193.

[Cover] [Contents] [Index]

Page 271

Chapter 7
Multisource classification

Earth observing satellites currently in operation or planned for the near future carry sensors that operate in the visible, infrared and microwave regions of the spectrum. In addition, the widespread availability of GIS means that digital spatial data have become more accessible. Hence, greater attention is now being given to the use of multisource data in remote sensing image classification. Such data may consist of images produced by different sensor systems, or digital spatial data. The assumption is made that classification accuracy should improve if additional features are incorporated. Such features may be derived from the image data set itself, or from two or more co-registered image sets from different sensors, or from associated geographical information such as surface elevation, soil type, or drainage pattern. This assumption is generally reasonable, because the greater the amount of relevant information that is included in a classification the greater the probability that interclass confusion will be reduced. Thus, it can be foreseen that the development of multisource classification methodologies will become increasingly important. It should, of course, also be remembered that the use of highly correlated features, or features that show no variation between the classes of interest can obscure rather than illuminate the problem. One should always bear in mind that weighting is generally applied to input features. If features are standardised to a 0−1 range, for example, then variation in feature a is considered to be equivalent to variation in feature b. Noting that variation can be considered to be equivalent to information, it is clearly not helpful to include ‘surface elevation’ as a feature if it varies only slightly over the region of interest, and if these small variations are not related to the spatial boundaries of the classes.

A further consideration is the varying reliability and completeness of different data sources. It is necessary, therefore, to take the reliability or uncertainty of each data source into account when classification of multisource data is attempted. Source reliability factors are important parameters that determine how strongly a given source contributes to the multisource consensus pool. In this sense, reliability factors are equivalent

[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