2008
1125
LAND COVER CLASSIFICATION FOR FUTURE-ORIENTED DEVELOPMENT PLAN
OF MULTIFUNCTIONAL ADMINISTRATIVE CITY OF KOREA
Han, Seung-Hee a ’ *, Lee, Jin-Duk b
a *Dept. of Civil & Environmental engineering, Kongju National University, Seoul KOREA - shhan@kongju.ac.kr
b Dept, of Civil & Environmental engineering, Kumoh National Institute of Technology, Seoul KOREA -
jdlee@kumoh.ac.kr
Commission VII, WG VII/6
KEY WORDS: KOMPSAT2, Land cover classification, Aster, Objected based classification
ABSTRACT:
In this study, land cover classification and NDVI evaluation are carried out in surrounding areas of Yeongi-gun, Chungcheongnam-
do (132km 2 ) where a project for multi-functional administrative city is promoted by the Korean government. Image acquired from
KOMPSAT 2, LANDSAT and ASTER is utilized and comparative evaluation on limitation in classification based on resolution was
carried out. The area mainly consists of arable land including mountains, rice fields, ordinary fields, etc thus special attention was
paid to the classification of rice fields and ordinary fields. For the classification of image acquired from KOMPSAT 2, segmentation
technique for classification of high-resolution image was applied. To evaluate the accuracy of the classification, field investigation
was conducted to examine the sample and it was compared with the land usage and classification of land category in land ledger of
Korea. Acquired results were made into theme map in shape file format and it would be of great help in decision making of policy
for the future-oriented development plan of multi-functional administrative city.
1. INTRODUCTION
1.1 General Instructions
To acquire recent middle classification land coverage, coverage
classification using high resolution satellite image is effective
method. However, in case of applying the high resolution image
to previously pixel based coverage classification method. In
case of high resolution panchromatic image or multi-spectral
image, necessary information shall be extracted based on
spatial/spectral traits. It is trend to classify using the region of
interest defined as the spatial/spectral traits and surface traits in
image, namely, object based classification method.
In this study, not only classification with previous pixel based
classification method regarding the Seijong-si, the multi
functional administrative city, using Aster image but also object
based coverage classification using high resolution image of
KOMPSAT2 was carried out. Also, it shall be compared with
the result of image classification using the naked-eye distinction
and spectral traits for the verification on coverage classification
of high resolution image. Also, land use thematic map was
created using the land category of cadastral map and it was
compared with result of coverage classification.
2. OBJECT BASED CLASSIFICATION
When applying high resolution image to previous pixel based
classification method, it is difficult to use with variation of each
trait of pixel since it is classified in scatter. Namely, high
resolution image has severe spectral variation and there is
limitation for classification in unit of pixel. Ultimately, previous
classification algorithm has no meaning in high resolution.
(Van der Sande, et al.,2003). Therefore, it is a strategy to
classify large unit in previous unit of pixel and apply spatial
classification method in unit of shape for small unit. In this
study, object based classification method as Figure 1 was
applied. Looking into the stage of classification in Picture 1,
region of pixel is separated by determining the cluster size to
classify segment regarding the image, namely resolution.
Determination of resolution cannot be complete at once and it
should go through trial & error process based on experience.
The next stage is to calculate the attributes of each region such
as spatial, wavelength, band ratio, etc and save the data to
memory. Supervised classification rule based classification is
processed after defining the objects within image. Rule based
classification is the method to grant the attribute by rule set
intervened of subjectivity of operator with form of segment,
NDVI, texture, ratio of length and height.
Figure 1. Workflow for Objected based classification