The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
410
has been evaluated for discrimination of vegetation area from
others. Atmospheric correction and other pre-processing
operations on the hyperspectral imagery have been performed
by using routine procedures.
Spectral unmixing algorithms use a variety of different
mathematical procedures to endmember extraction and estimate
abundances. Unmixing problem comprises three sequential
steps: dimension reduction, endmember determination, and
inversion. Because hyperspectral scenes can include extremely
large amount of data, some algorithms for spectral unmixing
first use image itself to estimate endmembers present in the
scene. The dimension-reduction stage reduces the dimension of
the original data in the scene. This step is optional and is
invoked only by some algorithms to reduce the computational
cost of subsequent processing; it also selects bands with higher
signal to noise ratio (SNR) to separate endmembers spectra
without any data loss. We use the MNF method to achieve this
goal. In this paper, we introduce that the site study and EEA that
applied on image then emphasize on the second step of
unmixing problem. Finally represent the result of this research.
1.2. THE SITE STUDY
An agricultural area located in southern parts of Tehran, known
as Ahmadabad has been selected as the study site. Wheat and
barley are the main agricultural crops in the area. More than 30
fields of detailed ground-truth dataset have been visited in the
field and their records have been used as a ground truth data for
training and verifying the results of the classification.
1.3. OPTIMAL BANDS IN HYPERION IMAGES
Hyperion data were acquired over the Ahmadabad village on
May 21, 2002 at 06:57:56 GMT. The EO-1 satellite is a sun-
synchronous orbit at 705 km altitude. Hyperion data includes
256 pixels with a nominal size of 30 m on the ground over a
7.65 km swath. Well-calibrated data (Level 1B1) is normally
available. Post-Level 1B1 processing of the dataset, as
performed in this study, included correction for bad lines,
striping pixels and smile, atmospheric correction and co
alignment. Hyperion data is acquired in pushbroom mode with
two spectrometers. One operates in the VNIR range (including
70 bands between 356-1058 nm with an average FWHM of
10.90 nm) and the other in the SWIR range (including 172
bands between 852- 2577nm, with an average FWHM of 10.14
nm). 44 of 242 bands including bands 1-7, 58-76 and 225-242
are set to zero by TRW software during Level 1B1 processing
[ 2 ][ 3 ]. Post-level 1B1 data processing operations for preparation
of the Hyperion data for classification including band selection,
correction for bad lines, striping pixels and smile, a pixel-based
atmospheric correction using FLAASH [2] and a co-alignment
were performed.
2. IDENTIFY OPTIMAL ENDMEMBER
An important problem in Hyperspectral image processing is to
decompose the mixed pixels into the materials that contribute to
the pixel, endmember and a set of corresponding fractions of the
spectral signature in the pixel, abundances, and this problem is
known as the unmixing problem.
According to the definition, an endmember is an idealized pure
signature of a class. Endmember extraction is one of the
fundamental and crucial tasks in hyperspectral data exploitation.
It has received considerable interest in recent years, with many
researchers devoting their effort to develop algorithms for
endmember extraction from hyperspectral data.
Hyperspectral imaging sensors on environmental applications
have high spectral resolution and low spatial resolution so that
numerous disparate substances can contribute to the spectrum
measured from a single pixel or in the field of view of the
sensor. A mixed pixel occurs due to high spectral resolution or
low spatial resolution when more than one material substance
present in a pixel, in this case these substances are considered to
be mixed linearly or nonlinearly in the pixel.
Recent researches aim to identify the individual constituent
materials existing in the mixture pixel, as well as the
proportions in which they appear. Spectral unmixing is the
procedure by which the measured spectrum of a mixed pixel is
decomposed into a collection of constituent object spectra that
present in scene. In estimating the constituent members and
abundances in pixel spectra, unmixing algorithms incorporate
philosophical assumptions regarding the physical mechanisms
and mathematical structure by which the reflectance properties
from disparate substances combine to yield the mixed pixel
spectra. Detection algorithms make similar assumptions when
testing for the existence of a specific substance in a mixed pixel.
Not surprisingly, the hyperspectral detection and unmixing
problems are closely related.
2.1. PPI
The Pixel Purity Index (PPI) has been widely used in
hyperspectral image analysis for endmember extraction due to
its publicity and availability in the Environment for Visualizing
Images (ENVI) software.[4] In this experiment, the PPI was
implemented to find endmembers for the image scene in Fig 1,
using the same set of randomly generated initial skewers. Fig 2
shows the endmember pixels extracted by the PPI, respectively.
Fig 1 Fig 2
Fig 1 :Hyperio image scene subset
(R=650.67,G=569.27,B=477.69)
Fig 2 :Hyperion PPI
2.2. SMACC 1
A new endmember extraction method has been developed that
is based on a convex cone model for representing vector data.
The endmembers are selected directly from the data set. The
algorithm for finding the endmembers is sequential: the convex
cone model starts with a single endmember and increases
incrementally in dimension. Abundance maps are
simultaneously generated and updated at each step. A new
endmember is identified based on the angle that it makes with
the existing cone. The data vector which is making the
maximum angle with the existing cone is chosen as the next
- Sequential Maximum Angle Convex Cone