Multi-Hypothesis Structures and Taxonomies for Combat ...Multi-Hypothesis Structures and Taxonomies for Combat Identification Fusion Tod M. Schuck, J. Bockett Hunter, Daniel D. Wilson

Multi-Hypothesis Structures and Taxonomies for Combat Identification Fusion

Tod M. Schuck, J. Bockett Hunter, Daniel D. WilsonSeptember 2004

Copenhagen DenmarkPaper #141

9th International Command and Control Research and Technology Symposium

9th ICCRTS Sept 2004 paper #141 -2Copyright Lockheed Martin Maritime Systems and Sensors

Agenda

Presentation Objective

Architecture Definition for Disparate Information

Taxonomy f-refinement

Partition Refinement

Canonical Mapping

Response Mapping

CID Realization of JDL Model

CID Bayesian Network

CID Situational Awareness Fusion Expansion

Conclusions and Future Work


Presentation Objective

Address movement of information across multiple hypothesis classes

Relate it to developing the identification of objectsDescribe how it can be combined both within and between JDL levels

Result will be an information architecture that is naturally adaptive to information regardless of quality, level, specificity

We want to answer the question, “How can we establish the relationship between information sets such as

needed for a fusion process????”


An Environment of Disparate Information…


Solution Requires Good Architecture Definition

Defining architecture requires investigation into detailed taxonomic relationships between information sets and subsequent canonical mappingsSubsequent response mapping can be definedResults can be tied into JDL model

Taxonomy – a classification scheme for objects with mutually exclusive labels (parallels study of ontologies)

CID {Friend, Assumed Friend, Hostile, etc.}Nationality {US, Russia, France, Iraq, etc.}Category {Air, Sea, Land, etc.}Platform {Fighter, Bomber, Civil, etc.}Type {F-14, F/A-18, F-22, etc.}Class {F-14A, F/A-18D, etc.}


Taxonomy f-refinement

Taxonomy A is an f-refinement of taxonomy B if:f is a function such that if then

where ϕ = empty set BAf →: 21 bb ≠

ϕ=∩ −− )()( 21

11 bfbf

F/A-18A F/A -18B F/A - 18C F/A -18D F/A –18E F/A –18F

F - 16 F/A-18

A

B

f

F - 16A F - 16B F - 16C F-16D F/A-18A F/A -18B F/A - 18C F/A -18D F/A –18E F/A –18F

F - 16 F/A-18

A

B

f

F - 16A F - 16B F - 16C F-16D

If a taxonomy of A is an f-refinement of taxonomy B and a ∈A, b ∈ B, and f(a) = b, we say that a is an f-refinement of b

Example: F/A-18A in the Class taxonomy is a refinement of F/A-18 in the Type taxonomy


Partition Refinement

Given a set S of objects, a taxonomy imposes a partition on the setEach element of the partition is the set of all elements of S for which a single element of the taxonomy is the appropriate nameExample: an element of the partition imposed on aircraft by the Type taxonomy is the set of all F-15s, all 747s, etc.A taxonomy T1 is a refinement of another taxonomy T2 if the partition imposed by T1 is a refinement of the partition imposed by T2

F - 16 F/A-

18 747

Mig-29

A340

F/A-18FF/A -18EF-16A

F - 16B

F-16C F-16D

F/A-18A

F/A-18D

F/A-18B

F/A-18CF - 16

F/A-18 747

Mig-29

A340

F/A-18FF/A -18EF-16A

F - 16B

F-16C F-16D

F/A-18A

F/A-18D

F/A-18B

F/A-18C


Canonical Mapping

Canonical mappings provide a way to exploit information about an object from different taxonomies to categorize the object in one of those taxonomies, or in another, completely different, taxonomySet of canonical mappings must be defined between any two related taxonomies

T 1

T 3

T 4

T 2

T 5

T 6

m 1m 2

m 3

In the case of a collection of taxonomies that are successive refinements, the canonical mappings reflect the hierarchical nature of the taxonomies themselvesExample: sets T6 and T3 are related through both the mapping m3 and the composite mapping m2*m1 (so not a true canonical mapping)


Response Mapping

Response mapping is a way to interpret a response with elements from one taxonomy in terms of another taxonomyProvides a means of interpreting a response with elements from more than one taxonomy in the various referenced taxonomies

