sensor models - university of arizonadial/ece531/sensor_models.pdf2 sensor models 3 fall 2005...

1

ECE/OPTI 531 – Image Processing Lab for Remote Sensing Fall 2005

Sensor Models

Reading: Chapter 3

Fall 2005Sensor Models 2

Sensor Models

• LSI System Model• Spatial Response• Spectral Response• Signal Amplification, Sampling, and

Quantization• Simplified Sensor Model• Geometric Distortion

2


Overall Sensor Model

• Remote sensors are complex systems of optical,mechanical and electronic components– These components determine the quality of the data

from the sensor– The sensor may be considered a “black-box” that

converts at-sensor radiance to DNs


LSI System Model

• Model the various systems as Linear Shift-Invariant (LSI)– A linear transformation of the input x results in a

similar transformation of the output y• Superposition principle• If T[f1] = g1 and T[f2] = g2 , then T[a1f1 + a2f2] = a1g1 + a2g2

– Shifting the input results in a similar shift of theoutput

• Shift invariance• If T[f(x)] = g(x) , then T[f(x-x0)] = g(x-x0)

• LSI model is generally applicable over thenominal range of operation for these systems– Model will break down as performance limits are

approached (i.e., system response becomes non-linear)

3


Instrument Response

• Any signal can be written as a sum of weighteddelta functions using the “sifting” property

• What happens when this input form is put intoa linear system?

• Knowing the transformation of a deltafunction, the “impulse response”, completelycharacterizes the LSI system


Instrument Response (cont.)

• The precision of measurement is determined by theinstrument response, r

• The transformation from the input physical quantity tothe measurement is described mathematically by aconvolution

– where• i(α) is the input signal, a function of time, space, etc.• r(z0-α) is the instrument response, inverted and shifted by z0

• o(z0) is the output signal at z = z0

• W is the range over which the instrument response is significant• Shorthand notation, o(z) = i(z) ∗ r(z), read as “the output

signal is the input signal convolved with the instrumentresponse.”

4


Input and Impulse Response

Convolution Operation

Output

1-D Convolution Example

• The measured value at z0 is anaverage of the input signal inthe vicinity of z0, weighted overthe range W by the instrumentresponse


Resolution

• Any instrument that measures a physicalquantity is limited in the amount of detail itcan capture– This limit is referred to as the instrument’s “resolution”– “Resolution” is a term that is widely used, but often

misunderstood• The width W of the instrument response defines

the spatial resolution, or effective GIFOV

5


Sensor Models




Spatial Response

• The spectral signal is convolved with the sensorspatial response

– where the spatial response of an imaging system isnow called the Point Spread Function (PSF)

• The net sensor PSF is a convolution of individualresponses from:– optics PSFopt

– image motion PSFIM

– detector PSFdet (defines the geometrical GIFOV)– electronics PSFel

6


PSF Properties

• The net sensor PSF is wider than the GIFOVbecause of– PSFopt in both directions– PSFIM

• cross-track for whiskbroom scanners• in-track for pushbroom scanners

– PSFel cross-track for whiskbroom scanners• Reasonable assumption in many cases is that

the PSF is separable in the cross-track and in-track directions


PSF Comparison

0

10

20

30

40

0

10

20

30

40

0

0.2

0.4

0.6

0.8

1

x 10-3

0

10

20

30

40

0

10

20

30

40

-2

0

2

4

6

8

x 10-4

0

10

20

30

40

0

10

20

30

40-1

0

1

2

3

4

5

6

7

x 10-4

0

10

20

30

40

0

10

20

30

40-2

0

2

4

6

8

x 10-4

0

10

20

30

40

0

10

20

30

40

0

0.2

0.4

0.6

0.8

1

normalized detector PSFdet

AVHRR MSS

SPOT HRV TM

cross-trackin-track

significant “out-of-pixel” response in all four systems

7


Sensor Models




Spectral Response

• The at-sensor radiance is transferred to thesensor image plane by the camera equation

– where• τo(λ) is the optics spectral transmittance• N is the optics f-number, given by the ratio of the optical

focal length divided by the aperture stop diameter• the optical magnification is assumed to be one

• The spectral responsivity Rb(λ) weights the imageplane irradiance to yield a signal value

8


Spectral Response (cont.)

