ul 2004 
Geosci 
olution 
. IEEE WAVELE-BASED REDUCTION OF HYPERSPECTRAL IMAGERY 
image a 
) B.Salehf^ , M.J.Valadan Zoej" 
Department of Remote Sensing, Faculty of Geodesy and Geomatics Engineering, K.N.Toosi University of Technology, Tehran 
Iran 
a : e b, N : 
(salehi bahram(@yahoo.com), > (Valadanzouj(@kntu.ac.ar) 
PS WG VIVI 
KEY WORDS: remote sensing, hyper spectral, analysis, classification, reconstruction, spectral 
ABSTRACT: 
New sensor technology has made it possible to gather multispectral images in hundreds and potentially thousands of spectral bands, this 
tremendous increase in spectral resolution should provide a wealth of detailed information, but the techniques used to analyze lower 
dimensional data often perform poorly on high dimensional data. Therefore, it is necessary to investigate the problem and to explore 
effective approaches to hyperspectral data analysis. Studies indicate that the key problem is to need very large number of labeled samples. 
It has been found that the conventional approaches can be retained if a preprocessing stage is established. 
Dimension reduction is a preprocessing stage that brings data from a high order dimension to a low order dimension. Some stochastic - 
based techniques are used for dimension reduction such as Principal Component Analysis (PCA), Discriminant Analysis Feature 
Extraction (DAFE) and Decision Boundary Feature Extraction (DBFE).But these techniques have some restrictions. For example PCA is 
computationally expensive and does not eliminate anomalies that can be seen at one arbitrary band; the number of training samples is 
usually not enough to prevent singularity or yield a good covariance estimate in DBFE. 
Spectral data reduction using Automatic Wavelet Decomposition could be useful because it preserves the distinction among spectral 
signatures. It is also computed in automatic fashion and can filter data anomalies. This is due to the intrinsic properties of Wavelet 
Transform that preserve high and low frequency feature therefore preserving peaks and valleys found in typical spectra. Compared to 
PCA, for the same level of data reduction this paper shows that automatic wavelet reduction yields better or comparable classification 
accuracy. 
1. INTRODUCTION declines as the number of spectral bands increases, which is 
often referred to as the Hughes phenomenon (Hughes, 1968), as 
Multispectral sensors have been widely used to observe Earth shown in figure 1. 
surface since the 1960's. However, traditional sensors can only 
collect spectral data less than 20 bands due to the limitation of 
sensor technology. In recent years, spectral image sensors have 
been improved so as to collect spectral data in several hundred 
bands, which are called hyperspectral image scanners. For 
example, the AVIRIS scanners developed by NASA JPL 
provide 224 contiguous spectral channels (Hsu, ParHui, 2000). 
Theoretically, using hyperspectral images should increase our 
abilities in classifying land use/cover types. However, the data 
classification approach that has been successfully applied to 
multispectral data in the past is not as effective for 
hyperspectral data as well (Hsieh and Landgrebe, /998). 
As the dimensionality of the feature space increases subject to 
the number of bands, the number of training samples needed 
  
Mean Recognition Accuracy 
ta 
Measurement Complexit n [total discrete value) 
for image classification has to increase too. Fukunaga (1989) Figurel. Mean recognition accuracy vs. measurement of 
proved that the required number of training samples is linearly complexity for the finite training cases ( Houghes, 1968) 
related to the dimensionality for a linear classifier and to the 
square of dimensionality for a quadratic classifier. It has been One of the approaches to improve the classification 
estimated that as the number of dimensions increases the performance is to reduce dimensionality via a preprocessing 
training samples size need to increases exponentially in order to method, which takes into consideration high dimensional 
have an effective estimate of the multivariate densities needed spaces properties. Dimension reduction is the transformation 
to perform a non-parametric classification. If training samples that brings data from a high order dimension to a low order 
are insufficient for the need, which is quite common for the dimension, similar to lossy compression method, dimension 
case of using hyperspectral data, parameter estimation becomes reduction reduced the size of the data but unlike compression, 
inaccurate. The classification accuracy first grows and then dimension reduction is applicant-driven (Kaewpijit et a/, 2003) 
61