coniferous f.
deciduous f.
mixed f. ^
permanent cultures'^"
seasonal cultures
residential area
industrial area
An object-oriented formalism has been used to design the
knowledge base of the RESEDA expert system. The
central component of the RESEDA knowledge base is a
taxonomy of remote sensing targets, as indicated in figure
2. The nodes in this hierarchical graph represent target
classes, whereas the annotations of these classes (shown
in italics) represent target attributes. The target taxono
my is static; that is, it is independent of individual cases
or analyses. The target taxonomy describes all the ab
stract features with which a remote sensor data analysis
may deal. Although the target taxonomy is static, it is
nonetheless extendable and may be continuously updated
by a knowledge engineer during a knowledge acquisition
dialog. Currently, this hierarchy is tailored for some
purposes of environmental management, but it may be
easily adapted to cover other cases.
The purpose of this taxonomy is twofold. On the one
hand, this hierarchy is used by the RESEDA Assistant
system to generate a top-level menu, on which the user
may indicate the information of interest. On the other
hand, the hierarchy serves as the most general represen
tation of the microworld that can be handled by remote
sensing methods and, as such, is used by the inference
mechanism of the expert system.
Target Classes
Target classes are static knowledge base items describing
case-independent properties of a certain class of land
coverage, such as the following:
- quantitative or qualitative characteristics of reflec
tance behavior;
- expected phenological changes over the year (e.g., of
seasonal cultures);
- stability of the land coverage over time (e.g., unde-
finite time for forest or residential areas, 1 year or
less for most agricultural areas);
- probability of land-use changes (e.g., caused by crop
rotations) (Janssen, 1990).
Each target class may possess crosslinks to related know
ledge-base items, such as the target attributes defined for
the class, or to processing models for recognizing the
class (so-called classification models). Since the target
classes are organized in a generalization hierarchy, they
inherit the descriptions from their parent nodes.
Target Attributes
Target attributes stand for abstract properties of geogra
phic locations (not for their concrete values). Every target
attribute is defined for all members of a particular target
class, including all of its subclasses. Examples of target
attributes are the following: