Here is a plain-text order form to order reports by mail. Some of the following reports can be downloaded as PostScript, or compressed PostScript files.
Randall P. Shumaker and Laura C. Davis
"Research in Advanced Software Technologies at the Naval Research Laboratory:
Machine Intelligence and Formal Methods," Proceedings of the 6th Annual
Software Technology Conference, Salt Lake City, UT, April 15, 1994,
(NCARAI Report: AIC-94-010).
Not available on-line at this time. Please see order form.
Abstract: The Department of Defense is critically dependent upon software, with software representing a very large and rapidly growing fraction of the defense budget. In many respects software engineering methods have improved greatly in the last decade, but demands on these methods to produce ever larger and more complex systems have outpaced these improvements. The Naval Research Laboratory (NRL) Information Technology Division conducts applied research in artificial intelligence and formal methods in software engineering aimed at demonstrating the applicability and effectiveness of advanced software methods to practical problems. Several NRL efforts have matured sufficiently to be distributed for application or further development. While some of these methods represent quite different approaches to software specification, validation and testing than conventional software engineering practice, they are designed to fit within existing frameworks. Each of the systems shares a developmental philosophy -- construct a reusable tool useful within, and compatible with, existing methods of software engineering. This paper describes several of these current efforts and concludes with a discussion of issues in technology transition drawn from our experiences in tool development.
Patrick R. Harrison and P. Ann Harrison
"Validating an Embedded Intelligent Sensor Control System," IEEE Expert, 49-53,
June 1994, (NCARAI Report: AIC-94-023).
Not available on-line at this time. Please see order form.
Abstract:
This paper develops a theoretical model for the design of intelligent,
real-time, sensor control systems. It then discusses system implementation
using this model in the context of what the authors call validation based
design. The design and implementation of an actual system using these
concepts for sensor control on a high speed jet aircraft is then described.
Finally, validation techniques and issues are discussed.
Abstract:
A knowledge-based system called VEG was expanded to infer nadir or
any off-nadir reflectance(s) of a vegetation target given any combination
of other directional reflectance(s) of target for a constant sun angle.
VEG determines the best technique(s) to use in an array of techniques,
applies the technique(s) to the target data, and provides rigorous
estimate of the accuracy of the inference(s). The knowledge-based system
VEG facilitates the use of diverse knowledge bases to be incorporated into
the inference techniques. In this study, VEG used additional techniques
that only use spectral data from the unknown target in a simplistic
manner. VEG used spectral data and a normalized difference technique to
infer the percent ground cover of the unknown target. This estimate of a
percent ground cover of the unknown target along with information on the
sun angle were then used to search a historical data base for targets that
match the unknown target in these characteristics. This data captured the
general shape of the reflectance distribution of the unknown target. This
historical information was used to estimate the coefficients of the
techniques for the condition at hand and to test the accuracy of the
techniques. The tests used in this study were different ones. For
example, techniques were tested that make long angular extensions using
one, two, or four input view angles to predict an unknown nadir value.
Furthermore, a wide variety of unknown targets were tested. The errors
(+/-proportional rms) obtained were on the order of 0.15. In addition
techniques were tested that use seven or nine multiple view angles to
predict the entire hemispherical reflectance distribution of an unknown
target. The accuracy of these tests was relatively good considering the
relatively dynamic and noisy nature of directional reflectance
distributions and the amount of historical data available that closely
matches the unknown target.
Abstract:
An Intelligent Workbench (VEG) has been developed for the
systematic study of remotely sensed optical data from vegetation. A goal
of the remote sensing community is to infer physical and biological
properties of vegetation cover (e.g. cover type, hemispherical
reflectance, ground cover, leaf area index, biomass and photosynthetic
capacity) using directional spectral data. Numerous techniques that infer
some of these vegetation properties have been published in the literature.
A fundamental problem is deciding which technique to apply to the data and
then estimating the error bounds on the results. Studies have found that
using conventional techniques produced errors as high as 45%.
VEG collects together in a common format technique previously
available data from many different sources in a variety of formats. The
decision as to when a particular technique should be applied is
nonalgorithmic and requires expert knowledge. VEG has codified this expert
knowledge into a rule-based decision component for determining which
technique to use. VEG provides a comprehensive interface that makes
applying the techniques simple and aids a researcher in developing and
testing new techniques. VEG also allows the scientist to incorporate
historical databases into problem solving. The scientist can match the
target data being studied with historical data so the historical data can
be used to provide the coefficients needed for applying analysis
techniques. The historical data also provides more accurate error estimates
than were previously available. VEG also enables the scientist to try
"what-if" experiments on data using a variety of different techniques and
historical data sets to do comparative studies or test experimental
hypotheses.
VEG also provides a classification algorithm that can learn new
classes of surface features. The learning system uses the database of
historical cover types to learn class descriptions of one or more classes
of cover types. These classes can include broad classes such as soil or
vegetation or more specific classes such as forest, grass, and wheat. The
classes can also include subclasses based on continuous parameters, e.g.
0-30% ground cover. The learning system uses sets of positive and negative
examples from the historical databases to find the most important features
that uniquely distinguish each class. The system then uses the learned
classes to classify an unknown sample by finding the class that best
matches the unknown cover type data. The learning systems also include an
option that allows the user to test the system's classification
performance.
VEG was developed using object oriented programming, and the
current version consists of over 1500 objects.
Abstract:
This article reports on the extension of the VEG remote sensing
analysis workbench to include the capability to reason about nadir or any
off-nadir reflectance(s) of a vegetation target given any combination of
other directional reflectance(s) of the target. VEG determines the best
techniques in an array of techniques, applies the techniques to the target
data, and provides a rigorous estimate of the accuracy of the inferences.
In this study, VEG used additional information to make more accurate view
angle extension techniques than traditional techniques that only use
spectral data from the unknown target. VEG used spectral data and a
normalized difference technique to infer the percent ground cover of an
unknown target. This estimate of percent ground cover of the unknown
target along with information on the sun angle were then used to search a
historical data base for targets that matched the unknown target in these
characteristics. This historical information was used to estimate the
coefficients of the techniques for existing conditions. Extensive testing
was done to test the accuracy of the new techniques. The results were very
positive for a wide range of problems and are detailed in the report.
Abstract: The paper provides a practical discussion of an overall design concept for real-time intelligent sensor control systems. The discussion focuses on one component of this design concept. This component is an embedded, intelligent, decision aid called the Advice Generator. The application is the sensor control system on the F-14D aircraft. Intelligence is introduced into the sensor management system of the aircraft within a control context. The essence of the design separates control of induction from control of deduction in the sensor management system. The design allows the expert system component to do what expert systems do best, deduce, while overall control of induction is managed by the human (RIO) in the loop. The paper also discusses validation of the Advice Generator component in the context of the design model.
Intelligent M4 Systems
Interface Design And Evaluation
Machine Learning
Sensor-Based Systems