Example: let R be a response from a source of information composed of a set of attributes, let the canonical mapping from taxonomy Ti to taxonomy Tj be mij- Each taxonomy potentially has elements that are part of the response (R1 and R2 in the figure), as well as elements that are the images, under a canonical mapping, of elements in other taxonomies (m12(R1), m21(R2),m13(R1), m23(R2)

m12(R1)

T 1 T2 T3

R R2R1

m12m23

m13

m12(R1) m13(R1)

m23(R2)

m21

m12(R1)m21(R2)m12(R1)

T 1 T2 T3

R R2R1

m12m23

m13

m12(R1) m13(R1)

m23(R2)

m21

m12(R1)m21(R2)m12(R1)m12(R1)

T 1 T2 T3

R R2R1

m12m23

m13

m12(R1) m13(R1)

m23(R2)

m21


JDL Model and CID Realization

JDL Model CID Implementation


A priori CID Bayesian Network (Level 1)

IDF 1 4 AF 1 4 BF 1 4 DF 1 6 AF 1 6 BF 1 6 CF 1 6 DF A 1 8 A CF A 1 8 B DF A 1 8 EF A 1 8 FB o e in g 7 3 7 2 0 0B o e in g 7 3 7 3 0 0B o e in g 7 3 7 4 0 0B o e in g 7 3 7 5 0 0B o e in g 7 3 7 7 0 0B o e in g 7 3 7 8 0 0B o e in g 7 3 7 9 0 0

8 .7 80 .5 70 .5 72 0 .03 .4 02 9 .51 5 .31 2 .94 .0 00 .4 30 .4 30 .3 20 .5 00 .6 51 .1 40 .5 00 .4 20 .5 6

T yp eF 1 4F 1 6F A 1 8B o e in g 7 3 7

9 .9 26 8 .21 7 .84 .1 0

C a te g o r yA ir 1 0 0

C la s sF 1 4 AF 1 4 BF 1 4 DF 1 6 AF 1 6 BF 1 6 CF 1 6 DF A 1 8 A CF A 1 8 B DF A 1 8 EF A 1 8 FB o e in g 7 3 7 2 0 0B o e in g 7 3 7 3 0 0B o e in g 7 3 7 4 0 0B o e in g 7 3 7 5 0 0B o e in g 7 3 7 7 0 0B o e in g 7 3 7 8 0 0B o e in g 7 3 7 9 0 0

8 .7 80 .5 70 .5 72 0 .03 .4 02 9 .51 5 .31 2 .94 .0 00 .4 30 .4 30 .3 20 .5 00 .6 51 .1 40 .5 00 .4 20 .5 6

N a t io n a lity U S B e lg iu m D e n m a rk E g y p t In d o n e s ia Is ra e l N e th e r la n d s N o rw a y P a k is ta n P o rtu g a l S in g a p o re T a iw a n T h a ila n d V e n e z u e la B a h ra in G re e c e K o re a T u rk e y U n ite d A ra b . . . C a n a d a A u s tra l ia K u w a it F in la n d S w itz e r la n d M a la y s ia I ra n A rg e n t in a C h in a E n g la n d J a p a n P o la n d G e rm a n y P h il lip in e s M o ro c c o S a u d iA ra b ia S w e d e n A fr ic a B ra z il F ra n c e S p a in

6 7 .6 0 0 0

0 .2 5 3 0 .9

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

1 .2 6

P la tfo rmF ig h te rC o m m e rc ia l

9 5 .94 .1 0

Given a-priori state of universe for F-14, F-16, F/A-18 and Boeing 737 aircraft

Shaded areas (Nationality and Platform) represent where new info will be inputted


A priori CID Bayesian Network (Level 1)BN after new sensor inputs:

Fighter (0.85)COM Air (0.15)Israel (0.7)US (0.1)Indonesia (0.1)Spain (0.1)

ID

F 1 4 AF 1 4 BF 1 4 DF 1 6 AF 1 6 BF 1 6 CF 1 6 DF A 1 8 A CF A 1 8 B DF A 1 8 EF A 1 8 FB o e in g 7 3 7 2 0 0B o e in g 7 3 7 3 0 0B o e in g 7 3 7 4 0 0B o e in g 7 3 7 5 0 0B o e in g 7 3 7 7 0 0B o e in g 7 3 7 8 0 0B o e in g 7 3 7 9 0 0

