OBJECT-BASED CLASSIFICATION USING HIGH RESOLUTION SATELLITE DATA
AS A TOOL FOR MANAGING TRADITIONAL JAPANESE RURAL LANDSCAPES
K. Takahashi 1 * 3 ’*, N. Kamagata 3 , K. Hara b
3 Graduate School of Informatics, Tokyo University of Information Sciences, 4-1 Onaridai, Wakaba-ku, Chiba 265-
8501, Japan - (h07001kt, h05002nk)@edu.tuis.ac.jp
b Dept, of Environmental Information, Tokyo University of Information Sciences, 4-1 Onaridai, Wakaba-ku, Chiba
265-8501, Japan - hara@rsch.tuis.ac.jp
WG IV/9 - Mapping from High Resolution Data
KEY WORDS: Ecology, Land Cover, Mapping, Classification, IKONOS, Object, Segmentation, Texture
ABSTRACT:
A system for object-based classification of landscape types incorporating topographic data was designed and tested in a Japanese
countryside target area. IKONOS data (Japan Space Imaging - multi-spectral resolution 4 meter, panchromatic resolution 1 meter)
acquired on 23 April 2001 was utilized, and the classification used Definiens Ver.5 software (Definiens). Initial segmentation was a
muli-resolution, bottom-up system, and each segment identified was considered to be one object. Topographic data was derived
from field surveys and topographic maps. Using a master landscape map for assessment of accuracy, two classifications, employing
the same landscape types and ground truth, were implemented. One classification employed only the extant classification system
based on spectral characteristics; while the other featured integration of the topographic data. The results showed that in terms of
overall classification accuracy, the object-based using topographic data (56%) scored higher than the object-based alone (52%). In
terms of overall Kappa index as well, object-based with topographic (0.51) outscored the object-based only (0.47). Several
misclassifications seen in the extant system were eliminated by addition of the topographic data. These results indicate that object-
based classification of very high resolution data combined with topographic data produces a system that promises to be an effective
tool for landscape classification. The results also indicate that additional gains in accuracy may be achieved by developing a system
for adjusting scale parameter according to the characteristics of individual landscape types.
1. INTRODUCTION
Accurate identification and continuous monitoring of landscape
elements is essential for conservation of regional environments.
This is especially so in many areas of Japan, where traditional
countryside ecosystems can be found on the outskirts of major
cities. These ecosystems are highly susceptible to disruption by
various factors, including urbanization, illegal dumping of
industrial refuse, and changes in lifestyle patterns.
Remote sensing technology has been identified as an effective
means for fast and continuous landscape monitoring. In
particular, very high resolution satellite data such as IKONOS,
which became available after 2000, provides image quality high
enough to allowed accurate extraction of even small landscape
elements. This is of especial importance in the Japanese
countryside, which often consists of various small landscape
elements mixed together in a complicated mosaic pattern.
In addition, recent research (Kamagata et al., 2006) has shown
that the object-based classification method, when applied to
very high resolution data, can be an effective tool for
classification of vegetation and land cover. In addition, Hara et
al (2007) have used this approach to extract landscape elements.
Extant object-based image analyses, however, rely on
classification using spectral characteristics of the image. When
dealing with landscapes, classifications should include not only
land-use data but topographic elements as well. The goal of
this research is to add topographic data to object-based
classification of IKONOS very high resolution data, and to test
the ability of this system to extract and classify landscape types.
2. STUDY AREA AND METHODS
2.1 Study Area
This research was implemented on a 1500 meter by 2000 meter
test plot, located in an agricultural area of Sakura City in
northern Chiba Prefecture, about 40 kilometers northeast of
Tokyo. Sakura City is situated in the center of the Shimousa
Uplands (Figure 1). The topography in the study area consists
of relatively flat, terrace-like uplands, around 30 meters above
sea level in elevation, into which narrow, branching valleys
have been cut. The traditional land use pattern is irrigated rice
paddy on the marshy soils that form the valley bottoms, and dry
vegetable fields and orchards on the uplands. The short but
steep slopes between the valley bottoms and the uplands were
used for bamboo groves and oak coppices (Fujiwara et al.,
2005).
Corresponding author.