2
Pramod K. Varshney, Manoj K. Arora
for information extraction from hyperspectral, multi-source, multi-sensor and
multi-temporal data sets.
A number of textbooks on conventional digital processing of remote sensing data are available to the academic community (e.g. Jensen 1996; Richards
and Jia 1999; Mather 1999; Campbell 2002). These deal with the basic digital
image processing techniques together with fundamental principles of remote
sensing technology. They are excellent resources for undergraduate and beginning graduate-level teaching. However, the researchers and graduate students
working on the development and application of advanced techniques for remote sensing image processing have to depend on texts available in electrical
engineering, and computer science literature, which do not focus on remote
sensing data. This book contains chapters written by authors with backgrounds
in different disciplines and thus attempts to bridge this gap. It addresses the
recent developments in the area of image registration, fusion and change detection, feature extraction, classification of remote sensing data and accuracy
assessment. The aim of the book is to introduce the reader to a new generation
of information extraction techniques for various image processing operations
for multispectral and particularly hyperspectral data. We expect that the book
will form a useful text for graduate students and researchers taking advanced
courses in remote sensing and GIS, image processing and pattern recognition
areas. The book will also prove to be of interest to the professional remote
sensing data users such as geologists, hydrologists, ecologists, environmental
scientists, civil and electrical engineers and computer scientists. In addition to
remote sensing, the algorithms presented will have far-reaching applicability
in fields such as signal processing and medical imaging.
What is Hyperspectrallmaging?
Since the initial acquisition of satellite images, remote sensing technology has
not looked back. A number of earth satellites have been launched to advance our
understanding of Earth's environment. The satellite sensors, both active and
passive, capture data from visible to microwave regions of the electromagnetic
spectrum. The multispectral sensors gather data in a small number of bands
(also called features) with broad wavelength intervals. No doubt, multispectral
sensors are innovative. However, due to relatively few spectral bands, their
spectral resolution is insufficient for many precise earth surface studies.
When spectral measurement is performed using hundreds of narrow contiguous wavelength intervals, the resulting image is called a hyperspectral
image, which is often represented as a hyperspectral image cube (see Fig. 1)
(JPL, NASA). In this cube, the x and y axes specify the size of the images,
whereas the z axis denotes the number of bands in the hyperspectral data.
An almost continuous spectrum can be generated for a pixel and hence hyperspectral imaging is also referred to as imaging spectrometry. The detailed
spectral response of a pixel assists in providing accurate and precise extraction
of information than is obtained from multispectral imaging. Reduction in the
cost of sensors as well as advances in data storage and transmission technolo-
Pramod K. Varshney, Manoj K. Arora
for information extraction from hyperspectral, multi-source, multi-sensor and
multi-temporal data sets.
A number of textbooks on conventional digital processing of remote sensing data are available to the academic community (e.g. Jensen 1996; Richards
and Jia 1999; Mather 1999; Campbell 2002). These deal with the basic digital
image processing techniques together with fundamental principles of remote
sensing technology. They are excellent resources for undergraduate and beginning graduate-level teaching. However, the researchers and graduate students
working on the development and application of advanced techniques for remote sensing image processing have to depend on texts available in electrical
engineering, and computer science literature, which do not focus on remote
sensing data. This book contains chapters written by authors with backgrounds
in different disciplines and thus attempts to bridge this gap. It addresses the
recent developments in the area of image registration, fusion and change detection, feature extraction, classification of remote sensing data and accuracy
assessment. The aim of the book is to introduce the reader to a new generation
of information extraction techniques for various image processing operations
for multispectral and particularly hyperspectral data. We expect that the book
will form a useful text for graduate students and researchers taking advanced
courses in remote sensing and GIS, image processing and pattern recognition
areas. The book will also prove to be of interest to the professional remote
sensing data users such as geologists, hydrologists, ecologists, environmental
scientists, civil and electrical engineers and computer scientists. In addition to
remote sensing, the algorithms presented will have far-reaching applicability
in fields such as signal processing and medical imaging.
What is Hyperspectrallmaging?
Since the initial acquisition of satellite images, remote sensing technology has
not looked back. A number of earth satellites have been launched to advance our
understanding of Earth's environment. The satellite sensors, both active and
passive, capture data from visible to microwave regions of the electromagnetic
spectrum. The multispectral sensors gather data in a small number of bands
(also called features) with broad wavelength intervals. No doubt, multispectral
sensors are innovative. However, due to relatively few spectral bands, their
spectral resolution is insufficient for many precise earth surface studies.
When spectral measurement is performed using hundreds of narrow contiguous wavelength intervals, the resulting image is called a hyperspectral
image, which is often represented as a hyperspectral image cube (see Fig. 1)
(JPL, NASA). In this cube, the x and y axes specify the size of the images,
whereas the z axis denotes the number of bands in the hyperspectral data.
An almost continuous spectrum can be generated for a pixel and hence hyperspectral imaging is also referred to as imaging spectrometry. The detailed
spectral response of a pixel assists in providing accurate and precise extraction
of information than is obtained from multispectral imaging. Reduction in the
cost of sensors as well as advances in data storage and transmission technolo-