NTRS - NASA

0 downloads 0 Views 1MB Size Report
produced by a remote sensor system and they all affect the performance of a data classification ..... _To access data values at spectral grid points required.

NASA Contractor

Report

172393

P

NASA-CR-172393 19840024713

A Simulation

of Remote

Sensor

Systems and Data Processing

Algorithms for Spectral Feature Classification

R.

F.

Arduini,

R.

M. Aherron,

and

R.

W. Samms

Information & Control Systems, Incorporated Hampton VA 23666 :

Contract

NASI-16870

_u,_ ,_.

'-

N/ A

National Aeronautics and Space Administration Langley Research Center

Hampton, Virginia23665

LIBRfI, tlY_,OP\' S!:P2 _' LANGLEY RESEARCH CENTER LIBRARY, NASA I..IA_P_[O_N= VIRGIN_IA

.-....,

1 RN/NASA-CR-17239J

L.~

nlCDt AI:.! U.J..Jl

!{TTl"= ~~:

I L-*

I-.n.

· l ...1 ....~ .... ~ oe--rr £....~ .I

84N32784**

iSSUE 22

LlrJCLASS IF I ED

DOCtJr~ErJT

:\. n

,-. -i ~

U"..; ;.

1. -,. ~ -;. .-.....

I 1IIU • C

... ~,.-#.•-..+

;,0:-,

~r-"~'-~'

0

l. t '",."1 I

:.}f

.....

OOT*~

:.......

.--" :J..,.

f&mot2

'!. I

AUTH; A/ARDUINI, CORP: information dnu Control

r-. 1'..

M

=

1'1.,

SYst~ms,

He AE:4 ,.. . MF AG1

RHJS; /*CLASSIFICATIONS/*DATA PROCESSING/*MULTISPECTRAL 8AND SCANNERS/* PERFORMANCE-PREDICTION;*REMOTE SENSORS/*SlMULATiON/*SPECTRAL REFLECTANCE ALGORl~YMSi ATMOSPHERIC EFFECTS.I PROGRAMS/. INSOLATION .:'~O,;:'\_=­ nun~

A

COffiPutdtioudl mod21 of the

d~t~rm~n;stlc

dnU stochdgtic

PfOC2££e£

invoiv~d in ffiultisp~ctral r~mote S~nSlng llias d~signed to eVcilUcite th~ Pf:r fOfffidf:Cf: of Sf:f"iSOr S1--i Stf?ff:S'- drid data Ptoc~ssirlg cilQori tilifiS fOr SPf=ctt":cil ff=aturf?CidSsificat~0fi. HCCUfdC1r: ir·i distirJg:~i-s~iir·ig uf::ttlieef. cat:2gofi~8 of .-... ,,-Vllir-CI .-.•-.." . :"-.. - .. -= C.-.. d ffif?ciflS ~t. ...... 7

r ...-.....- ....-·.-.. -.::: O"='II~·..,~I

-be madE of the

ci; gOt": 1 ti~iffiS. Ti~i;? ffiO:j=2 i of varlciuiiitvof th& atffi0sph~f~

~ff~cts rf?fi~ctdftC:&7 ciS uif:i-i as

of

tii&

f=ff~cts

d

.-.. i lotii8 stud; =28 +t. .......

-~.-...-:!

C:U I U

- .. ~ ••-.!.

CJIU

ABSTRACT

A computational in multispectral sensor

systems

Accuracy

model

remote

of the deterministic

sensing

has been

and data processing

in distinguishing

between

categories

Of surfaces

cessing

algorithms.

ability

of the atmosphere

channel

selection

allows

for comparing studies

and of surface noise.

to evaluate

for spectral

types is used as a criterion

and sensor

developed

algorithms

surface

The model

and stochastic

sensor

to be made

reflectance,

Examples

i

of these

processes

the performance

feature

systems

specific

and data pro-

of the effects

effects

of

classification.

or between

as well

involved

of vari-

as the effects are Shown.

of

TABLE

OF CONTENTS

page ABSTRACT LIST

...........................

