example, the main aim of this paper is to compare these 
two strategies and to see which one is better and more 
robust. 
2. PREVIOUS WORK 
In this part, we retrieve some previous work on land 
cover/use classification and on crop area calculation in 
an administrative division. It showed that much of these 
work was done based on strategy A (cut and classify) 
using 
classification method as well as visual interpretation. 
supervised or unsupervised automatic 
Strategy B (classify and cut) was only used by fewer 
investigators, in whose cases supervised classification 
method was mainly applied. 
2.1 Strategy A 
In inland wetland change detection in the Everglades 
Water Conservation Area 2A(WCA-2A), Rutchey and 
Vilchek(1994) use Landsat Multispectral Scanner(Mss) 
data and SPOT High Resolution Visible(HRV) 
multispectral data to inventory aquatic macrophyte 
changes. Firstly, the polygon boundary of the study area 
was rasterized to a UTM map projection. The mask was 
applied to each of the rectified images. Thus only land 
within WCA-2A was allowed to contribute to the cluster 
development in the cluster development in the 
classification phase. A standard statistical unsupervised 
classification of the study area was performed, yielding 
30 clusters and then aggregated to 7 classes(Jensen et al. 
1995). Multitemporal classification results were used 
to identify the change in the spatial distribution of 
aquatic macrophyte. 
Bauer et al.(1994) described the procedure to inventory 
Minnesota forest resources using multitemporal Landsat 
TM data. Because of the large study area, three scenes 
were mosaiced and then cut down the target area(5 
counties). Consequently, unsupervised classification 
was conducted and 11 forest cover type map and their 
241 
corresponding areas were got. Change trend analysiscan 
be conducted with multi-year areas. It should be noted 
that the last forest covertype map was not obtained from 
mosaiced TM data. 
images were processed individually and the 
were merged as classified(GIS) files. 
classifying the Three separate 
results 
Strategy A was applied by previous researchers not only 
in unsupervised classification but also in supervised 
classification. Ray et al.(1994) estimated cotton 
production in India using IRS-1B and meteorological 
data. After mosaicing two scenes of LISS -1 data, study 
area boundary pixels were used to cut down the target 
image. Maximum likelihood classification scheme was 
used for crop identification and acreage estimation. 
The same strategy was applied in Turkey in acreage 
estimation of wheat and barley(Pestemalei et al., 1995). 
2.2 Strategy B 
Contrary to the above most authors who use strategy A 
in their experiment, fewer authors applied strategy B in 
their project. Hall-Konyves(1990) introduced basic 
principals and techniques in Sweden to monitor the crop. 
Maximum classification technique was applied by the 
author and to improve the accuracy of the classifier, a 
post-classification clean-up filter was applied. The 
author states "no area was masked out prior to 
classification". 
In the pro-harvest state level wheat acreage estimation of 
Punjab, India, Makey et a1.(1993) obtained the wheat 
acreage for six stratausing supervised classification. But 
the author hasn't mentioned which strategy was applied. 
3. EXPERIMENTAL 
Since strategy A has been applied by most pervious 
researchers insupervised and unsupervised classification 
and strategy B in supervised classification, the main aim 
of our experiment is to see the effect of strategy B in 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996