7 .1 10 .4 10 .4 11 7 .34 .0 72 3 .46 .1 01 2 .24 .0 70 .3 00 .3 02 .0 32 .0 33 .0 59 .1 52 .0 33 .0 53 .0 5

T yp eF 1 4F 1 6F A 1 8B o e in g 7 3 7

7 .9 35 0 .81 6 .92 4 .4

C a te g o r yA ir 1 0 0

C la s sF 1 4 AF 1 4 BF 1 4 DF 1 6 AF 1 6 BF 1 6 CF 1 6 DF A 1 8 A CF A 1 8 B DF A 1 8 EF A 1 8 FB o e in g 7 3 7 2 0 0B o e in g 7 3 7 3 0 0B o e in g 7 3 7 4 0 0B o e in g 7 3 7 5 0 0B o e in g 7 3 7 7 0 0B o e in g 7 3 7 8 0 0B o e in g 7 3 7 9 0 0

7 .1 10 .4 10 .4 11 7 .34 .0 72 3 .46 .1 01 2 .24 .0 70 .3 00 .3 02 .0 32 .0 33 .0 59 .1 52 .0 33 .0 53 .0 5

N a t io n a l it y U S B e lg iu m D e n m a rk E g y p t In d o n e s ia Is ra e l N e th e r la n d s N o rw a y P a k is ta n P o r tu g a l S in g a p o re T a iw a n T h a ila n d V e n e z u e la B a h ra in G re e c e K o re a T u rk e y U n ite d A ra b . . . C a n a d a A u s tra lia K u w a it F in la n d S w itz e r la n d M a la y s ia I ra n A rg e n t in a C h in a E n g la n d J a p a n P o la n d G e rm a n y P h il lip in e s M o ro c c o S a u d iA ra b ia S w e d e n A fr ic a B ra z il F ra n c e S p a in

6 1 .4 2 .0 9 0 .8 9 2 .6 5 0 .3 3 3 .2 0 3 .2 4 0 .8 9 0 .9 4 0 .2 1 0 .6 8 1 .7 8 0 .7 4 0 .2 5 0 .2 6 1 .8 2 2 .2 6 3 .3 4 1 .0 1 1 .6 3 1 .5 1 0 .4 9 0 .7 0 0 .3 7 0 .6 0 0 .8 4 0 .1 8 0 .4 3 0 .4 3 .0 6 5 0 .2 1 0 .8 3 .0 5 6 0 .3 7 0 .1 6 1 .0 4 0 .1 7 0 .6 2 0 .3 7 0 .9 0

P la t fo rmF ig h te rC o m m e rc ia l

7 5 .62 4 .4


CID Situational Awareness (SA) Fusion Expansion

Level 1 SA: Perception of the environmental elements – The identification of key elements of “events” that, in combination, serve to define the situation

JDL – numeric processing of tactical componentsSA – symbolic processing of these entities

Level 2 SA: Comprehension of the current situation – This combines level 1 events into a comprehensive holistic pattern (or tactical situation)

JDL and SA virtually identicalLevel 3 SA: Projection of future status – Projection of the current situation into the future, so as to predict the course of an evolving tactical situation

SA more general than JDL, includes projection of ownship/aircraft/etc., and friendly intent


Tactical Elements Employed by Fusion Level

What?

Level 1: Object Refinement

Where?

Who?

Level 2: Situation Refinement

When?

What if?

Level 3: Threat Refinement

Why?Ph

ysica

l Obj

ects

Even

ts

Thre

at En

viron

men

t, En

emy

Tacti

cs, &

Thr

eat B

ehav

iors

Enem

y doc

trine

,

objec

tives

& ca

pabil

ity

Frien

dly vu

lnera

bilit

ies

and m

ission

s

individual

organizations

Beha

viora

l Obj

ects

tactical elements

local

global

local

global

specific

aggregatedoptions

needs


Proposed Threat Evaluation Tool

I&WIndicators & Warnings

Threat Assessmentstate, confidence, rationale, & source

Intent Determinationstate, confidence, rationale, & source

“Impact” Assessment (Threat Evaluation)

“Intent” Determination (Mission, Activity)

Alert Generation

“Trigger” ThresholdsCID Mission Planning (IPB)

HypothesisManager

Flight plans, FAA updates, ADS-B

Anticipated scenarios, threats and environments

Defensive missions, resources, & assets

Courses of ActionAssessment/

Comprehension

Predictive Profiling

Threat Assess.Assessment/ Comprehension

Impact Modeling

Hypothesis Generation/ Refinement


Hypotheses on “Intent” (mission, activity, etc.)

