los alamos national laboratory from sensing to information...

64
From Sensing to Information: Everything Under the Sun (LA-UR 09-02038) Andrea Palounek Space and Remote Sensing Sciences (ISR-2) Los Alamos National Laboratory [email protected] Los Alamos National Laboratory

Upload: vukhanh

Post on 01-Feb-2018

218 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

From Sensing to Information:Everything Under the Sun

(LA-UR 09-02038)

Andrea PalounekSpace and Remote Sensing Sciences (ISR-2)

Los Alamos National Laboratory

[email protected]

Los Alamos National Laboratory

Page 2: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

We go from wild ideas to useful applications –with all the steps in between

Nonproliferation Space WeatherSensor ArraysLightning

Stellar

LunarProspector

Vela

Gamma-ray Bursts

Solar Wind Physics

RF/EMP DetectionMachine Learning

Theory and ModelingAutonomous Computing

O ti l/RF R t S i

Planetary Composition

Nuclear Detonation DetectionStellar

Formation

Treaty Monitoring & Verification

V Sensor

Atmospheric science

RF EMPGPS

National SecurityMissions

CoreScience andTechnology

Solar Wind PhysicsOptical/RF Remote SensingAnomalous Change Detection

Image Analysis & InterpretationComplex Algorithm Dev. & Sim

Reconfigurable Computing I f ti S i d

Space Situational Awareness Fast Transient Astrophysics

Reconfigurable ComputingNeutral Atom ImagingSatellite TechnologiesX-ray, γ-ray Detection

Microcalorimetry

Energy and Water SecurityInformation Science and

Knowledge ExtractionHigh Power Electrodynamics

Ionospheric Physics

μcalorimetry

SAVE/SABRS Water on Moon & MarsCometary Physics

MicrocalorimetryNeutron ImagingRadiation Effects

Neutron Physics

Persistent SurveillanceIonospheric Physics

Forte

LAPP

Meteorites

PetaVision

Forte

CFE MTI

Page 3: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Our Division Represents 40+ Years of Space Experience: 1400 sensors, 400 instruments,

60 satellitesFORTÉ

ALEXIS

VELA HOTEL

ALEXIS

DSPGPS

Multi-Spectral Thermal Imager

Cibola Flight Experiment-Space-based R&D Flight Demo w/RCCImager Demo w/RCC

Page 4: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

We Began with Nuclear Detonation DetectionDetection…

Space • Gamma Rays• Neutrons• X-rays

~100 km~100 km

Transition Region • Optical

30 km30 kmBelow Ground• Seismic

Low Altitude

• Gamma Rays• Neutrons

• Seismic• Hydro-acoustic

Low Altitude • Optical• Electromagnetic Pulse• InfrasoundInfrasound

Page 5: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

…and Now Include Proliferation Detection100 km100 km

30 km30 kmThink of the problem through the process of acquiring and using a weapon

Nuclear ProliferationNuclear Proliferation Proliferation Detection and

Raw Material Extraction & PreparationRaw Material Extraction & Preparation EnrichmentEnrichment Production ReactorProduction Reactor

Detection and Response is a complex, multidimensional

bl

Chemical &Chemical &

Storage & Storage & DeploymentDeployment

Device Fabrication & Device Fabrication & NonNon--nuclear testingnuclear testing

Weapons Weapons Material Material ProcessingProcessing

ReprocessingReprocessing

problem

Chemical & Chemical & Biological Biological ProliferationProliferation

Field TestingField Testing

gg

Storage & Storage & D l tD l t

Factory Production Factory Production f A tf A t

Weaponization & Weaponization & P k iP k i DeploymentDeploymentof Agentof Agent PackagingPackaging

Page 6: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Electro-magnetic pulse sensor heritage is a classic example of Los Alamos development strategy

UNCLASSIFIED

example of Los Alamos development strategyFocused short-term R&D supports eventual operational capability

Operational SensorsOperational SensorsGPS IIA/IIR W-Sensor GPS IIF V-SensorVela W-Sensor GPS III V-Sensor

???R&D Sensors (“Free-flyers”)

