International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
(j) Multiplicative transform of ERS and LISS III (k) Brovey transform of ERS and LISS III
Figure 2. The visual comparison of the merging methods from IRS 1 D Pan, LISS III and ERS images.
3.4.2 Classification
Maximum likelihood supervised classification and classification
accuracy assessment were applied to all fused images. In order
to select training sites for the classification 1/5000 scaled
standart topographic maps and aerial photographs were used.
Urban, grass land, barren land, water and highways were
selected as classes to represent the surface of study area. After
signature extraction, accuracy assessment of classification was
calculated using an error matrix, which showed the accuracy of
both the producer and the user (Lillesand and Kiefer 2000). The
classification accuracy in remote sensing shows the
correspondence between a class label allocated to pixel and true
class. For accuracy assessment, 250 pixels were randomly
selected for each image. Land use maps and aerial photographs
were used as reference data to observe true classes. The overall
accuracy and a Kappa analysis were used to perform a
classification accuracy assessment based on error matrix
analysis. Table 1 shows overall accuracy and kappa values for
the classification of fused images.
Fusion technique ‘Overall % Kappa %
Brovey (IRS Pan+LISS) 80 74,5
Multiplicative (IRS Pan+LISS) 77,6 70,1
HIS (IRS Pan+LISS) 83,2 78,9
PCA (IRS Pan+LISS) : 82,8 77,6
Multiplicative (ERS +LISS) 70,4 58,3
Brovey (ERS +LISS) 12 61,5
HIS (ERS +LISS) 75,6 66,6
PCA (ERS +LISS) 72 63,1
Table 1. Classification accuracy results of fused images.
4. CONCLUSIONS
By using fusion techniques visual interpretability of images
were improved. Furthermore, detailed land cover/use
information was obtained by classification of fused data. As a
result of containing very complex and irrregular features, land
cover/use categories are highly spatially intermixed in the study
site. Due to similar reflectance values of urban, road and bare
land class, mixed pixel problems were occurred during the
classification process. The reason of similar reflectance
578
between these categories is the result of irregular and roofless
buildings. According to classification results, the best result for
the study site was obtained by using IHS colour transformation
technique with overall accuracy of 83.2 % and kappa value of
77.6 %. In order to investigate appropriateness of fusion
techniques a pilot area that has regular urban structure was
selected and maximum likelihood classifications were applied
to this data. The results showed that classification accuracy
assessment values are better for the pilot area. The best result
was obtained for the pilot site by means of IHS colour
transformation with overall accuracy of 88.4 % and kappa value
of 82.3 %. The classification accuracies of radar and LISS
merged images were around 70 % because of the complex
structure of the area. Merged radar and multi spectral data
improved visual interpretability of radar image, these images
can be used as ancillary data for classification procedure. For
the aim of this study, the results of merged optic images were
satisfactory for producing land cover map of the area.
5. REFERENCES
Chavez, P.S., Sides, S.C., Anderson, J.A., 1991. Comparison of
three different methods to merge multi resolution and multi
spectral data: Landsat TM and SPOT Panchromatic.
PhotogrammetricEngineering and Remote Sensing, 577 (3), 295—
303.
Chena, C. M., Hepnerb, G. F., and Forsterb R. R., 2003. Fusion
of hyperspectral and radar data using the IHS transformation to
enhance urban surface features. ISPRS Journal of
Photogrammetry & Remote Sensing, 58 , pp. 19— 30.
Gamba, P., Houshmand, B., 1999. Three-dimensional road
network by fusion of polarimetric and interferometric SAR
data. Proceedings of IGARSS’99 I, 302- 304.
Janssen L. F. and Vander W. J., 1994. Accuracy A Review.
Photogrametric Engineering and Remote Sensing, pp.419-425.
Jensen, J.R., 1996. Merging different types of remotely sensed
data for effective visual display. Introductory Digital Image
Processing: A Remote Sensing Perspective, 2nd ed. Prentice-
Hall, Upper Saddle River, NJ, pp. 100— 103.
Interi
Lille:
Imag
Niem
1998.
with
Interi
174—
Pohl,
in T
Interi
Sara
apprc
Inter
Wald
Archi
Welc
with
Jourr
Weyc
ERS-
Proce
Sensi
Yao,
aperti
sensi
Remo
Zhan
spatia
(10),
Zhou.
transí
Panct
(4), 7