OF TABLES

"

"

...........................

iv

LIST OF FIGURES .......................... I. II.

INTRODUCTION REMOTE A.

B.

SIGNAL

MODEL

3

GENERATION ....................

3

i.

SOLAR

2.

ATMOSPHERIC

3.

MULTIPLE

SCATTERING .................

6

4.

SURFACE

REFLECTANCE ..................

7

5.

SIGNAL

6.

PROGRAM

IRRADIANCE

..................

TRANSMITTANCE

COMPUTATION STRUCTURE

SIGNAL PROCESSING.

............

OF STOCHASTIC

2.

DECISION

RULES

BOUNDARY

b.

MEAN

SAMPLE

SQUARE

5

i0 ii

• ......

PROCESSES

12

..........

13

..................

APPROXIMATION

IMPLEMENTATION

DISTANCE

•. 15

METHOD .......... AND MAXIMUM

15

LIKELIHOOD

.................

OF DECISION

RULES

16

.........

18

BNDARY

b.

RPLPROG .....................

19

c.

CLASIFY .....................

20

APPLICATIONS REMARKS

.................



a.

CONCLUDING

REFERENCES

.............

FOR STIMULA ............

ANALYSIS

a.

5

.................

i.

3.

IV.

i

....................

CLASSIFICATION

III.

v

........................

SENSING

i

....

19

.....................

21

.....................

22

............................. iii

23

_LIST OF TABLES

page TABLE i.

SUMMARY

OF PARAMETERS

TABLE

TARGETS

AND THE ASSUMED

TABLE

2.

3.

USED

IN SIMULATION ........

STANDARD

REFLECTANCE

......................

SENSITIVITY

CHARACTERISTICS

_v

DEVIATION

OF THEMATIC

25

OF THEIR 26

MAPPER ......

28

LIST OF FIGURES

page FIGURE

i.

SCHEMATIC

OF SIGNAL GENERATION

MODEL

FIGURE

2.

SCHEMATIC

OF SIGNAL

OPTIONS

FIGURE

3.

SOLAR IRRADIANCE

FIGURE

4.

MEAN

FIGURE

5.

SIMULATED

SPECTRAL

PROCESSING

......... ........

6.

MEAN

REFLECTANCES

SPECTRAL

SPECTRAL

RADIANCES

31

..............

REFLECTANCE

TWO SOLAR INCIDENCE RANGES,

30

AT TOP OF ATMOSPHERE .........

32

VARIABILITY

FOR TWO

TARGETS .................. FIGURE

29

• ..... INCIDENT

ANGLES,

ON REMOTE

e o , AND THREE

V, USING OATS AS TARGET

SENSOR

34 FOR

VISUAL

AND BARE MOIST

SOIL

AS BACKGROUND ................ FIGURE

7.

TYPICAL

REALIZATIONS

FOR TWO VIEWING BARE MOIST 8.

SIMULATED

FIGURE

9.

STIMULAPROGRAM

FIGURE

i0.

FIGURE

Ii.

SIGNAL

CONDITIONS,

SPECTRAL

SENSOR

PLOT

COVARIANCE

AND BARE MOIST 12.

TM SIGNAL

USING

VARIABILITY

OATS AS TARGET

AND

FOR TWO TM CHANNELS,

37

OATS AS TARGET

FOR THREE

LISTED

IN TABLE

SOIL AS BACKGROUND

WITH

38 USING

SOIL AS BACKGROUND

CONTOURS

HISTOGRAM

DISTRIBUTION

36

RESPONSES ..........

AND BARE MOIST

USING THE SUBSTANCES

FIGURE

RADIANCE

STRUCTURE ...............

SCATTER

AS TARGET

OF SPECTRAL

SOIL AS BACKGROUND ..........

FIGURE

SIGNAL

35

AND BARE MOIST

i

v

.....

