ENDMEMBER EXTRACTION FOR HYPERSPECTRAL IMAGE
BASED ON INTEGRATION OF SPATIAL-SPECTRAL INFORMATION
Xiang-bing Kong
1,2
, Zui Tao
2
, Er Yang
1
, Zhihui Wang
1
, Chunxia Yang
1
1. Key Laboratory of the Loess Plateau Soil Erosion and Water Loss Process and Control, Ministry of
Water Resources, Yellow River Institute of Hydraulic Research, Zhengzhou Henan 450003, China
2. State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing and Digital Earth,
Chinese Academy of Sciences, Beijing 100101, China
ABSTRACT
Endmember extraction (EE) plays an extremely important
role for hyperspectral mixture analysis, and many EE
methods have been proposed in recent years. However, most
approaches have been designed from a spectroscopic
viewpoint and thus, tend to neglect the existing spatial
correlation between pixels. In this paper, a novel algorithm
is proposed to integrate both spatial and spectral information
for automatic EE (ISEE). At first, the image is divided into
some subspaces for improvement of spectral contrast. Then,
the subset of the image is projected to the feature space
related to the image endmembers, and the candidate
endmember spectra are extracted through orthogonal
subspace projection analysis. At last, the endmember spectra
are refined under the constraint of image spatial context and
spectral information. The performance of different
endmember extraction methods is compared using both
synthetic hyperspectral image and real hyperspectral image.
The experimental results demonstrate that ISEE
incorporated with spatial information is effective, and the
endmember spectra extracted by ISEE is more accurate than
by some common EE methods.
Index Terms—Hyperspectral remote sensing,
Endmember extraction, Orthogonal subspace projection,
Spatial information
1. INTRODUCTION
Hyperspectral imagery is acquired by high-spectral-
resolution imaging sensors, containing hundreds of
contiguous narrow spectral band images, with lots of
valuable spectral, spatial and radiation information.
However, due to the distribution complexity of the materials
and the low spatial resolution of the sensors, there are lots of
mixed pixels in hyperspectral imagery, which is the biggest
obstacle of quantitative analysis and application of
hyperspectral data. The linear mixture model (LMM) is well
known for its simple structure and clear physical meaning
and endmember extraction (EE) plays an extremely
important role for hyperspectral mixture analysis.
Some automatic or semi-automatic endmember
extraction methods have been proposed, including Pure
Pixel Index(PPI)
[1]
, Iterative Error Analysis (IEA)
[2]
,
Vertex Component Analysis (VCA)
[3]
and Orthogonal
Subspace Projection (OSP)
[4]
and some new endmember
extraction methods based on specific mathematical theory
[5-
6]
. However, most approaches have been designed from a
spectroscopic viewpoint and thus, tend to neglect the
existing spatial correlation between pixels. Endmember
extraction based on integration of spatial and spectral
information has become an important research direction in
recent years, and several approaches have been developed
[7-
10]
.
This paper proposes a novel algorithm integrating both
spatial and spectral information for automatic EE (ISEE).
Comparing with the traditional end-member extraction
method
[1-4]
using only spectral information, the spatial
information plays an important role in the EE process of
ISEE. Unlike some typical endmember extraction methods
[7-
10]
, ISEE can give the final hyperspectral endmemers with
high efficiency through feature space projection analysis.
2. ISEE
There are two issues for typical endmember extraction
methods. First, it is difficult to reflect local hyperspectral
image information for the final endmember set extracted
through the entire image. Second, the effectiveness of
enmember set using only spectral information is easy
affected by the noise and spectral mixture in hyperspectral
image. Based on the analysis of the relevant issues, a
combination of spatial and spectral information in
hyperspectral image endmember automatic extraction
methods (ISEE) is proposed, containing four key steps, i.e.
image subspace dividing, feature space projection analysis,
endmember spectrum optimization with spatial and spectral
information constraint.
2.1. Image subspace dividing