a stochastic model-based approach to sar atr lee montagnino electronic systems and signals research...

25
A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering Washington University St. Louis, Missouri Supported in part by ONR grant N00014- 98-1-06-06

Upload: jewel-carter

Post on 02-Jan-2016

216 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

A Stochastic Model-Based Approach to SAR ATR

Lee Montagnino

Electronic Systems and Signals Research Laboratory

Department of Electrical and Systems Engineering

Washington University

St. Louis, Missouri

Supported in part by ONR grant N00014-98-1-06-06

Page 2: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 2

Presentation Overview

Problem DefinitionLikelihood Approach to ATRConditionally Gamma ModelConditionally K distributionAzimuth Correlation ModelConclusions

Page 3: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 3

Problem Definition

Typical Recognition Scenario

Imaging Platform

Target Classifier

Orientation Estimator

65ˆ

72ˆ Ta

Page 4: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 4

Problem Definition

Model-Based Recognition

65ˆ

72ˆ TaTarget Classifier

Orientation Estimator

form)Simply Unior (Known ClassTarget on Prior -

form)Simply Unior (Known n Orientatioon Prior -

Model Data lConditiona - ,,

ap

ap

ap

A

A

A

rR

Page 5: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 5

Problem Definition

Model-Based Recognition

Functional Estimation

Training Data

Scene and Sensor Physics

ProcessingImage

Inference ,aL r 72Tˆ a

Page 6: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 6

Problem Definition

Use Modular Software Test Bed to Perform: Direct comparisons of different

stochastic models Performance analysis under a wide

range of testing and training scenarios Detailed study of performance vs.

models and model parameters

Page 7: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 7

Likelihood Approach to ATR

Target Class and Pose Estimates

,,ˆminargˆ

,maxargˆ

2

HSˆHS

,Bayes

aOOEa

dapapaPa AAa

rr

rr R

cossin

sincos

where

O

Page 8: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 8

Likelihood Approach to ATR

Generalized Likelihood Ratio Test and Maximum-A-Posteriori Estimation

,ln,lnmaxargˆ

ln ln,lnmaxmaxargˆ

,MAP

,GLRT

apapa

Apapapa

AA

AAa

rr

rr

R

R

Page 9: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 9

MSTAR DATA SET

A collection of spotlight mode SAR images from a number of target classes Using 4 target classes from the public

release set Using 10 target classes from the public

release set MSTAR Program sponsored by DARPA and

Wright Laboratory

Page 10: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 10

MSTAR Data Set

Partitioned into two subsets: 17 ° depression images used for

estimating likelihood functions 15 ° depression images used for

experimentally assessing performance

For testing, we assume a uniform prior on orientation and target class

Page 11: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 11

Gamma Distribution

Multi-parameter distributionRelaxation of the quarter-power

normal modelRelates to Work in MSTAR Program

at WPAFB

,/

1,

1GAMMA ,,,

1, ar

a

i

in

i ii

ii

i

ea

r

aaap

r

Page 12: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 12

Gamma Estimates

Maximum Likelihood Estimates

where

ii

ii

AG

A

ˆˆ

lnˆˆln

) of derivative ic(logarithmfunction digamma theis

and , ,/1/1

11

nn

i i

n

i i rGrnA

Page 13: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 13

Gamma Results

Percentage of Correct Classification and Orientation Estimation Error

Page 14: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 14

K Distribution

Multi-ParameterMixture modelModels Specular and Diffuse

Reflectivity

n

i iiaa

ii

aai

arK

aa

rap

ii

ii

11,2/1,

2/,3,

K ,

2

,,

2,

r

Page 15: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 15

K Estimates

Expectation-Maximization

ki

ki

kik

i

ki

ki

ki

kik

iki

A

G

A

ˆ,ˆ,ˆˆ

,ˆ,ˆ

,ˆ,ˆlnˆˆln

1

11

r

r

r

n

iki

ki

n

ii

ki

ki

n

i

iki

ki

ikik

iki

kik

i

ki

rKrG

rK

rK

nA

/1

ˆ111

ˆ1

ˆ

ˆ2

lnexpˆ2

,ˆ,ˆ ,

ˆ2

ˆ2

ˆ2

1,ˆ,ˆ

rr

Page 16: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 16

K Results

Percentage of Correct Classification and Orientation Estimation Error

Page 17: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 17

Azimuth Correlation

Radar data correlated in azimuth

M

Mmmm wZ R

mlo

M

Mlml

M

Mmml NZZERREK ,212121,

mlo

M

Mlml

M

Mmm NK ,212

21,

2/12

2sin

122

21

sin

122

1

Mm

Mm

M

Mm

Page 18: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 18

Azimuth Correlation Functions

EM Algorithm to Find Estimates of

pm

p

m

p

m

p

m

pppp

M

MmNom

p

mp

EK

KKKE

INKK

traceˆˆˆ

ˆˆˆ

ˆˆ

22212

11

2

rr

2ˆm

Page 19: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 19

Azimuth Correlation Covariance Images

Page 20: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 20

Azimuth Correlation Results

Percentage of Correct Classification and Orientation Estimation Error

Page 21: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 21

Conclusions

Gamma Distribution Model low recognition rates poor orientation estimation

K distribution model comparable recognition rates to the

zero-mean conditionally Gaussian presented by DeVore

Page 22: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 22

Conclusion

Azimuth Correlation comparable recognition rates to the

zero-mean conditionally Gaussian model presented by DeVore

best orientation estimation error rates of any distribution

correlation models don’t match actual data

Page 23: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 23

Questions

Page 24: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 24

References

Page 25: A Stochastic Model-Based Approach to SAR ATR Lee Montagnino Electronic Systems and Signals Research Laboratory Department of Electrical and Systems Engineering

February 6, 2004 Montagnino: SAR ATR 25

References