314.

[Cover] [Contents] [Index]

Page 81

reducing computation times and achieving higher classification accuracy. A comprehensive discussion of these topics is presented in Chapter 3. Recent interest in neural network construction is also driven by understanding the human-like performance in the area of speech and pattern recognition. However, ANN are perceived to be difficult to apply successfully, and questions such as the type of network, the network architecture, the initial values of parameters such as learning rate and momentum, the number of iterations required to train the network, and the choice of initial weights are all difficult to answer. Chapter 3 provides some guidance on these and related questions.

2.3.6 Knowledge-based methods

Knowledge-based methods represent an attempt to develop automated methods of pattern recognition by simulating the brain’s inference mechanism. Knowledge can be heuristic, i.e. based on experience and reasoning, and may not necessarily involve a statistical model. For example, knowledge of multipolarisation radar backscatter coefficients for a range of cover types such as water, bare rock, crops and forest could be used to generate an inference system to perform classification (Pierce et al., 1994). Two categories of knowledge-based classification methods, based on hierarchical decision-tree methods and automatic fuzzy rules extraction, are described here. The hierarchical decision-tree method is the most general type of knowledge-based classifier, while automatic fuzzy rule extraction is a newer technique that shows some interest and promise.

2.3.6.1 Decision trees

Categorisation of data using a hierarchical splitting (or ‘top-down’) mechanism has been widely used in the environmental and life sciences. The purpose of using a hierarchical structure for labelling objects is to gain a more comprehensive understanding of relationships between objects at different scales of observation or at different levels of detail. Its simplest representation takes the form of an inverted tree in which the different levels of classification are represented by the separate levels of the hierarchy. When applied to multispectral data, the design of a decision tree is based on knowledge of the spectral properties of each class and of the relationships between the classes.

A hierarchical decision tree classifier is characterised by the fact that an unknown pattern is labelled using a sequence of decisions. A decision tree is composed of a root node, a set of interior nodes, and terminal nodes, called ‘leaves’. The root node and interior nodes, referred to collectively as non-terminal nodes, are linked into decision stages. The terminal nodes represent final classifications, which provide the most detail in terms of the

[Cover] [Contents] [Index]


Classification Methods for Remotely Sensed Data
Classification Methods for Remotely Sensed Data, Second Edition
ISBN: 1420090720
EAN: 2147483647
Year: 2001
Pages: 354

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net