???ALEXIS/Blackbeard (1993 launch) FORTE (1997 launch) CFE (2007 launch)

UNCLASSIFIED

Page 7: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

ISR-2 Sensor Development for SatellitesUNCLASSIFIED

V-Sensor• Sophisticated radio receiver designed to

detect emissions from nuclear detonations• Supports treaty verification for the US

Government• Hosted on the US Air Force GPS (Global

Positioning System) satellite constellation• ISR Division satellite sensor heritage dates

back to the Vela program of the 1960sback to the Vela program of the 1960s• V-Sensor is designed and produced by the

LANL space science/engineering groups

Technicians adjust a V-Sensor antenna on a GPS Block IIF space vehicle

(B i h t h)(Boeing photograph)

UNCLASSIFIED

POC: David A. Smith, [email protected]

Page 8: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

The Los Alamos Portable PulserUNCLASSIFIED

The LAPP produces a pbroadband VHF signal that mimics what would come from a nuclear weapon The signalnuclear weapon. The signal, broadcast into space through this dish antenna, is used to calibrate EMP sensors on orbiting satellites.

UNCLASSIFIED

POC: Kalpak A. Dighe, [email protected]

Page 9: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Lightning is a violent atmospheric event: an excellent remote sensing tracer to study severe weatherremote sensing tracer to study severe weather

Lightning has complex structure and a broad spectrum. A significant portion of the RF transientspectrum. A significant portion of the RF transient is below 10 MHz, undetectable from space. ⇒

Terrestrial RF remote sensing

Lightning SpectrumLightning Spectrum

Lightning W fWaveforms

Page 10: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Terrestrial RF remote sensing is a multidisciplinary effort of physics, collection, and information integrationeffort of physics, collection, and information integration

Deployment

Signal Propagation• Ionosphere limits signal propagation for RF <~10-30 MHz

Sensor• Meas. physics: power, E, B• Antenna• Filtering & Tuning

Deployment• Ground, air, sea

propagation for RF <~10-30 MHz• Reflection, refraction, transmission, waveguide

Filtering & Tuning• Digitization & recording

RF Signatures• Persistent: Radio• Transient: Lightning

Information• Multiple sensor geolocation• Signal characterization and classification

Page 11: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

The Los Alamos Sferic Array (LASA) exploits modern advances for distributed terrestrial RF remote sensingadvances for distributed terrestrial RF remote sensing

Key Modern Advances• IT components: computer, digitizerAntenna Signal

conditioning Digitization digitizer• IT infrastructure: Internet• Absolute reference: GPS

conditioning g

Event Capture &Triggering

InternetCapture &

Report (Wireless) WAN

GPS Timing &

Triggering

Deployed Sensors Multi-site info integration 3D Geolocation

GPS receiver Location

Reference

Power &Event profile retrieval & processing

Sensor managementDeployment

Host

Power & Mechanical

Processing ServerHost

POC: Cheng Ho, [email protected]

Page 12: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

LASA has a rich history of deployment and scientific results

2005 GPN Deployment

Four-station detection allowed us to

l tgeolocate and classify.

LASA GPN array 2005 observed d h t i d H i Ritand characterized Hurricane Rita

Page 13: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

LASA is being reinvigorated in 2008-2009 with the Gulf as the observation focal point

Ci l f 1000Circle of 1000 km radius

LASA2005

Deployed as of 20081015

Planned deploymentC ll b ti ( t ti )Collaborative sensors (representative)

Page 14: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

We continue to improve on LASALASA Improvements• System robustness• Optimization• Component upgrades and obsolescence

defeat• Front-end adaptiveness

LASA O ti & S iLASA Operation & Science• Data production• Pipeline processing• User interfaceUser interface• Scientific exploitation• Collaboration

General Terrestrial Remote SensingGeneral Terrestrial Remote Sensing• Expand RF signal measurement space and

mission areas• Systematic approach to collection/processing

adaptivenessadaptiveness

POC: Dr. Cheng Ho, 505-667-3904, [email protected]

Page 15: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Predicting Hurricane Intensification Using LASA Measurements of Eyewall LightningMeasurements of Eyewall Lightning

