODUCTION
The specific objectives of image fusion are to improve the
spatial resolution, improve the geometric precision, enhanced
the capabilities of features display, improve classification
accuracy, enhance the capability of the change detection and
replace or repair the defect of image data [1] .
But for a long time, remote sensing image fusion is mainly used
to enhance the visual interpretation, and it not usually used in
the research of improving the image classification, the main
reasons are shown as follows [2] : (D Image fusion is mostly
based on the fusion of different satellite. Because of the
difference of the various parameters and phase between
different sensors, as well as the inevitably registration error, led
to the fusion classification results unsatisfactory;© Although
the same sensor system provided different spatial resolution
images, because of its low spatial resolution, resulting in poor
classification effect; ©Because of the unreasonable fusion
algorithm or classification method make the failure of
classification.
In this paper, using Landsat ETM + images panchromatic bands
and multi-spectral bands to fuse, to research the fusion
technology of different spatial resolution based on the same
sensor system and the classification technology, evaluate the
infection of each fusion method with the land use classification.
2. THE CHOICE OF DATA SOURCES AND
SUMMARIZE OF THE PRETREATMENT
In this paper, using the image data of Landsat 7 ETM +
panchromatic and multispectral images of August 2001,the
study area is Shenyang.There are many types of feature in this
area ,the main features include rice, dry land, forest, water
bodies, residents of villages and towns and so on.
2.1 Bands Selection
Bands combination is a key step of fusion technique, bands
combination optimization must be followed by two principles:
firstly, the physical significance of the selected bands are good
and they are in different light area, that is to say the relevance
of each bands are small; secondly, we should choose the bands
with the largest information [3 lln this paper calculate the
correlation coefficient matrix and OIF index, select the bands
combinations in turn (table 1 and table 2).
In the table of OIF index we can see that, the OIF index of the
combination of bands ETM+3,4,5 is the biggest, and the
correlation coefficient of bands3,4,5 is the smallest, so choose
bands 3,4,5 as a fusion experiment.
2.2 Image Registration
The essence of image registration is according to the geometric
correction of the remote sensing images; adopt a geometric
transform to make the image unified to a same coordinate
*Corresponding author: Yao Jing, School Of Geomatics,Liaoning Technical University , 123000 , 359 mailbox,Zhonghua
street,Fuxin,China, yaojingl 124@163.com, telephone number:13941830365