• Spectral responsivities are not ideal rectangularbands (Fig. 3–8)

0

0.2

0.4

0.6

0.8

1

400 500 600 700 800 900 1000 1100

Rel

ativ

e R

esponse

wavelength (nm)

AVHRR1 AVHRR2

0

0.2

0.4

0.6

0.8

1

400 500 600 700 800 900 1000 1100

Rel

ativ

e R

esponse

wavelength (nm)

MSS1 MSS2 MSS3 MSS4

0

0.2

0.4

0.6

0.8

1

400 500 600 700 800 900 1000 1100

Rel

ativ

e R

esponse

wavelength (nm)

TM1 TM2 TM3 TM4

0

0.2

0.4

0.6

0.8

1

400 500 600 700 800 900 1000 1100

Rel

ativ

e R

esponse

wavelength (nm)

SPOT1

SPOT2

SPOT3

0

0.2

0.4

0.6

0.8

1

2.5 4.5 6.5 8.5 10.5 12.5 14.5

Rel

ativ

e R

esponse

wavelength (µm)

AVHRR3 AVHRR4 AVHRR5

0

0.2

0.4

0.6

0.8

1

1200 1400 1600 1800 2000 2200 2400

Rel

ativ

e R

esponse

wavelength (nm)

TM5 TM7

AVHRR

TM

SPOT MSS


Hyperspectral Responsivity

• Hyperspectral sensors have relatively constantspectral resolution across a wide spectral range

band center wavelengths linear with

band number

0

500

1000

1500

2000

2500

8.5

9

9.5

10

10.5

11

11.5

12

1 51 101 151 201

center wavelength

bandwidth

cente

r w

avel

ength

(nm

)ban

dw

idth

(nm

)

band

average bandwidth

about 10nm

Spectral band properties for AVIRIS (Fig. 3–9)

9


Sensor Models




Signal Amplification

• The electronic signal produced by the detectorsis amplified

• Some sensors have multiple gain settings, e.g.SPOT HRV (Chavez, 1989) and ETM thermal band,to increase signal level for dark objects

signal rangerequired at

A/D input for fullrange DN output

ab

eb

anticipated range

optional high gainb

offsetb

all scenes

standard gainb

of detected signals low radiance scenes

Linear amplification characteristics (Fig. 3–17)

10


Sampling and Quantization

• The amplified signal is sampled in time (duringscan) and quantized into Digital Numbers (DNs)

DN

0

2Q – 1

input signal to A/D converter

ab

• Quantization is a low-levelnoise superimposed on thedata values– For Q bits/pixel quantization

there are 2Q integer DNs overthe range [0...2Q-1]

• Radiometric resolution = 2-Q

Linear quantization transfercharacteristics (Fig. 3–18)


Sensor Models



11


Effective Sensor Model

• Total measured signal at pixel p in band b

– where• DNpb is the Digital Number at pixel p in band b• Lλ(x,y) is the at-sensor spectral radiance from scene

location (x,y)• Kb is a gain coefficient for band b that includes sensor

gain, detector spectral responsivity and spectral filtertransmittance

• offsetb is the sensor offset coefficient for band b• the gain and offset are effective quantities, averaged

over an effective spectral band


Gain-Offset Model

• The three integrals are over:– the effective spectral response range of band b

(spectral resolution)– the effective spatial response range in-track and cross-

track (spatial resolution)• Assume a band- and space-integrated at-sensor

radiance Lpb at pixel p, band b. Then,

– DNs are linearly proportional to the total at-sensorradiance

– Ignores radiometric quantization and nonuniformresponse within spectral bands and the GIFOV

– Simplifies modeling and radiometric calibration of thesensor

12


Sensor Models




Sources of Distortion

• All remote sensing images are distorted relativeto a map– platform motion, especially airborne sensors– scanning distortion of the Ground Sample Interval (GSI)– topography Distortion caused by airplane

motion (ASAS airborne sensor)

cross-track ground distance from nadir (km)in-track

ground distance (km)

01.12.2

3.34.4

5.5

0 200 430 750 1400

1024 20481280 1536 1792

cross-track pixel numberorbital track

Bow-tie distortion in AVHRR data (Fig. 3–23)

13


Scanning Distortion

0

1

2

3

4

5

6

7

8

0 0.2 0.4 0.6 0.8 1

flat earthtrue earth

rela

tive

GS

I

off-nadir scan angle (radians)

!

FOV

nadir

H

"!

f

GSIe(!)

GSIf (!)

line and whiskbroom scanners (Fig. 3–22)

0.5

1

1.5

2

2.5

0 0.2 0.4 0.6 0.8 1

flat earthtrue earth

rela

tive

GS

I

off-nadir view angle (radians)

!

FOV

nadir

H

"!

W

f

GSIe(!)

GSIf (!)

pushbroom scanners (Fig. 3–24)


Topographic Relief

• Image offset proportional to elevation abovebase plane, or “datum”

• Stereo pair of images can be used to findelevation

• Imagery corrected for topographic distortion iscalled “orthographic”

AA0

ground point at A actually appears to

come from A0

because of topography

sensor models - university of arizonadial/ece531/sensor_models.pdf2 sensor models 3 fall 2005...

Documents