imagery, various hyperspectral vegetation indices have been proposed for studying
specific properties of different targets covering Earth’s surface, including soil, water,
and vegetation. However, this topic is not covered herein as it can form the topic of
another chapter by itself and lies outside the objectives of this chapter. Interested
readers are referred to a detailed recent review of hyperspectral indices as regards
particular vegetation properties by Thenkabail et al. (2000).
15.3.2 Land Cover Mapping from Hyperspectral Remote Sensing Imagery
The ability of hyperspectral sensors to better discriminate ground cover classes than
traditional multispectral sensors has been demonstrated elsewhere (Zhang and Ma,
2009). Nowadays, hyperspectral remote sensing imagery is regarded as one of the
most significant Earth observation data sources (Du et al., 2010) and is being used for
various applications, including land cover classification (Li and Liu, 2010). Producing
land use/cover mapping thematic maps using hyperspectral remote sensing data is
commonly performed by digital image classification (Chintan et al., 2004). A
comprehensive review of the variety of classification approaches applied to remote
sensing data, including hyperspectral sensors, was made available recently by Lu and
Weng (2007). Generally, a widely used categorization of classification techniques
includes three main groups of approaches: pixel-, subpixel-, and object-based
classification. These are briefly reviewed next.
15.3.2.1 Pixel-Based Techniques Techniques that belong to this group perform
classification by assigning pixels to land cover classes using either supervised or
unsupervised classifiers. Unsupervised classifiers group pixels with similar spectral
values into unique clusters according to some statistically predefined criteria where
the classifier combines and reassigns the spectral clusters into information classes.
Supervised classifiers use samples of known identity for each land cover class, known
as “training sites,” to classify image pixels of unknown identity (Campbell, 1996).
Supervised classifiers are also commonly divided into parametric and nonparametric.
In comparison to the nonparametric [such as artificial neural networks (ANNs) or
support vector machines (SVMs) (Vapnik, 1995)], parametric pixel-based classifiers
[e.g., the maximum-likelihood (ML)] (Harris, 1998) require prior knowledge/assumptions regarding the statistical distribution of the data to be classified for the different
classes used, information often difficult to attain in practice. A pixel-based classification technique that has gained popularity, particularly when implemented with
hyperspectral imagery, is the spectral angle mapper (SAM) (Kruse et al., 1993). This
is a supervised classification technique based on the computation of spectral angle
similarity between a reference source and the target spectra. The popularity of SAM is
due to its simplicity and rapid implementation for the examination of the spectral
similarity of image spectra to reference spectra. It is also a very powerful classification
method because it suppresses the influence of shading effects to accentuate the target
reflectance characteristics (De Carvalho and Meneses, 2000). Despite the potential of
pixel-based methods, some of these techniques require making assumptions regarding
HYPERSPECTRAL REMOTE SENSING IN LAND COVER EXTRACTION
305
specific properties of different targets covering Earth’s surface, including soil, water,
and vegetation. However, this topic is not covered herein as it can form the topic of
another chapter by itself and lies outside the objectives of this chapter. Interested
readers are referred to a detailed recent review of hyperspectral indices as regards
particular vegetation properties by Thenkabail et al. (2000).
15.3.2 Land Cover Mapping from Hyperspectral Remote Sensing Imagery
The ability of hyperspectral sensors to better discriminate ground cover classes than
traditional multispectral sensors has been demonstrated elsewhere (Zhang and Ma,
2009). Nowadays, hyperspectral remote sensing imagery is regarded as one of the
most significant Earth observation data sources (Du et al., 2010) and is being used for
various applications, including land cover classification (Li and Liu, 2010). Producing
land use/cover mapping thematic maps using hyperspectral remote sensing data is
commonly performed by digital image classification (Chintan et al., 2004). A
comprehensive review of the variety of classification approaches applied to remote
sensing data, including hyperspectral sensors, was made available recently by Lu and
Weng (2007). Generally, a widely used categorization of classification techniques
includes three main groups of approaches: pixel-, subpixel-, and object-based
classification. These are briefly reviewed next.
15.3.2.1 Pixel-Based Techniques Techniques that belong to this group perform
classification by assigning pixels to land cover classes using either supervised or
unsupervised classifiers. Unsupervised classifiers group pixels with similar spectral
values into unique clusters according to some statistically predefined criteria where
the classifier combines and reassigns the spectral clusters into information classes.
Supervised classifiers use samples of known identity for each land cover class, known
as “training sites,” to classify image pixels of unknown identity (Campbell, 1996).
Supervised classifiers are also commonly divided into parametric and nonparametric.
In comparison to the nonparametric [such as artificial neural networks (ANNs) or
support vector machines (SVMs) (Vapnik, 1995)], parametric pixel-based classifiers
[e.g., the maximum-likelihood (ML)] (Harris, 1998) require prior knowledge/assumptions regarding the statistical distribution of the data to be classified for the different
classes used, information often difficult to attain in practice. A pixel-based classification technique that has gained popularity, particularly when implemented with
hyperspectral imagery, is the spectral angle mapper (SAM) (Kruse et al., 1993). This
is a supervised classification technique based on the computation of spectral angle
similarity between a reference source and the target spectra. The popularity of SAM is
due to its simplicity and rapid implementation for the examination of the spectral
similarity of image spectra to reference spectra. It is also a very powerful classification
method because it suppresses the influence of shading effects to accentuate the target
reflectance characteristics (De Carvalho and Meneses, 2000). Despite the potential of
pixel-based methods, some of these techniques require making assumptions regarding
HYPERSPECTRAL REMOTE SENSING IN LAND COVER EXTRACTION
305