International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
2.6 Image classification 
Image analysis was performed using supervised and a 
new hybrid approach (digital and visual) classification 
method. At first, the image were classified in classes 
including forest and  non-forest using traditional 
maximum likelihood classifier without any knowledge 
of a-priori possibilities. All multispecral bands (except 
thermal band), fused bands and synthetic bands such as 
those derived from PCA and ratios were used by 
classification process. Required training areas were 
defined trough fieldwork. The best band sets were 
selected using Bhatacharrya distance criterion and the 
defined training areas. To eliminate the isolated 
classified pixels, the resultant classifications were 
filtered with a majority filter in a 5*5 moving window. 
The forest/non-forest classification was also carried out 
using visual interpretation at computer display, a new 
approach called hybrid. The main advantage of hybrid 
interpretation is that contextual information and expert 
knowledge can be used in the analysis more easily. To 
perform hybrid classification, the most accurate map 
derived from maximum likelihood classification was 
used. This map was converted to vector format and then 
it was edited on the basis of various color composites, 
fused images. 
3. RESULTS AND CONCLUSION 
The main purpose of this study is comparison of the 
potential of the Landsat7- ETM-- data and SPOT5-HRG 
image for forest area mapping at the scale of 1:25000 in 
northern of Iran. The quality analysis of the satellite data 
indicated that the quality of the level 1 ETM+ was very 
good. In contrast, a  non-systematic geometric 
misregistration between HRG-XS and HRG-Pan of 
SPOTS data could be recognized. It ranges from 3 to 15 
pixels. There were no other radiometric or geometric 
distortion. Both satellite images were orthorectified very 
precisely in comparison with the digital topographic 
map. The geometric misregistration of the SPOTS bands 
could be corrected through the orthorectification. To 
assess the capability of the landsat7-ETM+ and the 
SPOTS-HRG data to discriminate forest area, the results 
of the classification were compared pixel by pixel to the 
ground truth. The maximum likelihood classifier 
concluded overall accuracies and Kappa coefficients 
equal to 89% and 0.84 for Landsat7 and 93% and 0.89 
for SPOTS. Better results have been achieved from the 
hybrid classifications since this approach pays particular 
attention to texture and knowledge of expert. Similar 
conclusion was reported by Rafieyan ef al. (2003). The 
achieved overall accuracies and Kappas through hybrid 
approach are equal to 93% and 0.89 for Landsat7 and 
97% and 0.93 for SPOTS. The spectral data fusion 
technique DIRS, which preserves the spectral 
424 
characteristics of each multispectral band, had improved 
the classification results at 1% by both data sets. The 
performance of SPOTS data is on behalf of its high 
spatial resolution, which permits to distinguish small 
forest and non-forest polygons. Revision of forest roads 
was also precisely possible trough SPOTS image. 
Furthermore, determination of forest /non-forest 
boundary by SPOTS data could be done more precisely 
than by Landsaty7 data. Three additional spectral bands 
of ETM+ (related to HRG), which lie in the blue, mid 
infrared and thermal infrared were not selected by the 
best bandsets for the classification. It indicates that the 
spectral resolution of SPOTS is sufficient for such 
proposes. 
The results of this investigation can lead to the 
conclusion that in such regions both Landsat7 and 
SPOTS data are suited for forest mapping. But SPOTS 
data meets specification of large scale mapping 
(1:25000) and offers forest managers valuable sources of 
data for fine forest mapping. Its potential with regard to 
forest stands mapping should be evaluated. 
4. ACKNOWLEDGMENTS 
The authors would like to thank the Forest, Range and 
Watershed Organization of Iran for providing SPOTS 
data and National Cartographic Center (NCC) which 
provided the current aerial photographs. 
5. REFERENCES 
Brockhaus, J. A. and S. Khorram, 1992. A 
comparison of SPOT and Landsat-TM data for 
use in conducting inventories of forest 
resources. /nternational Journal of Remote 
Sensing, Vol.13, NO. 16, pp. 3035-3043. 
Cheng, P. T. Toutin, and V. Tom, 2002. 
Orthorectification and Data Fusion of Landsat7 
Data, Unpublished. 
Darvishsefat, A. A., 1995. Einsatz und Fusion 
von Multisensoralen Satellitenbilddaten zur 
Erfassung von Waldinventuren. Remote Sensing 
Series, Vol. 24, Department of Geography, 
University of Zurich, Switzerland. 
Joffre, R., 1991. Estimation tree density in Oak 
Savana-like of southern Spain from Spot data. 
International Journal of Remote Sensing, Vol. 
14, NO. 16, pp. 685-697. 
Munechika, C. Ka 1990. Merging 
Panchromatic and Multispectral Images for 
Enhanced Image Analysis. M.S. Thesis, Center 
Int 
2.6 | 
Ima, 
new 
met] 
inch 
max 
of a 
ther 
thos 
clas: 
defi 
sele 
defi 
clas: 
filte 
The 
usin 
app! 
intel 
kno 
perf 
dert 
usec 
it w 
fuse 
3.R 
The 
pote 
ima, 
nort 
indi 
200! 
mist 
SPC 
pixe 
dist 
prec 
map 
coul 
asse 
SPC 
of tl 
grot 
conc 
equi 
for | 
hybı 
atter 
cone 
achi 
appt 
97% 
tech