This study aims to establish a practical image analysis method for the use of middle-scale resolution images acquired by
the multi-spectral sensors such as Landsat-7/ETM+, Terra/ASTER and ALOS/AVNIR-2 as the complementary data
sources of higher resolution images such as Quickbird for the purpose of environmental monitoring of wide-range areas.
For this purpose, an image analysis based on mixture is investigated as one of the effective approaches. As the
information target, we selected vegetation cover rate (VCR) in urban area because it is one of the important
environmental factors to affect urban environment issue such as heat island phenomena.
In order to realize easy and efficient computation for estimating the mixture rate of vegetation categories, the linear
mixture model using two main categories including vegetation and non-vegetation, is applied in combination with the
least square estimation of multi-regressive coefficients for vegetation cover rate (VCR) and non-vegetation cover rate
(non-VCR) with several bands data by multi-spectral sensors. In addition, two sub-categories for both of vegetation and
non-vegetation categories are considered to specify representative pixel values as correct as possible, that is, trees and
grasses for vegetation, and buildings and bare-soils for non-vegetation respectively, and their optical mixture rates are
estimated as well as the mixture rate of vegetation and non-vegetation categories. For this purpose, an iterative procedure
is adopted, in which each mixture rate of two sub-categories for vegetation and non-vegetation is varied by ten percent
steps and the least square estimation is applied with all combinations of mixture rates of sub-categories for vegetation
and non-vegetation.
The experiments for VCR extraction were conducted in the test site of Hiroshima-city and by using multi-spectral data
acquired by Landsat-7/ETM+, Terra/ASTER, and ALOS/AVNIR-2. The accuracy for VCR extraction was evaluated
based on the comparison with the VCRs obtained by means of pixel-wise vegetation/non-vegetation classification of a
Quickbird multi-spectral image. The result shows that the number of bands is one of the important parameters in general.
However, it was verified that the combination of wavelength regions is more important than the number of bands. The
result of this study suggests that the combination of wavelength regions is essential in middle-resolution multi-spectral
images for vegetation cover rate estimation based on mixture analyses.
|