Data fusions from SAR and TM, SPOT and TM, ASTER and TM, MODIS and ETM, etc are the common methods. But
that from TM and CBERS-02B is rare. With HR camera working in September 19th 2007, Chinese-Brazil Earth
Resources Satellite 02B (CBERS-02B) became the first civilian high-resolution satellite in China. It could provide 2.36m
panchromatic image which is better to Landsat TM. Meanwhile the spectral resolution of TM is better than CBERS-02B.
So it's a good idea to take advantage of benefits from CBERS-02B HR and TM through data fusion.
In this study, images of TM and CBERS-02B HR in 2007 were used as data sources. After image registration and noiseremoval
process, data fusion methods of IHS and PCA were adopted. Then unsupervised classification and supervised
classification were used for land use classification. Finally, classification accuracy between original image and fusion
image was compared and evaluated.
The result shows:
(1) Compared with original TM or CBERS-02B HR image, the fusion image not only retains abundance spectrum but
also enhances the object details. Residential texture, lake morphological, the relative position between roads, industrial
and mining sites, etc, was identified easily.
(2) Results from IHS and PCA are different. IHS image had higher spatial resolution but more spectral distortion.
Spectral differences between some objects became smaller and classification accuracy was lower. Supervised
classification accuracy assessment shows that overall Kappa index and overall land use classification accuracy decreased
by 0.237 and 11% respectively. Meanwhile PCA image not only had high spatial resolution, but also smaller spectral
distortion. Different land use / cover types can be better distinguished.
(3) Disadvantages of low spatial resolution in TM and single color in CBERS-02B HR image are overcome in PCA
fusion image to a certain extent. In this research under supervised classification in PCA image Kappa index of farm land,
forest land and bare land increased by 0.097, 0.176 and 0.242 respectively. Overall Kappa index and overall land use
classification accuracy were improved by 0.092 and 7.24% respectively.
|