Sensor Modeling in DIRSIG June 10, 2004

1
Sensor Modeling in DIRSIGJune 10, 2004

• Cindy Scigaj
• Dr. John Schott
• Scott Brown Dr. Bob Kremens Dr.Carl Salvaggio
• Paul Lee Jason Faulring
• Niek Sanders

2
Overview

• What is DIRSIG?
• Project definition
• Background information
• MTF, Spectral Response ( spectral smile), Noise
• Lab experiments
• Field experiments
• Summary

3
DIRSIG

• Physics based model developed at RIT to simulated
  remotely sensed data
• Various platforms
• Line scanner, framing array, pushbroom scanner

DIRSIG Megascene Image
4
Project Definition

• Historical perspective and justification
• Users wishing to incorporate rigorous sensor
  modeling to DIRSIG simulations needed to
  oversample the image spatially and spectrally.
• Large intermediate images were required
• Project Goals
• Add a flexible sensor model that allows users to
  incorporate sensor models during rendering.
• Vary properties for each detector element (pixel)
• Create and distribute pre-built sensor models for
  out-of-the-box use.
• Potential systems AVIRIS, WASP, HYDICE, SEBASS,
  COMPASS, NVIS
• Allow users to compare results with these
  standardized models
• Combine efforts to create sensor model
  cook-book
• Benefit algorithm developers/testers and
  instrument designers
• Long term handle tabulated data
• Overall provide an easy way to incorporate
  sensor artifacts

5
Implementation Overview

• Response function
• A set of one or more channel or band responses
• Different types of channels
• Pass-band channels have a tabulated spectral
  response.
• Spectrometers channels have a center, width and
  shape.
• Both have gain and bias terms per channel
• Dead detectors introduced with zero gain values
• Spectral Response Function
• Pushbroom spectrometer specific issues
• Smile and frown effects can vary channel
  locations for each spatial detector.
• Point-Spread Function (MTF)
• A combination of atmosphere (turbulence),
  platform (jitter), optics, detector and
  electronics effects.
• Ideally, a series of PSFs stored in a functional
  form to ease computation and allow for different
  sub-detector sampling schemes.
• Noise
• A combination of photon arrival, detector
  read-out and electronics.
• Store the noise covariance and use a Principle
  Component (PC) synthesis method to compute a
  unique noise spectrum for each scan/read-out of
  the detector elements.

6
Pushbroom Scanners

• AIS (grating)
• HYDICE (prism)
• SEBASS (prism)
• Hyperion (EO-1)

7
Response FunctionSpectral Smile and Frown

• In pushbroom spectrometers, the spectral channel
  locations of pixels on the edge of the focal
  plane are different than the ones in the center
  of the focal plane
• This effect can be modeled with these
  enhancements.

8
Point-Spread Function

• A spectrally dependent function that introduces
  the blur from a variety of sources in the image
  formation process.
• Atmospheric turbulence, platform jitter, optics,
  detector and electronics effects.
• Ideally, each of these would be in a flexible,
  functional form.
• Radially symmetric (Gaussian, Lorentzian, etc.)
• X/Y separable (Gaussian, Sinc, Sinc2, etc.)
• A functional form could allow the user to change
  the sub-detector sampling (finer grid, N random
  locations, etc.)

PSF PSFa(r,l) PSFp(r) PSFo(r) PSFd(x,y)
Cn2
jitter
optics
detector
Pixel Detector
9
NoiseSpectral Structure

• We want to introduce noise for each detector
  element and each scan
• Each detector can have unique noise statistics
• Non-repeating noise that can be described by
  higher order statistics (spectral
  covariance/correlation)
• Due to focal plane design (shared electronics)
  the sensor noise is correlated.
• This effect can be modeled with these
  enhancements.

1
2
3
4
AVIRIS Noise Correlation
10
Modeling Concept

• To accurately model the radiance at this pixel
  detector
• Spectrally oversample
• Convolve to channel resolution using response.
• Spatially oversample inside and outside the
  physical detector element
• Convolve to detector resolution using PSF.
• Add spectrally correlated noise
• Pull from statistical noise model.

Many spatial samples
Focal Plane
Many spectral samples
PSF Contribution Region
Projected Detector Extent
11
Data Flow
PSF (by channel)
Noise (by channel)
Channel Responses
Apply Response
N Spatial Samples Many Spectral Samples Without
Noise
N Spatial Samples M Spectral Samples Without Noise
1 Spatial Sample M Spectral Samples Without Noise
1 Spatial Sample M Spectral Samples With Noise
12
Wildfire Airborne Sensor Program(WASP)

• System in development at RIT
• Four separate camera systems
• Terra Pix 0.4-0.9 microns
• SWIR 0.9-1.8 microns
• MWIR 3.0-5.0 microns
• LWIR 8.0-9.2 microns
• Characterization
• equipment specifications
• actual lab scenes and measurements
• Bayer pattern artifacts

WASP
13
WASP Modeling Criterion
14
Lab Targets
15
Field Experiments 06/07/04

• Data collected on scene
• GPS location
• ASD radiance
• ASD reflectance
• DP radiance
• Temperature
• Thermocouples
• Exergen
• Thermistors _at_ pier

16
Summary

• Sensor modeling status
• A survey of different sensors has been performed
  to gather information on popular imaging
  platforms.
• This information will be used to construct
  distributed sensor models.
• Recently became in contact with COMPASS
  instrument group
• Precal data and calibration documentation
• Correlated noise has been demonstrated
• Simple smile case has been demonstrated
• Goal supply significantly better default sensor
  models
• These new features, user interfaces and pre-built
  sensor models will be in DIRSIG 4
• Experiments status
• WASP Lab data acquisition complete
• Image analysis programs being developed
• using ISO standard procedures
• WASP imagery with supporting ground truth
  measurements have been acquired and need to be
  sorted and organized
• Waiting for SEBASS and COMPASS data distribution
  30 days?
• Pre-calibrated data as well

17
Contact info

• WASP Terra Pix _at_ 3,000 ft
• Cindy Scigaj
• cls8343_at_cis.rit.edu