39

TM CHANNELS, 2(a) AS TARGETS

..........

AND "EQUIVALENT" EQUAL MEAN

OATS

40

GAUSSIAN

AND VARIANCE,

USING

SOIL AS BACKGROUND.

. 41

LIST

OF FIGURES

(CONCLUDED)

page FIGURE

FIGURE

13.

14.

DISCRIMINATION

ACCURACY

BETWEEN

GROUPS

OF CATEGORIES,

LOCATED

AT 0.67 _m ..................

FEATURE

CATEGORIZATION

CONDITIONS

USING

VERSUS

THRESHOLD USING

ACCURACY

THE TM CHANNELS

BOUNDARY

THE TM CHANNEL 43

FOR SEVERAL LOCATED

IMAGING

AT 0.67,

0.84, and 1.68 _m .................. FIGURE

15.

FEATURE

IDENTIFICATION

CONDITIONS, RESPONSES

16.

TM, SPOT, AND KON

LOCATED

AT NEARLY

THE SAME WAVELENGTHS

MLH OR MSD CLASSIFICATION

SIGNALS

............

FEATURE

IDENTIFICATION USING

WITH EITHER 17.

FEATURE

IMAGING

THE THREE

CONDITIONS

FIGURE

FOR SEVERAL

USING

WITH EITHER

FIGURE

ACCURACY

44

EITHER

........... ACCURACY THREE

TM CHANNELS

FOR SEVERAL

AND TWO IMAGING

IMAGING

OF SIGNALS.

FOR SEVERAL CONDITIONS,

WITH MLH CLASSIFICATION

vi

45

OR FOUR TM CHANNELS

MLH OR MSD CLASSIFICATION

IDENTIFICATIONACCURACY

NOISE LEVELS

OF UNRATIOED

. . 46

SENSOR

USING

FOUR

OF SIGNALS ....

47

I.

The amount is used

of data

for global

spectral

Scanner

reducing

the data

sensors

water

resource

or Thematic

Mapper)

transmission

and processing

enough

could differentiate

At a higher between

different

or clay and sand.

spectral

discrimination

complete

smart sensor

spectral

have

system,

differences

to distinguish ofclouds

types

of vegetation such tasks

to be augmented

however,

classification

(or similarities)

multispectral the data of in-

vegetation

which

from

obscure

the smart or s0il,

the

sensor

e.g.,

it is understood

by other

Multi-

to drastically

only

of classification,

To accomplish

would

One approach

the presence

system which

(such as the Landsat

and transmit

be designed

level

sensor

load is to design

to identify

might

remote

and land use is enormous.

land and to determine

view of the surface.

and wheat

by a multispectral

monitoring

Such a smart sensor

or bare

their

generated

which are "smart"

terest.

INTRODUCTION

oats

that

information

of signals

is fundamental

in a

according

to the smart

to sensor

approach. This sensing sensor

report

process design

describes

which was developed

concepts

The modeling radiance

effort

sensing

was divided

atmospheric

sor sensitivity

model

makes

as a tool in the study

accounting

two major

tasks:

by the sensor

A major

(i) to

of smart

in this effort

of the elements

different

the and

and the accuracy

objective

reflectance

simulate

into a "signal"

for the stochastic

and of surface this model

remote

algorithms.

of the variability

transfer

is what

of the multispectral

so that they may be classified

Thisexplicit

radiative

into

may be measured.

a realistic

process.

to be used

and its conversion

these signals

of their classification to include

model

and data processing

at the satellite

(2) to process

a computational

in the remote

properties

and modeling

from most

was

other

of of sen-

remote sensing

i

models

(Refs _ 1-

7).

This report

will

pies of how the model may be used. followed

by a discussion

implemented, results

including

to show

First

of the signal

classification

the applicability

describe

the model

and give a few exam-

the signal generation

processing algorithms.

of the model.

algorithms

isdescribed,

which

Thi s is followed

have been by a few

II.