Impact Proj.

Track Fusion

CID Fusion

JDL Level 1 Hypotheses


Capability & effectiveness of classDesign mission of class

Source of class

Source of ID

Goals of ID

Knowledge Base

Initial Hypothesis Generation


“Mission” Recognition

Object Behaviors “Event”

Recognition

“Object” Aggregation

Domain of Level 3

Fusion

I&WIndicators & Warnings

Threat Assessmentstate, confidence, rationale, & source

Intent Determinationstate, confidence, rationale, & source

“Impact” Assessment (Threat Evaluation)

“Intent” Determination (Mission, Activity)

Alert Generation

“Trigger” ThresholdsCID Mission Planning (IPB)

HypothesisManager

Flight plans, FAA updates, ADS-B

Anticipated scenarios, threats and environments

Defensive missions, resources, & assets

Courses of ActionAssessment/

Comprehension

Predictive Profiling

Threat Assess.Assessment/ Comprehension

Impact Modeling



Hypotheses on “Intent” (mission, activity, etc.)

Impact Proj.

Track Fusion

CID Fusion



Capability & effectiveness of classDesign mission of class

Source of class

Source of ID

Goals of ID

Knowledge Base



“Mission” Recognition

Object Behaviors “Event”

Recognition

“Object” Aggregation

Domain of Level 3

Fusion


Conclusions and Future Work

Contextual relationship of information is paramountFusion process must incorporate these relationshipsCID information wrt context, time, timeliness, quantity, and quality must be knownFuture work:

Metrics for information value, completeness, and costs of decisions can be developed and integratedContextual reasoning leading to predictive SA


Author Information

Tod Schuck is a Lead Member of the Engineering Staff at Lockheed Martin MS2, Moorestown, NJ. He is the recipient of numerous awards for his technical work including the ‘01 Lockheed Martin MS2 Author of the Year, finalist for best paper by OASD C3I at the 7th International Command and Control Research Technology Symposium (ICCRTS) ‘02, best paper (2nd place) IEEE National Aerospace and Electronics Conference (NAECON) ‘00, and a group award winner of Vice President Al Gore’s “Silver Hammer”Award for Extraordinary Effort for Changing the Way That Government Does Business (July ‘97) as the technical lead for the Mk 17 NATOSeaSparrow radar SDP. His areas of expertise are in identification sensor design and development, fusion algorithm development, information architecture design; and requirements generation, and testing. He holds a BSEE from Georgia Tech, an MSEE from Florida Tech, and is pursuing a Ph.D. from Stevens Institute of Technology in Systems Engineering.

J. Bockett Hunter is a Senior Member of the Engineering Staff at Lockheed-Martin MS2, Moorestown, NJ. He has been on the faculty of Indiana University, worked on Navy operations research problems, and a wide variety of intelligence systems. His areas of expertise include algorithm development, systems engineering for very large systems including system concept definition, requirements development, design, and testing. He holds a BS in mathematics from Caltech and an MA in mathematics from Indiana University.

Daniel D. Wilson is a Principal Systems Engineer at Lockheed-Martin MS2, Burlington, MA. He has worked on Military Operations Research problems in support of numerous DoD programs. His areas of expertise include C4ISR systems, operations analysis, knowledge fusion, predictive situational awareness, dynamic replanning strategy development, collaborative mission planning, system effectiveness modeling, decision analysis, risk analysis, and cost estimation. He holds a BS in Computational Mathematics from the University of Vermont and an MS in Operations Research from Stanford University.

[email protected]@lmco.com

[email protected]

Documents

Multi-Hypothesis Structures and Taxonomies for Combat ...Multi-Hypothesis Structures and Taxonomies for Combat Identification Fusion Tod M. Schuck, J. Bockett Hunter, Daniel D. Wilson