References
7
ship probability for each pixel. The complete ICAMM algorithm has been
explained in a simplified manner. Unsupervised classification of hyper spectral
data is performed using ICAMM and its performance evaluated vis a vis the
most widely used K-means algorithm.
Chapter 10 describes the application of SVM for supervised classification of
multi and hyperspectral data. Several issues, which may have a bearing on the
performance of SVM, have been considered. The effect of a number of kernel
functions, multi class methods and optimization techniques on the accuracy
and efficiency of classification has been assessed.
The two classification algorithms, unsupervised ICAMM and supervised
SVM, discussed in Chaps. 9 and 10 respectively, are regarded as per pixel
classifiers, as they allocate each pixel of an image to one class only. Often the
images are dominated by mixed pixels, which contain more than one class.
Since a mixed pixel displays a composite spectral response, which may be
dissimilar to each of its component classes, the pixel may not be allocated to
any of its component classes. Therefore, error is likely to occur in the classification of mixed pixels, if per pixel classification algorithms are used. Hence,
sub-pixel classification methods such as fuzzy c-means, linear mixture modeling and artificial neural networks have been proposed in the literature. In
Chap. 11, a novel method based on MRF models has been introduced for
sub-pixel mapping of hyperspectral data. The method is based on an optimization algorithm whereby raw coarse resolution images are first used to
generate an initial sub-pixel classification, which is then iteratively refined to
accurately characterize the spatial dependence between the class proportions
of the neighboring pixels. Thus, spatial relations within and between pixels
are considered throughout the process of generating the sub-pixel map. The
implementation of the complete algorithm is discussed and it is illustrated by
means of an example
The spatial dependency concept of MRF models is further extended to
change detection and image fusion applications in Chap. 12. Image change
detection is one of the basic image analysis tools and is frequently used in
many remote sensing applications to quantify temporal information, whereas
image fusion is primarily intended to improve, enhance and highlight certain
features of interest in remote sensing images for extracting useful information.
Individual MRF model based algorithms for these two applications have been
described and illustrated through experiments on multi and hyperspectral
data.
References
Campbell JB (2002) Introduction to remote sensing, 3rd edition. Guilford Press, New York
DeFries RS, Chan JC (2000) Multiple criteria for evaluating machine learning algorithms for
land cover classification from satellite data. Remote Sensing of Environment 74: 503-515
Jensen JR (I996) Introductory digital image processing: a remote sensing perspective, 2nd
edition. Prentice Hall, Upper Saddle River, N.J.
JPL, NASA. AVIRIS image cube. http://aviris.jpl.nasa.gov/html!aviris.cube.html
7
ship probability for each pixel. The complete ICAMM algorithm has been
explained in a simplified manner. Unsupervised classification of hyper spectral
data is performed using ICAMM and its performance evaluated vis a vis the
most widely used K-means algorithm.
Chapter 10 describes the application of SVM for supervised classification of
multi and hyperspectral data. Several issues, which may have a bearing on the
performance of SVM, have been considered. The effect of a number of kernel
functions, multi class methods and optimization techniques on the accuracy
and efficiency of classification has been assessed.
The two classification algorithms, unsupervised ICAMM and supervised
SVM, discussed in Chaps. 9 and 10 respectively, are regarded as per pixel
classifiers, as they allocate each pixel of an image to one class only. Often the
images are dominated by mixed pixels, which contain more than one class.
Since a mixed pixel displays a composite spectral response, which may be
dissimilar to each of its component classes, the pixel may not be allocated to
any of its component classes. Therefore, error is likely to occur in the classification of mixed pixels, if per pixel classification algorithms are used. Hence,
sub-pixel classification methods such as fuzzy c-means, linear mixture modeling and artificial neural networks have been proposed in the literature. In
Chap. 11, a novel method based on MRF models has been introduced for
sub-pixel mapping of hyperspectral data. The method is based on an optimization algorithm whereby raw coarse resolution images are first used to
generate an initial sub-pixel classification, which is then iteratively refined to
accurately characterize the spatial dependence between the class proportions
of the neighboring pixels. Thus, spatial relations within and between pixels
are considered throughout the process of generating the sub-pixel map. The
implementation of the complete algorithm is discussed and it is illustrated by
means of an example
The spatial dependency concept of MRF models is further extended to
change detection and image fusion applications in Chap. 12. Image change
detection is one of the basic image analysis tools and is frequently used in
many remote sensing applications to quantify temporal information, whereas
image fusion is primarily intended to improve, enhance and highlight certain
features of interest in remote sensing images for extracting useful information.
Individual MRF model based algorithms for these two applications have been
described and illustrated through experiments on multi and hyperspectral
data.
References
Campbell JB (2002) Introduction to remote sensing, 3rd edition. Guilford Press, New York
DeFries RS, Chan JC (2000) Multiple criteria for evaluating machine learning algorithms for
land cover classification from satellite data. Remote Sensing of Environment 74: 503-515
Jensen JR (I996) Introductory digital image processing: a remote sensing perspective, 2nd
edition. Prentice Hall, Upper Saddle River, N.J.
JPL, NASA. AVIRIS image cube. http://aviris.jpl.nasa.gov/html!aviris.cube.html