Goal is to perform the first-ever 3-D mapping of convective events in the hurricane eyewall using LASA, then

Demonstrate that rapid hurricane intensification (sudden large-scale transition and reorganization) can be forecast accuratelyscale transition and reorganization) can be forecast accurately using a novel model that assimilates real-time knowledge of critical small-scale processes

New RF array measurements of hurricane lightningNew RF array measurements of hurricane lightning

New cloud physics modeling of cloud electrification

New data assimilation scheme to ingest lightning observationsNew data assimilation scheme to ingest lightning observations, and thereby improve hurricane forecast accuracy

POC: Christopher A. M. Jeffery, [email protected]

Page 16: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

We observed lightning activity from Rita

Potential:World-wide tracking of

h i d hhurricanes and other severe storms

21 Sep 05 16:00-17:00 UTC

21 S 05 09 00 10 00 UTC21 Sep 05 09:00-10:00 UTC

Rita category 5: intense lightning marks boundary of eyewall

Shao et al., EOS, 86, 42, 18 Oct. 2005

y y

Rita category 3: lightning in outer rain bands, first signs of eyewall lightning

Page 17: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Rita intensification was coincident with the onset of lightning activity

Shao et al., Eos, 86, 42, 18 Oct. 2005

Eye-wall lightning detected at stage of hurricane intensification and at time of landfall

21 Sep 05 16:00-17:00 UTC 22 Sep 05 00:00-01:00 UTC 23 Sep 05 08:00-09:00 UTC

Eye-wall lightning detected at stage of hurricane intensification and at time of landfall Little eye-wall lightning while hurricane decays

Samples of Rita lightning observations

Lightning height increases while Rita intensifying

Lightning rate and eye-wall pressure Rita

Lightning rate and eye-wall pressure Katrina

Page 18: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Los Alamos Sferic Array (LASA)Shao et al., 2006, J. Atmos. Ocean. Technol.

Frontal storm in Florida 3-D lightning location

Lightning time sequence Lightning flash types

Return strokes

Radar echo and hail/tornado report Lightning activitySevere storm over Great Plains

p18:00-24:03 UT, June 2, 05

g g yNotice type changes when storm becomes

severe

300

km

Page 19: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Multidecadal variability of Atlantic hurricane activity: 1851–2007hurricane activity: 1851–2007

Chylek and Lesins, JGR 113, D22106

An Atlantic hurricane activity i d h iindex that integrates over hurricane numbers, durations, and strengths during the years 1851–2007,during the years 1851 2007, suggests a quasi-periodic behavior with a period around 60 years superimposed upon

li l i ia linearly increasing background.

(a) Annual hurricane activity index and its 21-( ) yyear running mean suggests a linear trend with an increase of about 0.6/yr. The detrended HAX preserves a quasi-periodic oscillation with a period of about 60 years.

(b) Alternating periods of high and low hurricane activity.POC: Petr Chylek, [email protected]

Page 20: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

From Sensors to Information: an End-to-End Approachan End to End Approach

Signal propagation

Signatures &backgrounds

Signal propagationand transduction

Data collectionand fusion

Knowledgegeneration

Infrared absorption

absorption andscattering of infrared

spectra

pspectrum

plant emission model

ammonia plumedetected by

hyperspectral techiques

p

plant emission model

Page 21: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Physics-Based Remote Sensing Analysis

Source Physics• Direct emission and reflected radiation from many sources

– Surface properties (emissivity & reflectivity) difficult to characterize– Wavelength, temperature, direction, and polarization dependent

• Differential emission/absorption/reflection characteristics compared to complex background

Cluttered scenes– Cluttered scenes – Large background signatures

Propagation Analysis• Radiative transfer through g

inhomogeneous atmosphere– Aerosols & absorbing gases– Inhomogeneities

• Vertical & horizontal• Dynamic Atmosphere• Dynamic Atmosphere

– Local weather patterns – Small scale fluctuations ,

e.g., turbulent eddies

Sensor ModelingSensor Modeling

Page 22: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Hyperspectral Imagery for Gas Detection

Core capabilitiesSystem modelingScene characterization Ch i l id tifi tiChemical identification

