Page 50 - Digital Analysis of Remotely Sensed Imagery

Overview 23

training, both being critical issues to the success of neural network
image classification. The potential of this endeavor is evaluated
toward the end of the chapter. Chapter 9 on decision tree classification
begins with an introduction to major decision trees that have found
applications in image classification, followed by a discussion on how
to construct a tree. The potential of this classification method is
assessed toward the end of this chapter. The focus of Chap. 10 is on
spatial image classification in which the spatial relationship among
pixels is taken advantage of. Two topics, use of texture and object-
based image classification, are featured prominently in this chapter.
In addition, image segmentation, which is a vital preparatory step for
object-oriented image classification, is also covered extensively.
Recently, image classification has evolved to a level where external
knowledge has been incorporated into the decision making. How to
represent knowledge and incorporate it into image classification forms
the content of Chap. 11. After presenting various types of knowledge
that have found applications in intelligent image classification, this
chapter concentrates on how to acquire knowledge from various sources
and represent it. A case study is supplied to illustrate how knowledge
can be implemented in knowledge-based image classification and in
knowledge-based postclassification processing. The performance of
intelligent image classification relative to per-pixel classifiers is assessed
in terms of the classification accuracy achievable.
The next logical step of processing following image classification
is to provide a quality assurance. Assessment of the classification
results for their accuracy forms the content of Chap. 12. Addressed in
this chapter are sources of classification inaccuracy, procedure of
accuracy assessment, and proper reporting of accuracies. Chapter 13
extends digital analysis of remote sensing data to the multitemporal
domain, commonly known as change detection. The results derived
from respective remote sensing data are compared with each other
either spatially or nonspatially. Many issues related to change
detection are identified, in conjunction with innovative methods of
change detection. Suggestions are made about how to assess and
effectively visualize change detection results. The last chapter of this
book focuses on integrated image analysis with GIS and global
positioning system (GPS). After models of integrating these geo-
informatic technologies are presented, this chapter identifies the
barriers to full integration and potential areas to which the integrated
analysis approach may bring out the most benefits.