587 
The interface prompts the scientist to enter the required Lisp functions for executing the technique and the 
left hand side of the rule that causes this technique to be selected. Detailed instructions about the arguments 
of the functions, the values they should return, and the format of the rule are displayed. The Add Techniques 
interface enables the scientist to add techniques to VEG without assistance from the system developer. 
All the options in VEG make use of the historical cover type database. This database contains results 
from experiments by scientists on a wide variety of different cover types. Currently the database contains 
directional reflectance measurements over the entire hemisphere of a wide variety of cover types, sun angles, 
and wavelengths. The data included come from Kimes 1983, 1990, Kimes and Kirchner 1982, Kimes and 
Newcomb 1987, Kimes et al. 1982, 1985a,b, Deering 1989, Deering and Middleton 1990. 
When the learning system is in use, VEG selects positive and negative training examples from the histor 
ical cover type database. From these training examples, VEG determines relationships that discriminate 
between classes of vegetation. These classes can then be used to classify unknown target(s). The historical 
cover type database is a fundamental part of the learning system. 
All the other options in VEG use the historical cover type database in order to estimate the error terms 
when various analysis techniques are applied to the target. A subset of historical data, referred to as the 
"restricted data set," is selected for each session. VEG can automatically select the restricted data set that 
best matches the target Alternatively, the user can indicate the bounds on each parameter of interest and 
instruct VEG to select the subset of the historical cover type data that falls within the set bounds. Once the 
restricted data set has been selected, the reflectance data in each cover type are interpolated and extrapolated 
so that they match the exact view angles of the input spectral data. The restricted data set contains the true 
results for each cover type. Each technique that is applied to the target that is being studied is also applied 
to each cover type in the restricted data set An error for the restricted data set is calculated for each tech 
nique (Eqn. 1). This statistic provides an estimate of the error involved in applying the technique to the tar 
get being studied. 
The historical cover type database is stored as a series of flat files that are external to VEG. An inter 
face between VEG and these files has been provided. The interface allows the user to select which files of 
historical data to use. The files are then read, and the data are stored in KEE objects. The interface also 
allows the user to delete some or all of the historical database objects from VEG and load new historical 
data from a file. 
2.4. Learning System 
The Learning System "learns" class descriptions from samples (both positive and negative) of spectral, 
directional reflectance data of natural surfaces (bare soils, natural vegetation, and agricultural vegetation). 
The system is designed to handle any combination of directional view angles. The system finds class 
descriptions which contain the most important features that distinguish each class from the others. The expli 
cit relationships used in the class descriptions include greater-than relationships between combinations of two 
view angles and maximum and minimum value relationships. The class descriptions are used to classify an 
unknown target using the same directional views. The details of the algorithm are discussed by Kimes et al. 
(1992). 
The learning system can be operated in both VEG modes. In the "Research Mode," the learning system 
presents the user with three different options. In option 1, the system uses the database of historical cover 
type data to learn class descriptions of one or more classes of cover types. In option 2, the system learns 
class descriptions for one or more classes and then uses the learned classes to classify an unknown target by 
finding the class that best matches the unknown cover type data. Option 3 allows the user to test the 
system’s classification performance. In this option, the system learns class descriptions for one or more 
classes and then classifies the appropriate samples in the database. The percentage of correctly classified 
samples is then used to summarize the degree of classification accuracy achieved by the learning system. 
2.5. Testing VEG 
VEG has been tested for accuracy and used in a variety of studies for inferring spectral hemispherical 
reflectance, percent ground cover, plant height, and view angle extension. These studies are summarized in 
the Results and Discussion Section.