WB 57 Ai ftWB-57 AircraftIndustrial facility

Ammonia “release” at LANL

Smaller, cheaper…

Page 23: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

The ORCAS Compact Long-Wave IR Hyperspectral Imager(PI: Steven P. Love, [email protected])

• Compact Optical Package

• High Spectral & Spatial Resolution in the LWIR Atmospheric Window

• Rapid, Sensitive Imaging and Chemical ID of Gas Plumes

Spectral Range: 7.6 - 13.5 µmPixel Size: 40 µmDetector: HgCdTeArray Size: 256x256 pixelsArray Size: 256x256 pixelsSampling Increment:

0.023µm (2.3 cm-1@10µm)F ratio: f/3.8Field of View: 6.94˚ vert., arb. horiz.Si l i l if 0 027˚ 0 47 d

ORCAS Broadband IR

Single pixel ifov: 0.027˚, 0.47 mrad

ORCAS Chemical Matched Filter Images (ethanol)

Ethanol plume

Ethanol plume

Page 24: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

QuickTime™ and aMicrosoft Video 1 decompressorare needed to see this picture.

Page 25: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Remote Ultra Low-Light Imaging (RULLI)POC: Robert Shirey, [email protected]

Persistent situational awareness reaching the quantum limitOptical sensors that simultaneously measure each photon’s

position & time of arrival

Active 3D imaging is enabled by g g yexquisite measurement of photon time-

of-flight with low-power pulsed laser

Moonless geo-registered image from i l h t i

Time-tagged photons are very amenable to platform motion correction and dynamic scenes

single-photon imager

Page 26: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

LANL has developed a Nocturnal Camera, NCam

• Delivers RULLI capabilities in a compact camera form• Camera is 5”x5”x13” and 10 lbs (plus COTS lens and GPS/IMU)• <30 W external power or a laptop-class battery

• User-friendly operation• USB 2.0 data acquistion

TCP/IP C d & C t l d SOH• TCP/IP Command & Control and SOH

• Interfaces to Canon EF lenses

Page 27: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

RULLI offers important capabilities

• Starlight-only optical imaging is challengingSingle photon limit ⇒ Long exposure and low noise– Single photon limit ⇒ Long exposure and low noise

– Motion smearing ⇒ Short exposure or high time resolution

• RULLI offers capabilities such as:p– Single photon sensitivity with very low noise– Large format/fill factor/duty cycle imaging AND timing– Passive low-light motion-immune imagingPassive low light motion immune imaging– Exploitation of three-dimensional info collection capability

3D imaging with active illumination; time-encoded color imaging, polarimetry, and hyper-spectral; time-resolved spectroscopyyp p ; p py

Page 28: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Moonless RULLI image

Contrast Imaging Under Moonless Conditionsat 5100’ AGL with 135mm f/2.0 lens

Daytime USGS image

1.5 km x 0.5 km image Moonless is 9 orders of magnitude fainter than sunlit

(1998)

Page 29: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

RULLI Image Formation & g

Motion Compensation

Daytime imagey g

Motion compensation

Attitude Knowledge RULLI

RULLI Motion-corrected image

Motion-compensation algorithm

Scene images

• Image formation performed with modern computers • Merges RULLI data with

g

attitude knowledge (not control)• Post acquisition processing Exquisite spatial-time info allows us

to correct complicated motion pattern

Page 30: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

A single-photon imaging sensor with sub-nanosecond timing is ideal for active characterization of complex terrain

• 3D imaging achieved via time-of-flight range measurements of pulsed laser

illumination

is ideal for active characterization of complex terrain

illumination

• Instantaneous coverage over a wide area

• 3D imaging from a single vantage point

• No intrinsic moving parts and room-temperature operation

• Low illuminator power, thus low mass/power/volume and eye safe

• Relative motion can be removed in softwareRelative motion can be removed in software

• Works under low ambient lighting conditions (limited to night-time operation)

100 light-picoseconds in air= 3 cm

Page 31: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Why is Photon Counting and Timing Good for Active (3D) Imaging?Good for Active (3D) Imaging?

