The output, yj, is a row vector of length D, where each element indicates the confidence that the input data from the multiple sensor set has membership in a particular class.. The Bayes
Trang 1The output, y(j), is a row vector of length D, where each element indicates the confidence that the input data from the multiple sensor set has membership in a particular class At time k, the output decision, d(k), is the class that satisfies the maximum confidence criteria of Equation 23.4
(23.4)
This implementation of weighted decision fusion permits future extension in two ways First, it provides a path to the use of confidence as an input from each sensor This would allow the fusion process
to utilize fuzzy logic within the structure Second, it enables an adaptive mechanism to be incorporated that can modify the sensor weights as data are processed through the system
23.3.2.3 Bayesian Inference
Bayes’ theorem16-18 serves as the basis for the Bayesian inference technique for identity fusion This technique provides a method for computing the a posteriori probability of a particular outcome, based
on previous estimates of the likelihood and additional evidence Bayesian inference assumes that a set of
D mutually exclusive (and exhaustive) hypotheses or outcomes exists to explain a given situation
In the decision-level, multisensor fusion problem, Bayesian inference is implemented as follows A system exists with N sensors that provide decisions on membership to one of D possible classes The Bayesian fusion structure uses a priori information on the probability that a particular hypothesis exists and the likelihood that a particular sensor is able to classify the data to the correct hypothesis The inputs to the structure are (1) P(O j), the a priori probabilities that object j exists (or equivalently that a fault condition exists), (2)
P(D k,i |O j), the likelihood that each sensor, k, will classify the data as belonging to any one of the D hypotheses, and (3) D k, the input decisions from the K sensors Equation 23.5 describes the Bayesian combination rule
(23.5)
The output is a vector with element j representing the a posteriori probability that the data belong to hypothesis j The fused decision is made based on the maximum a posteriori probability criteria given in Equation 23.6
(23.6)
A basic issue with the use of Bayesian inference techniques involves the selection of the a priori
probabilities and the likelihood values The choice of this information has a significant impact on performance Expert knowledge can be used to determine these inputs In the case where the a priori
probabilities are unknown, the user can resort to the principle of indifference, where the prior probabil-ities are set to be equal, as in Equation 23.7
(23.7)
The a priori probabilities are updated in the recursive implementation as described by Equation 23.8 This update sets the value for the a priori probability in iteration t equal to the value of the a posteriori
probability from iteration (t – 1)
w i i
N
=
=
1
j
j k
K
k j
j k
K
k j i
N K
1 1
( )= ( )
=
=
∏
∏
∑
K
( )= arg max [ ( 1,…, ) ]
P O
N
j
( )= 1
Trang 2Information Technology
for NASA in the 21st Century 24.1 Introduction
24.2 NASA Applications
Exploration • Earth Observation • Air Traffic
24.3 Critical Research Investment Areas for NASA
Automated Reasoning • Intelligent Data
24.4 High-Performance Computing and Networking
24.5 Conclusions
24.1 Introduction
The future of NASA is critically dependent on the development and implementation of new tools and methods from the information technology research community A few examples are worth noting The sophisticated unmanned exploration of Mars and other parts of the solar system, which will be aimed
at answering fundamental science questions, such as the existence of early life forms in these environ-ments, will require a new generation of automated reasoning tools In addition, NASA’s role in the development of new air traffic management tools and methods to be evaluated and deployed by the Federal Aviation Administration must involve new approaches to optimizing the combined performance
of experts on the ground (air traffic controllers) and in the air (pilots) and the supporting information systems Ongoing safe operation of the Space Shuttle depends on new capabilities for early identification
of the precursors to failure of safety-critical system components from maintenance data and sensors distributed throughout the system Use of the mountains of data generated by the Earth-observing satellites and next-generation space telescopes fielded by NASA demands fundamentally new methods
of data interpretation and understanding In addition, new aircraft and spacecraft designs depend on new high performance computing capabilities
