Multisource data fusion- fusing optical and sar data
for irrigated rice areas identification
3.1 Overlay
Overlay of multi temporal data and display in different color channel of RGB. This technique is
suitable to apply to single frequency or single polarization data like SAR. Not only this
technique renders some colors to the interpreter but it also present changes during the
acquisition period of the multidate data.
3.2 Principal Component Analysis
The purpose of applying principal component analysis (PCA) is to reduce the dimensionality of
input data into a smaller number of output channels. It is more suitable to multi spectral data
where each spectral band is always highly correlated. In PCA, the most information of input will
be transformed into the first component and the information content decreases with increasing of
the number of PCA component. As indicated in Figure 3, there is a scatter diagram plot of the
original gray level value in band A and band B. Superimposed on these original axis is the new
axis (axis I and axis II) which defined the direction of the first and the second principal
components. PCA is the relationship that transforms the original value from band A/band B
coordinate system to the new axis (axis I/axis II) system (Lillesand et al., 3
rd Edition) and it is
the linear relationship (see figure 3).
Figure 3: Graphical representation of Principal Component Analysis (PCA)
3.3 Thematic Combinations
The purpose of thematic combination is to combine the information rather than the base data. By
this mean, the source data are not necessary converted to one common format (e.g. resampling
to one resolution) before deriving information and therefore less information are lost. In this
study, information was derived from optical and SAR data sources separately and each
technique was evaluated using the correlation coefficient. NDVI image, PCA of JERS-OPS and
average of multitemporal SAR data were investigated for the purpose of irrigated rice area
identification.
4. Results
Results from applying the above fusion techniques are demonstrated and interpreted to show the
derived information.
Overlay three SAR images not only render different colors related to different information but
also provide changes information. In the RGB image (figure 4), gray, white and dark color
represented no changes area (during Sep’96 to Mar’97) meaning that the bright white color in
the image demonstrated a stable land use which has been no changes during such period. In this
image, those are settlements (indicated by red circle). Another primary colors, Red, Green and
Blue represented changes (changes of backscattered energy) area. Red color indicated the object
that gave higher backscatter in September than in November and March. Green and Blue color
are followed the same rule. Green indicated the area where the object provided higher
backscatter in November while lower in September and March. Blue represented the area
containing the object that gave high backscatter in March and low backscatter in September and
November. Usage of cropping calendar in this area will help in more detailed interpretation.
Principal components derived from JERS-OPS multispectral data (3 bands) were demonstrated
in figure 5. The first component (PC1) contained highest variance (indicated by more variation
of gray levels) and therefore contained most information compared to the succeeding
components PC2 and PC3. Thus PC1 were then used as one theme for thematic combinations
instead of using 3 bands of optical data.
Figure 4: Show original multitemporal Images (Sep’96, Nov’96 and Mar’97) and their overlay
result in RGB channel
Figure 5: Principal Components of JERS-OPS data from left to right PC1, PC2 and PC3
Figure 6 shows the information derived from JERS-OPS and JERS-SAR namely, NDVI image,
the first principal component and average of three date SAR images respectively. NDVI value
will determine how much green vegetation presented in a particular area. In the NDVI image
presented here the lowest area NDVI is indicated by blue color and the highest indicated by red
color. Same as PC1 image the lowest value is ranged from blue to the highest in red color.
Average SAR images is presented in gray scale image where dark meant low backscatter and
bright meant high backscatter.
All the information (figure 6) derived from both OPS and SAR were then be the input bands for
the classification process. Unsupervised method was applied 2 times and the result is shown in
figure 7. This classified image result was checked and it had the correspondence with the field
information.
Figure 6: From left to right: NDVI image, PC1 image and average of 3 SAR images
Figure 7: Unsupervised classification result
5. Conclusion
Combining of data from different sources using thematic combination seems to be the most
appropriate technique because less information were lost. However, the information obtained
from this study still requires more field information to verify the applicability of fusion
technique presented here. Further, fusion among multi sensor and multi systems could require
more parameters for instance satellite geometry, spectral band width etc. complexing the
interpretation.
References
-
Lillesand M.T., Kiefer W.R., Remote Sensing and Image Interpretation, 3rd Edition, P572, John
Wiley & Sons, Inc, New York.
-
Soldberg, A. H., 1994, Multisource classification of Remotely Sensed Data: Fusion of Landsat
TM and SAR Images, IEEE Transactions on Geoscience and Remote Sensing Vol.32 No.4 July
1994 pp768-776.
-
Wald L., 1999, Some Terms of Reference in Data Fusion, IEEE Transactions on Geoscience and
Remote Sensing Vol.37 No.3 May 1999 pp1190.