Lots of information from very few photons:

2-D (contrast) imagingrequires many photons per pixel (or resolution element)requires many photons per pixel (or resolution element)Need 100 photons/pixel to distinguish 10% albedo differences

3-D imagingNeed very few photons/pixel to distinguish a surfaceNeed very few photons/pixel to distinguish a surface2-3 photons in a pixel within a single range bin gives very high probability

of a surface.Low background noise helps (our detector: <0.01/pixel/s)

Page 32: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Active 3-D Imaging with RULLIwith RULLI

By measuring time-of-flight of photons from a pulsed laser, RULLI can measure range over entire FOV from single vantage point and with no moving parts.

literal 3D imaging

Top and side-view images cover 180-395 m in range

• 1 s (left) and 10 s (right) exposures• 150k and 1.5M photons respectively • 15-m data cubes, 5-cm voxels• 220-m distance• 300-mm f/2.8 lens• 1 MHz laser pulse rate

For Official Use Only

Page 33: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Cloud of photonsOutgoing laser pulse timing

From Photons to Topography

3D

Cloud of photonsOutgoing laser pulse timing

3D Georegistration

photons in detector spaceStatistics driven

topography extraction

F(X,Y,Z) → Z(X,Y)

Contemporaneous position & attitude information

topography

Page 34: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

• Airplane @ ~100 knots

Versatile Motion Correctionp a e @ 00 o s

• Serendipitous observation of a moving boat• Both the Airplane and boat’sBoth the Airplane and boat s motion can be corrected

Ai l @Airplane @ ~100 knots

Boat @ ~6 knots

Both airplane and boat motion

t d 3D view of boat in its own rest framecompensated 3D view of boat in its own rest frame

Page 35: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

A serendipitous image…

QuickTime™ and aBMP decompressor

are needed to see this picture.

Page 36: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

RULLI Summary

There are many techniques for low light and high time resolution imaging, ith i t th d kwith various strengths and weaknesses

There is no one “silver bullet”

Single-photon high-time-resolution imaging, as with the RULLI f ftechnology, is well suited for a number of applications:

low light imaging of rapidly changing or moving scenes

active characterization of complex terrain

remote characterization of complex 3D objects

highly flexible correction of motion without a priori knowledge

A li ti i h t t h i i dApplications in many areas such as astronomy, atmospheric science and biology

Further development is underway to extend the performance, particularly towards much higher photon detection ratestowards much higher photon detection rates

Page 37: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Microcalorimeter nuclear spectrometersp

241Am 238Pu 241Am 238Pu

First SNM array spectrum World record resolution: ΔE=1 06 keV at 5 3 MeVResolution: 22 eV FWHM

T=0.1 K – no liquid He or N2

ΔE=1.06 keV at 5.3 MeVFirst α spec splitting of Pu peaksFirst mixed actinide α spec

resolving all peaks

Testing 66-pixel chips

Page 38: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

RaveGrid: Raster-to-Vector Graphics for Image Data

• RaveGrid enables• image scaling to the pixel

resolution of a particular digital display or Web-page layout;i i t d• image compression to reduce image-storage or bandwidth requirements;

• encryption of vectorized images inencryption of vectorized images in text files;

• image searches in large databases or on the Internet; and

• automatic analysis of reconnaissance or surveillance images

Page 39: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Think of image understanding as having two key tasks

Image Segmentation: Decomposing an image into constituent objects•Progressive synthesis of features into objectspixels edges contours shapes objects

Object Recognition: Identifying and understanding objects in an image in t f th i tit t tterms of their constituent parts•Analysis of objects into feature components•Characterization of objects in terms of their partsj p•Comparing and identifying objects based on their characterizationsobjects parts description recognitionobjects parts description recognition

2

Page 40: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Vectorized Image Segmentation

Raster image of peppers Image edges Triangulation of edges

Sampling triangle colorsPerceptual grouping oftriangles into polygons

Segmented image

Page 41: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Application to detection of marine fauna (WHOI)

Page 42: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

WHOI: how to tell texture from structure?

In the successive grouping of polygons, texture regions gsegment into polygons with highly wiggly boundaries, whereas objects tend jto have more regular boundaries.

