3 Processing Remotely Sensed Data: Hardware and Software Considerations
53
bit). The analyst can then density slice, magnify, roam, or manipulate the contrast
of the individual bands or color composites. Through the years, algebraic and
linear combinations of bands of remote sensor data have proven useful for hydrologic research. Therefore, the system must be able to build simple algebraic statements (e.g. ratio bands 4/5) or linear combinations of bands (e.g. the Kauth transform) to produce more sophisticated vegetation and hydrologic transformations of
the remote sensor data.
Spatial and frequency filtering algorithms can be used to enhance and display
subtle high and low frequency features and edges of these features in the remote
sensor data. Texture algorithms enhance areas of uniform texture (e.g. coarse,
smooth, rough). Some bands of remote sensor data are highly correlated with other
bands, therefore, there is redundant information. Principal components analysis is
often applied to reduce the dimensionality (number of redundant bands) used in
the analysis while still maintaining the critical essence of the data.
Digital elevation models (DEMs) are critical to successful modeling and understanding of many landscape processes. The analyst must be able to display aDEM
in a planimetric (vertical) view using analytical hill shading as well as in a pseudo
3-dimensional perspective view. Ideally, it is possible to drape thematic information such as a hydrologic network on top of the hill-shaded DEM. The DEMs and
orthophotos are produced using photogrammetric principles as discussed in the
soft-copy photogrammetry information extraction section.
Finally, it is important to be able to monitor change in the landscape by displaying multiple dates of imagery in an animated fashion. Change information can
be used to gather information about the processes at work.
3.2.3 Remote Sensing Information Extraction
The analyst must be able to read the brightness values (z) at any X,y location in the
image and along user-specified transects. The user must also be able to draw polygons around objects of interest on the screen and extract fundamental area, perimeter, and even volume information. Ideally, the polygon and its attribute information may be saved in a standard format for subsequent processing, e.g. in
ERDAS or ArcInfo coverage formats. Heads-up on-screen image interpretation
and digitization is becoming more important as very high spatial resolution satellite remote sensor data becomes available and fundamental photo-interpretation
techniques are merged with automated feature extraction (Jensen, 1995a; Firestone
et aI., 1996).
Reputable digital image processing systems allow the user to classify multispectral remote sensor data using supervised or unsupervised classification techniques. In a supervised classification the analyst supervises the training of the
algorithm. In an unsupervised classification the analyst relies on the computer to
identify pixels in the terrain that have approximately the same multispectral characteristics. These 'clusters' are then labeled by the analyst to produce a thematic
map. Most image processing systems still do not easily allow the incorporation of
contextual or other types of ancillary data in the classification. Furthermore, only a
few systems allow the user to apply expert system, neural network, or fuzzy clas-
53
bit). The analyst can then density slice, magnify, roam, or manipulate the contrast
of the individual bands or color composites. Through the years, algebraic and
linear combinations of bands of remote sensor data have proven useful for hydrologic research. Therefore, the system must be able to build simple algebraic statements (e.g. ratio bands 4/5) or linear combinations of bands (e.g. the Kauth transform) to produce more sophisticated vegetation and hydrologic transformations of
the remote sensor data.
Spatial and frequency filtering algorithms can be used to enhance and display
subtle high and low frequency features and edges of these features in the remote
sensor data. Texture algorithms enhance areas of uniform texture (e.g. coarse,
smooth, rough). Some bands of remote sensor data are highly correlated with other
bands, therefore, there is redundant information. Principal components analysis is
often applied to reduce the dimensionality (number of redundant bands) used in
the analysis while still maintaining the critical essence of the data.
Digital elevation models (DEMs) are critical to successful modeling and understanding of many landscape processes. The analyst must be able to display aDEM
in a planimetric (vertical) view using analytical hill shading as well as in a pseudo
3-dimensional perspective view. Ideally, it is possible to drape thematic information such as a hydrologic network on top of the hill-shaded DEM. The DEMs and
orthophotos are produced using photogrammetric principles as discussed in the
soft-copy photogrammetry information extraction section.
Finally, it is important to be able to monitor change in the landscape by displaying multiple dates of imagery in an animated fashion. Change information can
be used to gather information about the processes at work.
3.2.3 Remote Sensing Information Extraction
The analyst must be able to read the brightness values (z) at any X,y location in the
image and along user-specified transects. The user must also be able to draw polygons around objects of interest on the screen and extract fundamental area, perimeter, and even volume information. Ideally, the polygon and its attribute information may be saved in a standard format for subsequent processing, e.g. in
ERDAS or ArcInfo coverage formats. Heads-up on-screen image interpretation
and digitization is becoming more important as very high spatial resolution satellite remote sensor data becomes available and fundamental photo-interpretation
techniques are merged with automated feature extraction (Jensen, 1995a; Firestone
et aI., 1996).
Reputable digital image processing systems allow the user to classify multispectral remote sensor data using supervised or unsupervised classification techniques. In a supervised classification the analyst supervises the training of the
algorithm. In an unsupervised classification the analyst relies on the computer to
identify pixels in the terrain that have approximately the same multispectral characteristics. These 'clusters' are then labeled by the analyst to produce a thematic
map. Most image processing systems still do not easily allow the incorporation of
contextual or other types of ancillary data in the classification. Furthermore, only a
few systems allow the user to apply expert system, neural network, or fuzzy clas-