individual land use classes using the three child images and 
parent spectral image IRS 1C LISS III were. presented in 
Table 2. 
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002 
  
Thematic Accuracy of Classification Results (%) 
  
  
  
  
  
  
  
  
  
  
  
  
LISS I | poco | substituted | fused 
1 | Sugarcane 80.00 | 95.00 86.67 83.33 
2 | Banana 66.67 | 78.79 75.76 72.13 
3 | Plantation crops 64.29 | 71.43 67.86 71.43 
4 | Other Crops 72.31 80.00 83.85 50.00 
5 | Fallow Land 1 91.46 | 92.68 87.80 86.59 
6 | Stony Waste 61.54 73.08 57.69 53.85 
7 pue a a | 7-48 | 7143 | 5714 | 4444 
Overall Accuracy 78.40 84.24 80.78 68.43 
Kappa statistic 0.75 0.81 0.76 0.56 
  
  
  
  
  
  
Table 2 .The Thematic accuracy of natural land cover classification 
results 
Builtup 
Banana 
Coconut 
Stonv 
i ans 
  
Road 
LISIII data of study area Land use map prepared from LISSIII 
Fig.1. LISS III image of the study area and corresponding land 
use map 
A comparison of the accuracy estimates for LISS III and PAN 
merged LISS III (IHS) derived land use map reveals overall 
accuracy figures of 78.40 per cent and 84.24 per cent, 
respectively. The overall accuracy was 5.80 per cent greater 
than the parent spectral image. À significant enhanced accuracy 
of IHS fused image over the LISS III image was obtained for 
the land use classes, sugarcane, banana, fallowland and stony 
wasteland. 
  
p 
    
  
Other 
Fallow 
Coconut 
Stonv 
Tank 
Road 
   
PAN + LISIII (IHS) data Land use map 
sugarcane 
Fig.l. PAN + LISIII (IHS) image of the study area and 
corresponding land use map 
Merging LISS III digital data with the PAN data has resulted 
in a increase in overall classification accuracy was observed 
when the intensity, hue and saturation (IHS) transformation 
was applied to the LISS III data followed by replacing the 
Intensity (I) component by the PAN data, restoring 
subsequently to RGB components, and linearly stretching the 
resultant digital data. Because only one of the three 
components of the IHS transforms of the LISS III data were 
replaced by the PAN data while keeping the other two 
components — hue and saturation — undisturbed. The spectral 
information of the LISS III data could be better utilized. Kappa 
coefficient values also manifest a similar trend. 
A band substitution visually had a greater spatial resolutions 
but poor in spectral details. Owing probably to the poor 
spectral definition of land use/land cover in the latter sensor 
data resulting from replacement of the red ( 0.62 to 0.68um) 
bands of LISS III data by the PAN (0.55 to 0.75 um). The 
overall accuracy of band substituted image was 80.78 per cent. 
The band substitution visually had a greater spatial resolutions 
but poor in spectral details. In spite of the reduced spectral 
details the band substituted fused image showed a slight 
improvement in overall accuracy (1.80 %) from that of the 
parent spectral image. The Kappa statistic obtained was 0.76 
indicating that the classification was 76 per cent better than 
resulting from chance. 
However, the PCA fused image showed a larger degree of 
distorted spectral characteristics. The overall accuracy was very 
low when compared to other two fusion methods employed in 
the present study. PCA method exhibited 10 per cent 
degradation in overall accuracy from the parent spectral image, 
16.84 per cent from the IHS fused image and 12.35 per cent 
legradation from the band substituted fused image. But certain 
provements of accuracy in the PCA-fused image over the 
parent spectral image in some classes such as sugarcane (3.33 
%), banana (6.06%) and plantation crops (7.14%) were 
achieved. However, the accuracy of the remaining classes 
decreased. The Kappa statistic was 0.56. 
CONCLUSION 
Land cover classification works with the spectral information of 
an image. Among the three selected fusion approaches, IHS 
fusion approach produced high overall accuracy results than the 
parent image. Overall accuracy figures achieved from the 
various sensors data LISSIII, LISSIII and PAN merged (IHS), 
band substituted image, PCA fused image are 78.40 per cent, 
84.24per cent, 80.78 per cent and 68.43 per cent, respectively. 
The wealth of information that could be derived from space 
borne multispectral data will be further enhanced with the 
availability of higher spatial resolution data from the recently 
launched IKONOS - II and future earth observation missions 
such as Cartostat-1, IRS-P6 (Resourcesat), Cartosat-2, 
Quickbird etc. 
REFERENCES 
Benediktsson, J.A., and P.H. Swain, 1989. A method of 
statistical multisource classification with a mechanism to 
weight the influence of the data sources. Proceedings of IEEE 
symposium on Geosciencees and Remote Sensing (IGARSS), 
July, Vancouver, Canada, pp.517-520. 
    
Clic 
SPC 
ima; 
Sens 
Con 
met] 
intei 
Sen: 
Hay 
App 
mul 
Inte 
Aria 
Kim 
mul 
app! 
Mul 
Maı 
Leh 
hyp 
autc 
Woi 
Cali 
Lille 
Inte 
Mur 
Col 
Cor 
Rin 
mul 
Mul 
Mai 
Sud 
and 
use/ 
Jald 
à 
Bas 
soil 
Prir