N E A Full Optical Spectrum Hyperspectral Scene Simulation · Smp V b Image SampleLWIR Image...

Post on 03-Feb-2020

1 views 0 download

transcript

Motivation for Work

• Develop VIS-LWIR scene simulation– Realistic levels of clutter

•Spectral•Spatial

– Easy user control of terrain materials•Reflectance in Visible-MWIR (NASA SBIR)•Emissivity/Temperature in MWIR-LWIR•Statistical terrain definition

– Incorporate digital elevation maps– 3D variable atmosphere– 3D complex cloud fields– 3D faceted objects

• Simulation tool used to test algorithms– Atmospheric compensation– Object detection– Scene classification/segmentation– Sensor development

• Simulations to supplement data collects– Data usually collected on high visibility days– Simulate low visibility & bad weather

Full Optical Spectrum Hyperspectral Scene SimulationRobert Sundberg and Steven RichtsmeierSpectral Sciences, Inc., Burlington, MA

Raymond HarenAFRL/SNJT, Wright-Patterson AFB, OH

50 km uniform boundary

15-50 km100 m layers

0-15 km(100 m) voxels3

(1 m) pixels2

100 m

Atm

osp

heric

pro

fi le

Exit toSpace

BackwardPhoton

Trajectory

AtmosphericScattering

SurfaceScattering

Sum SolarScatterProbabilities

• Unified UV to LWIR RT• 3D atmospheres with complex clouds• 3D surfaces with BRDF’s (DEM)• 3D object insertion• 3D plume model• MODTRAN optical properties, profiles, and spectroscopy• Extended to underwater imaging• Active illumination sources (LIDAR)

MCScene Spectral Scene SimulationValidation

MCScene Summary

• New 3D code for computing visible to LWIR HSI/MSI data cubes• Spectral/spatial clutter with statistical behavior similar to nature• Applicable to wide variety of remote sensing problems• MCScene team

– SBIR & STTR funding•AFRL/SNJT phase I/II/II+•NIMA phase I/II•NASA phase I•US Army phase I

– Los Alamos National Lab

– IRAD funding•Spectral Sciences, Inc.

– Industrial partners•Kodak•Research Systems, Inc.•Rochester Institute ofTechnology

•Direct sun-ground-sensor termcalculated explicitly

•Weighting factor for molecularabsorption

• Integrate thru 3D world in steps•Sample probability distributionsfor distance traveled andscattering directions

•Atmospheric scattering (1st

order)

ω, Single scattering albedo P(θs), Scattering phase function

0.2 0.4 0.6 0.8 1.0 1.2 1.40.00

0.05

0.4

0.6

0.8

Ground Visibility5km

|10km

|23km

|

50km

|100km

|

!gnd

= 0

!gnd

= 1 16-STR DISORT

DSMC

MS

Appare

ntR

eflecta

nce

("Im

s/µ

F)

0.55µ Vertical Optical Depth

0.0 0.2 0.4 0.6 0.8 1.00.250

0.275

0.300

0.325

0.600

0.625

0.650

0.675

# = 0.800

#= 0.975

EXACT

104 PHOTONS

106 PHOTONS

Ap

pa

ren

tR

efle

cta

nce

("I

/µF

)

Cosine Solar Zenith (µ)

500 1000 1500 2000 2500-0.02

0.00

0.02

0.04

0.2

0.4

0.6

0.8

Multiple Scattering Residuals

Multiple

Scattering

Only

Total

MODTRAN4 w/ DISORT16

MOD4 w/ Cor- & DISORT 16k

Monte-Carlo

500 1000 1500 2000 2500-0.02

0.00

0.02

0.04

0.2

0.4

0.6

0.8!gnd=1

AP

PA

RE

NT

RE

FL

EC

TA

NC

E

WAVELENGTH (nm)

500 1000 1500 2000 2500-0.006

-0.003

0.000

0.003

0.05

0.10

0.15

0.20

0.25

Multiple Scattering Residual

Multiple Scattering Only

Single + Multiple Scattering

500 1000 1500 2000 2500-0.006

-0.003

0.000

0.003

0.05

0.10

0.15

0.20

0.25!gnd=0

AP

PA

RE

NT

RE

FL

EC

TA

NC

E

WAVELENGTH (nm)

MODTRAN4 w/ DISORT16

Monte-Carlo

Cloud Scattering Details

Data Driven Scene Simulation

Download DEM

RadianceCube

Correctwith FLAASH Deshadow

Register DEM& source

reflectanceMCScene

Pre-calculateMODTRAN

atmospheres

SpecifySun andSensor

GenerateCloudField

Simulation Flow Chart

Radiance Reflectance

ShadowMap

De-shadowedRadiance

• Atmospheric Compensation–FLAASH–ELM

10 m USGSDEM

Unit ReflectanceDEM Image

Sample Visible Image

Sample LWIR Image

• AVIRIS Scene– Virgin Mts., NV– 20 km nadir view– Solar geometry

•28° Zenith•317° Azimuth

– Composition• Vegetation• Minerals/Soils

Simple 3D Cloud Simulations

Self Shadow

Major Features

1 km

Cloud Over DarkVegetation

AtmosphericShadow

GroundShadow

200

m

τc=5

0 100 200 300 400 5000.0

0.2

0.4

0.6

0.8

1.0

Cloud

Self Shadow Cloud Shadow

Soil

Cloud 0.65µm

0.55µm

0.44µm

Radia

nce

(µW

cm

-2sr-1

/cm

-1)

Image Diagonal Pixel

24500 24600 24700 24800 24900 250000.0

0.1

0.2

0.3

0.4

0.5

0.6

Shadow

Line

Cloud

0.1km ThickRegion

0.2km ThickRegion

CLOUD BASE

7.8km

4.8km

1.8km

Appare

nt

Reflecta

nce

(!I

/µsun

F sun

)

Pixel (meters)

• AVIRIS source reflectance– Atmospheric compensation– De-shadowed

• 10 m USGS DEM• CSSM for cloud field

– Nimbostratus cloud– 20 m resolution

Realistic Scene Generation

Initialize photonposition and

view direction

Trace path optical depth,

! "= - ln

Does pathterminate within the

atmosphere?

Does pathterminate

at the ground?

Photon exit. PhotonNext

Sum solar scattersource term

& updatephoton weight

Sum solar scattersource term

& updatephoton weight

If trajectory is completed,

PhotonNext

Sample scatteringphase function

for new direction

Sample surfaceBRDF

for new direction

If trajectory is completed,

PhotonNext

Photon Trajectory Flow Chart

• Solar geometry:– SE, 44 º zenith angle

• Nadir sensor– 20 km altitude, 17 m GSD

• Atmosphere– Mid-latitude summer– Desert aerosol (50 km Vis)

• Scattering albedo weighted phase function

– Use MODTRAN3 form for Rayleigh phase function

• PR ay (θ s ) = 0.06050402 + 0.0572197 cos2θ s• Rayleigh function is sampled isotropically with a small anisotropy correction

• Aerosol and cloud scattering phase functions use H enyey-Greenstein

– Spectral and altitude dependent asymmetry factors, g

• The H-G function is sampled analytically with

w here α is a random number between -1 and 1

( ) ( ) ( )scldcldsaeraersRayRay PPPP !"!"!"" ++#

( )232

2

)cos21(4

1;

s

sGHgg

ggP

!"!

#+

#=#

( )2

2

)1(2

)2(

)1(22cos

g

gg

g

gGHs

!

!

!

!"

+

#++

++=

#

Phase Function Form & Sampling

( ) dTTP suns

Tsun

a

g

!"µ

#$ %=

1