IAPRS & SIS, Vol.34, Part 7, "Resource and Environmental Monitoring", Hyderabad, India,2002
Wavelet based fusion of ASTER (VNIR and SWIR) bands
for improved soil information extraction
K. VANI*, S. SANJEEVI** and A. RAVINDRAN*
*Institute of Remote Sensing, Anna University, Madras-25. India
**Centre for Geoscience and Engineering, Anna University, Madras-25. India
Email: vani@annauniv.edu ssanjeevi@annauniv.edu
KEYWORDS : FUSION, ASTER, SOIL, WAVELET, SWIR
ABSTRACT
Image fusion is a technique that has traditionally been attempted to improve the spatial resolution of multispectral data. However, not much
emphasis has been laid on the improvement of spectral details by fusing two multispectral images with differing spectral ranges. This study
is an attempt to fuse such images of an agriculturally dominant area, acquired in the VNIR and SWIR regions, by the ASTER (Advanced
Space-borne Thermal Emission and Reflection Radiometer) sensor on board Terra 1. Optimal combinations were tried out from amongst 9
bands (3 in VNIR + 6 in SWIR) and image fusion was attempted using Principal Component Analysis and wavelet transformation
techniques. The resulting images were interpreted to prepare soil maps of the area. An existing soil map was used as the basis for
comparison of the results. The combination that resulted in maximum contrast amongst the soil types is R=b3,G=b2,B=b1 for the VNIR
bands and R=b4,G=b6,B=b9 for SWIR bands. It is observed that the fused image obtained by wavelet transformation resulted in exhibition
of the maximum number (seven) of soil types when compared to the fused image obtained by PCA technique. Fusion using PCA technique
resulted in enhancing the moist/wet soil types, while the wavelet transformation technique enhanced the saline and alkaline soils. Thus,
improved soil mapping has been possible due to image fusion, and this study has demonstrated the need for image fusion in the multispectral
domain, apart from attempting in the spatial domain, to bring out more information about soil types.
1. INTRODUCTION
The currently available high spatial resolution images have
limitations in conveying information about the spectral
characteristics of certain land cover features. Such limitations,
however, could be overcome by multisensor image fusion
techniques. Fusion is defined as combination of multiple images to
form a new image for a certain application using a certain algorithm
(Pohl,1998). The enhanced image obtained by image fusion can be
used for identifying and discriminating land cover features such as
vegetation types, water, moisture, soil types, concrete, asphalt.
Much work has been carried out in development of image fusion
techniques and their applications. Chavez et al. (1991) explain
about IHS, PCA and HPF methods to merge multi-resolution and
multi-spectral data (LANDSAT TM and SPOT Panchromatic) and
compare the methods. The authors observed that the IHS method
distorted the spectral characteristics of the data the most, followed
by PCA and HPF methods. Crawford et al.(1999), fused airborne
polarimetric and interferometric SAR data for mapping coastal
environments. Neural network and Bayesian pair wise classifiers in
a multi-resolution framework was utilized and an accurately
classified map of the coastal test site was obtained. Li and Sheng
(2000) describe about a fusion scheme based on “a trous” wavelet
transformation for the fusion of IR and visible images. The authors
report that the spatial resolution improvement of IR image was
achieved and the salient information from both visible and IR image
were preserved. Garguet-Duport et al. (1996) attempted fusion
using multi-resolution analysis based upon wavelet transform for
SPOT PAN and SPOT XS data. The authors also found this method
very useful and particularly well adapted to vegetation analysis.
From the above examples, it is observed that the wavelet
transformation technique of image fusion seems to be
advantageous, compared to the older techniques.
Conventionally, image fusion has been attempted using high
resolution panchromatic and low resolution multispectral image set
or multispectral and radar images. The limitation of such studies is
that the multispectral information content about the land cover
features are made available only from one multispectral dataset, the
other set being panchromatic. In such studies, only spatial details
are enhanced and not much information about the spectral details is
conveyed (eg soil, water, and vegetation types). This study attempts
to fuse two image data sets of differing resolutions, both in the
multispectral domain (VNIR & SWIR). The highlight of the image
data used here is that both the multispectral image data sets are
provided by a single coverage by the same sensor (thus avoiding
multi date images, varying atmospheric conditions, differing look
angle and geometry). This study also aims to utilize the results of
such a fusion for soil mapping in an agriculturally dominant area in
Tamilnadu.
Soil spectral signatures result from the presence or absence, as well
as the position and shape of specific absorption features of its
constituents. The visible and NIR regions (0.4?m - 1.37m) are
characterized by broad spectral absorption features (ferrous iron
absorption feature near 1 ?m). Absorption features at 1.4?m -
1.9?m are due to unordered arrangement of water molecules in
soil, features in the 1.8? m - 2.5? m region are due to the presence of
OH , CO; and SO, molecules and absorption near 1.4? m -2.2?m is
due to layer silicate structure and moisture.
(v.
(the