This can be exploited pas a later stage perceptual cue that can distinguish texture from structure.

Page 43: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Then, mine the hierarchy for features

Page 44: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Repeatedly…

Page 45: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Use clever techniques to classify shapes

Page 46: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Scallops are lively little creatures!

QuickTime™ and aYUV420 codec decompressor

are needed to see this picture.

Page 47: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Image analysis: cumulative tracks can show signs of anomalous activities

U-TurnInefficient travel

routes and loitering

Correlated stops?

may imply non-transit activities

Suspect Location: pSafe house?

SchoolStrip Mall

Anomalous activity detection requires:• Continuous tracking of all movers• Characterization of “Normal” activity• Identification of hostile activity characteristics• Identification of hostile activity characteristics• Sophisticated activity modeling and recognition

algorithms• Fast processing

Page 48: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Petascale Synthetic Visual Cognition for Remote SensingSteven Brumby ISR-2 Garrett Kenyon P-21 Sriram Swaminarayan CCS-1 John Galbraith P-21 Kim Edlund HPC-5

DORSAL PATHWAY (WHERE?)

FRONTAL CORTEX

QuickTime™ and a decompressor

are needed to see this picture.

VISUAL CORTEX~10 billion

FRONTAL CORTEX (REASONING)

VENTRAL PATHWAY (WHAT?)

neurons

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

RETINA~ 6 Mpixel imager

Goal: Develop biologically inspired image processing algorithms forGoal: Develop biologically-inspired image processing algorithms forunique petascale computing (LANL Roadrunner >1 quadrillion FLOPS).

Page 49: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

LANL Roadrunner

The Hardware is Available Now

Xbox 360Intel Xenon Processor

LANL RoadrunnerIBM Cell processor

Hans Moravec, “When will computer hardware match the human brain?”, J. Evolution &Technology, 1998.

LANL Roadrunner exceeds estimates of full human brain processing requirements (1011 neurons, 104 synapses/neuron). Algorithms and software that can match human performance are the critical issues.

Page 50: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

level of biological detail →Our focus is on functional models of the cortex

Spiking DynamicsBiochemical ModelsBlue Brain (EPFL)

Existing ModelsFukushima Neocognitron

Poggio MIT modelsLeCun Deep networks

ΔWij

Feedback Loops

Synaptic Plasticity

ΔWij

ΔWij

Next generation Artificial Neural Networks: Our focus is on functional models of cortex based on recent breakthroughs in neuroscience. Traditional ANNs ignoreof cortex based on recent breakthroughs in neuroscience. Traditional ANNs ignore spiking dynamics, feedback loops, and synaptic plasticity. We will explore the functional value of these processes, without trying to model full biochemistry.

Page 51: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Modeling and Visualizing Primary Visual Cortex (region V1)

I t IOriented Gabor-filter prototypes for simple cells i V1 l 4Input Image in V1 layer 4

Responses of retinatopic arrays of Gabor-tuned simple cells to the input image

Active complex cells in output layer of primary visual cortex V1 layer 2/3

Page 52: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Automated Scene Classification with Machine Learning: translation of

expert knowledge into automated knowledge extractionGENIE i i l f “GEN ti I E l it ti ”GENIE: original acronym from “GENetic Imagery Exploitation”

Iterative algorithm improvement using

Image Data Classifier Classification

Result

improvement using biomorphic strategies

Ground Truth

CompareModify Classifier

Training Exploitationg Exploitation

Input Image

Training Image

Test Image Output of GENIEPOC: Neal R. Harvey [email protected]

Page 53: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

So much Data, So Little Information

• Satellite-based and other instrumentation todayinstrumentation today produces unprecedented quantities of raw image and signal data.signal data.

• Hidden in this data is information of interest to analysts and scientists.analysts and scientists.

• How can this information be extracted:– EasilyEasily– Rapidly– Reliably

Page 54: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

GENIE: Machine Learning

Easier to Easier to showshow a machine a machine

what to find…what to find…

...than to ...than to telltell a machine a machine

how to find ithow to find ithow to find ithow to find it

GENIEGENIE automaticallyautomaticallyGENIEGENIE automatically automatically generates an algorithm generates an algorithm for future usefor future use ExploitExploit

TrainTrain

Page 55: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Evolving Solutions• GENIE is an Adaptive System:

f f• It derives a general purpose image classifier from a limited set of user-supplied examples.

• It uses a hybrid genetic algorithm, combining evolutionary exploration with statistical machine learninglearning.

Page 56: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Issues in Pixel Classification

• Spectral information often inadequate.p q• Need to make use of textural and spatial context cues.• Many, many ways of describing/encoding such spatial

context information.• Best techniques are task-specific.

How do we do learn to map pixels to categories in• How do we do learn to map pixels to categories in general?

Page 57: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

The GENIE Approach

• Give GENIE a large and flexible “toolbox” of image g gprocessing algorithms.

• Use an evolutionary algorithm to explore which tools t i t f th t t kare most appropriate for the current task.

• Use statistical machine learning to learn how to combine those tools together to give an accuratecombine those tools together to give an accurate classification.

Page 58: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Genie Pro Architecture

GrayscaleMorphologyOperations

Spectral / Texture

OperationsCombination

Function

I iti l ClRaw Image

Spectral / Texture

Attributes

Initial Class Probabilities

CombinationFunction

Morphological Attributes

Class LabelsUser Markup

Assess performanceAssess performance

Page 59: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

Spectral / Textural / Morphological Attributes

Attributes used for classification

NormDiff StdDevAbsDiff Gabor

Smooth NormDiff

d b dRaw data bands

Page 60: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

GENIE Development1999: Initial funding from two NRO DIIs

Continued research funding from LANL, DOE and othersand others

2002: R&D 100 Award

2003: Transition to NGA funding for operational version: Genie ProGenie Pro.

2004: Genie Pro wins NGA Feature Extraction Evaluation (“bake-off”)

Page 61: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

GENIE Licensing

• Licensed to Observera of Chantilly, VA for remote sensing applications

• http://www.observera.com

Licensed to Aperio of Vista CA for bio medical• Licensed to Aperio of Vista, CA for bio-medical applications

• http://www.aperio.comp p

Page 62: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

GENIE Results: Cover of Laboratory I ti tiInvestigation

• “When tested on urothelial t l s i s ll t d t cytology specimens collected at

two separate institutions over a span of 4 years, GENIE showed a combined sensitivity and

ifi it f 85 d 95% specificity of 85 and 95%, respectively. Of particular note is that when ‘training’ was performed on cases initially di d ‘ i l’ diagnosed as ‘equivocal’ on cytology but with follow-up biopsy, surgical specimen or cytology, which was unequivocally b i li t GENIE benign or malignant, GENIE was superior to the cytopathologist interpreting the initial ‘equivocal’ cytology specimen.”

Page 63: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under

New Solutions Come from Our Wild Ideas to Create Useful Applications

National NeedDetect nuclear explosions in the

atmosphere and spaceeverywhere all the time

100 km100 km

30 km30 km

• Gamma Rays• Neutrons• X-rays• EMP

New Science ContributionsModels of human visual cortex

Lightning scienceIonospheric physics -everywhere, all the time

-Detect proliferation activities EMPIonospheric physics

Meteorites and meteoridsSensor characterization

Existing and Emerging S&TSatellite Instrumentation

EMP sensorsHyperspectral instruments

SolutionsTriggering and analysis codesImaging for homeland security

New detectors

Atmospheric modellingMachine learning

21 Sep 05 16:00-17:00 UTC

T h i i dj V S

Active complex cells in output layer of primary visual cortex

New Capabilities

Technicians adjust a V-Sensor antenna on a GPS space vehicle

Rita at category 5: intense lightning marks boundary of eyewall - EdotX sensors

Compact, innovative sensorsEnd-to-end system modelingAdvanced data exploitation

Anomaly and Change DetectionLightning Studies

Page 64: Los Alamos National Laboratory From Sensing to Information ...people.physics.tamu.edu/kamon/research/talk/pac_lunch/2009/090409... · From Sensing to Information: Everything Under