3.5 ISD Algorithm
The Iterative Self-Organizing Data Analysis Technique
(ISODATA) is a widely used clustering algorithm (Tou and
Gonzalez, 1977; Sabins, 1987). ISODATA represents a fairly
comprehensive set of heuristic (rule-of thumb) procedures that
have been incorporated into an iterative classification
algorithm. Many of the steps incorporated into the algorithm
are a result of experience gained through experimentation.
ISODATA calculates class means evenly distributed in the data
space and then iteratively clusters the remaining pixels using
minimum distance techniques. Each iteration recalculates
means and reclassifies pixels with respect to the new means.
This process continues until the number of pixels in each class
changes by less than the selected pixel change threshold or the
maximum number of iterations is reached.
3.6 Accuracy Evaluation for Classification Map
The overall accuracy and KAPPA analysis were used to
perform classification accuracy assessment based on error
matrix analysis. By using simple descriptive statistics
technique, overall accuracy is computed by dividing the total
correct (sum of the major diagonal) by the total number of
pixels in the error matrix. KAPPA analysis is a discrete
multivariate technique of use in accuracy assessment
(Congalton and Mead, 1983; Jensen, 1996). KAPPA analysis
yields a. K,. statistic (an estimate of KAPPA) that is a measure
of agreement or accuracy (Congalton, 1991). The K,. statistic
is computed as
o iei (11)
where r is the number of rows in the matrix, X; is the number
of observations in row i and column i, and X;, and X,, are
the marginal totals for row i and column i, respectively, and N
is the total number of observations.
4. RESULTS AND DISCUSSION
Remotely sensed data processing, including preprocessing,
vegetation index computation, false color composite, etc. were
carried out using two software packages, ENVI and
CITYSTAR. ENVI (the Environment for Visualizing Images)
is a "state-of-the-art image processing system designed from
the ground up to provide turn-key data visualization and
analysis of satellite and aircraft remote sensing data”
(Research Systems, Inc., 1996). CITYSTAR is an integrated
remote sensing, GIS and GPS multimedia system, developed
by the Institute of Remote Sensing and GIS, Peking University,
China. The microcomputer-based CITYSTAR system is mainly
composed of such subsystems as graphic edit, graphic view,
thematic map plot, remote sensing analysis, remote sensing
imagery mapping, remote sensing image processing, DEM &
DTM production, 3-D analysis, spatial analysis, GPS, etc.
4.1 Image Preprocessing
With constraints such as spatial, spectral, temporal and
radiometric resolution, relatively simple remote sensing
devices can not record the complex Earth’s land and water
surface well. Consequently, error creeps into the data
acquisition process and can degrade the quality of the remote
sensor data collected. Therefore, it is necessary to preprocess
the remotely sensed data prior to actually analyzing it. As a
kind of commercially remote sensor data, the radiometric and
systematic geometric errors of Landsat TM data have been
removed by the data provider, while the unsystematic
geometric error remains in the image. The unsystematic errors
of the Landsat TM data were corrected before the analysis of
land cover change.
4.2 Vegetation Index Computation
Vegetation index indicates the amount of green vegetation, ,
which is useful and important for land cover identification
because the cover of land on the earth are mostly vegetation.
Much of the research in the measurement of vegetative amount
and condition has involved in the analysis of remote sensing
spectral measurements (e.g., Landsat multispectral scanner
(MSS), TM, NOAA Advanced Very High Resolution
Radiometer (AVHRR), SPOT High Resolution Visible (HRV),
etc). Various models have been developed to express
vegetation index, such as Vegetation Index (VI), Ratio
Vegetation Index (RVI), Normalized Difference Vegetation
Index (NDVI), etc. NDVI model used to compute the
vegetation indexes in this study is (Rouse et al., 1973)
NDVI- TM4-TM3 (12)
TM4 + TM3
where TM3, TM4 are gray values of TM band 3 and band 4,
respectively.
43 False Color Composite
In order to perform training data collection, it is necessary to
make false color composite images. It's composed of two steps.
The first step is to test composite schemes by using Landsat
TM data acquired on May 20, 1993, with a) selecting 3
optimum bands from the 7 bands; b) using all 7 bands; and c)
using all 7 bands and NDVI. The second step is to perform
false color composition for all dates of Landsat TM data.
Based on the test study, the best composite scheme is
R=0.7 x TM3+03 x TM6 (13)
G=05 x TM2+0.5 x TM4 (14)
B=03 x TM1+0.3 x TMS +0.4 * TM7 (15)
where TM1, TM2, ..., TM7 are the gray values of TM band 1,
band 2, ..., band 7, respectively. The scheme was used to make
false color composite images, r85fcc.img, r87fcc.img,
r90fcc.img and r93fcc.img, from Landsat TM data obtained on
May 14 of 1985, May 20 of 1987, April 26 of 1990 and May 20
of 1993 respectively.
402 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
4.5
data
algo
pref
is St
is r-
Twc
coll
r93f
clas:
of cl
PAF
subr
clas:
