te and 
marsh, 
cation, 
d their 
jotoRS 
) times 
select 
has 12 
indary 
h and 
11,000 
f main 
of the 
, non- 
. The 
's and 
, most 
: 2001 
nages 
table 
IAPRS & SIS, Vol.34, Part 7, “Resource and E 
nvironmental Monitoring”, Hyderabad, India, 2002 
  
2.2.2 Administrative Boundary: The county’s boundary was 
traced from the rectified 1:50000 maps. 
2.2.3 Reference Material: 1-50000 maps for images 
rectification, merged IKONOS images for grassland and city 
expansion information extraction, 1:100000 maps (1983) for 
marsh interpretation, "Tianjin land use overall planning 
map(1996-2010)" for marsh interpretation and "Tianjin city 
overall planning map(1996-2010)" for marsh interpretation. 
2.3 Method 
The processing flow is shown in Figure 1: 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Image l Otimel 0 Image2(time2) 
E + c y 
matchin m 
| Rectification d 2 = 
Y E 
ymerge =. 
S 
| Merged Image | 
| v 
grassland) Y (marsh/grass) 
= 
Extract Change Information Supervised Classification B 
= 
l | e 
y = 
=. 
| Post-process changed information | S 
Y = 
sig a > 
| Statistic and Mapping | v S 9 
  
  
  
  
  
Figure 1. Processing Flow 
2.3.1 Data Preparation: Using our software package PhotoRS 
to rectify imagel of timel. The control coordinates were 
acquired from 1:50000 digital raster graph, which had been 
rectified by our other software package PhotoDRG. The 
method of matching image2 to imagel is to rectify image2 with 
the coordinates acquired from the geocoded imagel. The next 
step is to merge the two images into one after both had been 
enhanced. 
2.3.2 Marsh Information Extraction: 
1)  Supervised Classification 
Firstly, keep a group of target classes in mind, such as lake, 
river, sea, swamp and so on, which can provide a framework 
for organizing and categorizing the information to be extracted 
from the image, then create the appropriate signature from the 
data. Supervised classification was performed on the TM 
images of 1999 using ERDAS Imagine 8.4. But the results were 
not so ideal, because there were some noises and false 
information except the true marsh spots. 
2)  Editand Detailed Classified Information 
The method of modifying the classified information is visual 
analysis/interpretation and tracing. Using PhotoShop, a psd file 
can be created which have eight layers. One layer is for the 
thematic image (1999, obtained by step 1). One layer is for 
merged image of 2001 (TM + SPOT). The others were set to 
hold change information, including six types, such as “lagoon”, 
“river”, “lake”, “coast lagoon”, “manmade pools” and 
"swamp". Referencing to "Tianjin land use overall planning 
map" and according to the merged image, the boundary of the 
six kinds of features were traced and put into each layer. 
2.3.3 Grassland Information Extraction: 
1)  Supervised classification 
There are four kinds of grassland: "residential grassland", 
“focus grassland”(lawn in large square or in large park), 
“trees along road” and “trees or grass in the country”. 
Supervised classification was performed and a classified 
image was created. 
2) Edit classification information 
Some lawns among the buildings could not be extracted by 
supervised classification, because they were covered by 
building's shade. Those lawns and grassland which had not 
identified exactly by computer were traced by expert. The false 
information should be removed, for example, suburb's 
farmland had been classified as grassland. Because no recent 
material to help us judging each grassland type, the decision 
was made based on the image texture, color, spatial, spectral 
information and especially experts' experience. 
2.3.4 City's Reconstruction and Expansion Information 
Extraction: This information's extraction mainly depended 
upon individual analysis and interpretation. Because the image 
spectrum around city is so complex, the supervised or un- 
supervised classification method is not very effective. 
Therefore, the 1999 TM image and 2001 SPOT image were 
merged after each image had been enhanced. It is easy to see 
that the bright green pixels represent a new building on 
farmland or other vegetation. If they were outside the city, then 
represent the city expansion information. According to the 
referenced material it can make sure which buildings were 
created on farmland and which ones were created on non- 
farmland during last year. 
The renovation information within the Tianjin Outer Ring Road 
was extracted by comparing merged image of 1999 with 
merged image of 2001. It can be seen that new buildings have 
more perspicuity texture and brighter color than old ones. 
785