Keith Knox (AFRL)

Report
Air Force Research Laboratory
Lead ~ Discover ~ Develop ~ Deliver
Multimodal Data and Anomaly
Detection in SSA at AMOS
15 Oct 2012
Dr. Keith Knox
Air Force Maui Optical &
Supercomputing Site
Maui, Hawaii
Photo of
Briefer
Air Force
Maui Optical & Super Computing Site
2
Air Force
Maui Optical & Supercomputing Site
50 Years of Service to the Department of Defense
1960: The Advanced Research Projects Agency (ARPA) Midcourse
Optical Station
1963: Site construction started by ARPA
1994: High Performance Computer Center (HPCC) completed
1999: Advanced Electro-Optical System (AEOS) completed
2001: Air Force Research Laboratory
• Largest telescope in Department of
Defense with 3.6m primary telescope
• Highest resolution adaptive optics in
Department of Defense
• Largest electro-optical tracking facility in
the Pacific
3
Air Force Research Laboratory Directed Energy Directorate
Maui Space Surveillance System
MSSS (AFRL)
3.6m and 1.6m telescopes
Maui Space Surveillance System
1.6m
•
•
•
High-resolution Imaging
Orbital Tracking
Space Object Characterization
AEOS
3.6m
GEODSS (AF Space Command)
Ground-based Electro-Optical Deep
Space Surveillance
GEODSS
4
3.6-meter Telescope
AEOS 3.6
Telescope
Advanced Electro-Optical System
5
1.6-meter Telescope
1.6 Meter Telescope Inside Dome
6
Adaptive Optics Imaging
AEOS Visible Imager
Day
Terminator
Night
Hubble Space Telescope
Adaptive Optics (AO) plus
multi-frame blind
deconvolution processing
• Terminator Imagery
7
Long-Wave Infrared Imaging
AEOS Infrared Imager
Day
Terminator
Night
• Resolved thermal images
• Virtually diffraction-limited
• Nighttime Imagery
8
Speckle Imaging
Speckle Imaging 1.6m
Day
Terminator
Raw data
Processed Result
Night
• Daytime Imagery
• Terminator Imagery
9
Non-imaging Characterization
• As satellite image size decreases …
– Smaller satellite
– Greater distance
• the satellite becomes completely unresolved
– Satellites in geo orbit
– Cubesat-class satellites
10
Non-Imaging Technique
• Temporal filter photometry
– Measured brightness as function of time
• For an object facet to contribute to signal
– Facet must be illuminated by Sun
– Facet must be visible to sensor
Temporal photometry
• Sensor requirements are simple
– Calibrated light bucket
11
Astrodynamics & Tracking
High Performance Computing Software Applications Institute for Space
Situational Awareness
Now
15k Objects
Good orbit
knowledge and
some status info
Future
150k Objects
Accurate orbits and
uncertainty
Object identification
status and health
The Institute meets these challenges by bringing together:
• Supercomputing expertise
• World-class researchers from AFRL
12
Anomaly Detection and Multimodal
Data in Astrodynamics
13
Space Catalog Anomalies
• 22,000 objects in the catalog
• What is an anomaly?
–
–
–
–
–
New satellite is launched
Debris is created
Satellite maneuvers
Object drifts
Satellite status has changed
• Track 22K objects
– Look for deviations
• Paul Schumacher
– “The Future Space Catalog”
14
Multimodal Space Catalog:
Radar vs. Optical
2 radar observations
Range1
Az1 Elev1
3 optical observations
Range2
Az2 Elev2
RA1
Dec1
RA2
Dec2
RA3
Dec3
2 observations
6 scalars
( x y z vx vy vz )
6 scalars
3 observations
6 scalars
( x y z vx vy vz )
6 scalars
Radar and Optical Observations translated into 3-D spatial coordinates
15
Observation Association is Hard
?
?
SST
?
?
Space
Surveillance
Telescope
For one object, all
observations are connected
For many objects, all
observations are connected
16
Multimodal Data in Speckle Imaging
17
Speckle Imaging
using Short Exposure Sequences
• High resolution details are lost in long exposures
through the atmosphere:
• However, detail is encoded in short exposure
images:
• Assume that target is constant over period of a few
seconds. Then image reconstruction is possible:
18
Speckle Imaging
Multi-Frame Blind Deconvolution (MFBD)
Noisy and blurred images
Blurring functions
True object
h1x 
i1x  ox  h1x  n1x
o(x)
h2 x 
i2 x  ox  h2 x  n2 x
…
…
hN imagesx 
Restored
object
iN imagesx   ox   h N imagesx   nN imagesx 
MFBD Processing
Minimize this cost function with
respect to oˆ x  , hˆ 1 x  , … , hˆ N x  :
images
oˆ x 
Nim agesNpix els
 
k 1
n 1

i x   ˆi x 
σ x 
1
k
2
k
n
n
k
2
n
19
Multimodal Data
Improves Image Reconstruction
• Each wavelength experiences ~same optical path
difference (OPD) due to atmospheric turbulence
• Wavefront phase is θλ = OPD × 2π/λ
OPD in telescope pupil
Infrared images define OPD, which in turn improves visible reconstruction
Brandoch Calef, “Wavelength Diversity”
20
Multimodal Data in Non-Imaging
21
Spectrophotometry with BASS at AEOS
•IR spectrophotometry in 3-13.5 mm range
•Princeton CCD camera & filter collect
images simultaneously with IR spectra
22
22
Modeling Reflected & Emitted
Radiation
• Modeling space
debris to match
simultaneous IR
and visible
response
• HAMR objects
– High Area-to-Mass
Ratio
• Mark Skinner
– “Fusing Visible
and Thermal IR
Signature Data for
SSA”
23
Anomaly Detection in Non-Imaging
24

similar documents