The complexity and diversity of such critical needs for the future has motivated NASA to develop an expanded information technology (IT) research and development (R&D) portfolio The first step toward this end was an extensive strategic planning process for Computer Science/Information Technology R&D for the Agency Beginning in 1996, NASA’s Ames Research Center, located in the heart of Silicon Valley, assembled teams from the research community and the user communities served by the Agency to address two fundamental questions First, what are those NASA applications domains for which revolutionary advances in information technology are the critical enabler for the future? Second, in light of the
Robert J Hansen
University of West Florida
Daniel Cooke
NASA Ames Research Center
Kenneth Ford
Institute for Human and Machine Cognition
Steven Zornetzer
NASA Ames Research Center
Trang 3Data Fusion for a Distributed Ground-Based Sensing System 25.1 Introduction
25.2 Problem Domain
25.3 Existing Systems
25.4 Prototype Sensors for SenseIT
25.5 Software Architecture
25.6 Declarative Language Front-End
25.7 Subscriptions
25.8 Mobile Code
25.9 Diffusion Network Routing
25.10 Collaborative Signal Processing
25.11 Information Security
25.12 Summary
Acknowledgments and Disclaimers
References
25.1 Introduction
Sensor Information Technology (SenseIT) is a program of the U S Department of Defense Advanced Research Projects Agency (DARPA) that began in fiscal year 1999 A number of research groups are exploring networking and organizational problems posed by large networks of unmanned, intelligent, battery-powered, wireless sensors
The use of distributed sensor networks is of great interest to the Department of Defense (DoD) The work performed by this project is applicable to a number of heterogeneous unmanned sensing platforms The deployment of large numbers of sensing nodes poses several new technical problems Sensor networks resemble other wireless networks (such as cellular telephones) but have unique aspects, which are introduced in this chapter
Several applications exist for this technology within the DoD, the most obvious being in the realm of intelligence, surveillance, and reconnaissance (ISR): automating deployment of heterogeneous sensor networks and interpretation of their readings SenseIT could protect troops in the field as effectively as land mines, without many undesirable side effects Civilian applications include law enforcement, agri-culture, traffic monitoring, security, and environmental monitoring
SenseIT provides a new perspective to programming embedded systems It considers revolutionary approaches to coordination of multiple computers with power, time, bandwidth, and storage constraints
A chaotic network of small, inexpensive, unreliable systems form a large, powerful, dependable system
Richard R Brooks
The Pennsylvania State University
Trang 4An Evaluation Methodology for Fusion Processes Based
on Information Needs 26.1 Introduction
26.2 Information Needs
Database Analysis
26.3 Key Concept
26.4 Evaluation Methodology
References
26.1 Introduction
Fusion is a part of a larger Department of Defense (DoD) context — command, control, communication, computers, intelligence, surveillance and reconnaissance (C4ISR) C4ISR capabilities are enablers for the even larger context of information superiority A key question that is asked at decision-making levels of the DoD
is how C4ISR supports the military commander in the efficient execution of military operations The question
is germane at budget levels where, for example, C4ISR competes with weapons platforms for funding The Joint C4ISR Decision Support Center (DSC)* in DoD has performed numerous studies to deter-mine the value of C4ISR in general and for fusion in particular The DSC view is that value does not refer
to measures of the technical merits of alternative ISR approaches; it refers instead to the value of C4ISR
to support military command and control (C2) Increasingly, the C2 process has become a near-real-time decision based on perceived information Obviously better ISR improves the data and information, but can it do so in a timely manner with high confidence? And what is the value of that information? Fusion of information across several intelligence disciplines in the DoD context plays an essential role
in producing the knowledge provided by C4ISR systems The problem addressed by the DSC is to attempt
to quantify this statement by directly evaluating the value of fusion in the satisfaction of “information needs,” as defined by the DoD community This chapter gives an overview of the methodology used to perform this type of evaluation.1
*The DSC is a part of the Office of the Secretary of Defense, C3I Directorate and has performed for 4 years studies and analyses for both OSD/C3I and the Joint Staff.
Hans Keithley
Office of the Secretary of Defense Decision Support Center