The computational

model

R=_MOTE SENSING

to simulate

of multispectral

data encountered

described

The data acquisition

diance,

here.

the transfer

and the sensor's by a model

spectral

attenuation,

surface

noise

and sensor

produced

by a remote

data classification Figure eration able

1 shows

of

signals

and ELLIP

and CLASIFY

A.

satellite

irra-

reflection

may be described

components.

the physical

to the stochastic

the processes sensor

the signals.

Variability

state

nature

of the

of the signal

the performance

included

and Figur_

The model

simulation

in modeling

2 shows

actually

of a

the Signals

the signals.

generated;

These will

the gen-

the options

comprises

of the signal generation

for displaying

to Figure

altitude

sun is incident

is the result

or absorbed

unaltered.

avail-

five programs:

procedures; and BNDARY,

be discussed

that the radiance

of a number

through

by atmospheric

At the surface,

ing to some reflection trip

i it can be seen

of processes.

at the top of the atmosphere.

may be scattered

return

about

surface

is

SCATPLT, RPLPROG,

below.

Generation

Referring

mitted

schematically

for processing

Sisnal

for the solar

elements

and stochastic

surface

process.

for stochastic

HISTPLT,

account

system and they all affect

by the remote

for processing

STIMULA

sensor

of the earth's

of these

uncertainty

all contribute

must

and classification

the atmosphere,

Each

deterministic

in the atmospheric

sensing

process

through

response.

which has both

the acquisition

in remote

of radiation

MODEL

law

In transit constituents

the energy

(which, in general

the atmosphere,

after which

Energy

from

at

the

to the surface,

it

or it may be trans-

is reflected is quite

at the sensor

or absorbed

complex)

it is detected

accord-

and begins

its

by the sensor

and

is converted

into a signal.

The signal

in the j-thchannel

of a J-channel

multispectral

system

is

given by

sj = _ where

L(%)

spectral

L(%)

Sj(%) d%

is the spectral

response

incident

channel.

of the above

conditions

The spectral

radiance

of the j-th

L(%) and the evaluation and atmospheric

(i)

radiance

the sensor

Calculation

integral

is the purpose incident

upon

of the spectral

for a large

of program

on the sensor

and S.(%) J

number

is the

radiance

of surfaces

STIMULA.

can be modeled

as

(Refs

i L = _ (E° T o _o + Ld) OT + L p'

2 - 4)

(2)

where E =E o

o

(%)

- solar

To=To(%,T,_o)

irradiance

- atmospheric

at top of atmosphere

transmittance

along

the incidence

transmittance

along

the exitance

path T=T(%,T,_)

- atmospheric path

Ld=Ld(Eo,%,T,_o,0,0"

)

Lp=Lp(Eo,%,T,_o,_,¢,p,0"

- total downwelllng )

- path radiance

sky radiance

along

the path from surface

to

sensor p=pi%)

- spectral

reflectance

of target

0"=0"(%)

- spectral

reflectance

of background

_o =c°s @o

- where •

The other parameters

are wavelengthS,

@o is solar T

.

optical

,

.,,

zenith .

thickness

"

angle

o ,

T=[(_) of the atmosphere,

and azimuth

angle _ between

the plane

of incidence

of the total radiance

L which

direction

to as the beam radiance

is referred

These processes

given by _=L-Lp. discussed i.

from

and their

the surface

The component

into the viewing

Lb=_(Eo,%,T,Bo,_,_,P,P') implementation

and is

in the model

are

here. Solar

Irradiance

The solar irradiance well known

(compared

it does vary, others

is reflected

and exitance.

at the top of the Earth's

to some of the other

it is assumed

in this system

that its variability

and so is ignored.

from Labs and Neckels

(Ref. 8)

interpolation

to find the Value

is used

processes

is relatively

in thismodel)

is small

The data

and is stored

atmosphere

shown

in a table

of irradiance

and although

compared

to the

in Figure

3 comes

in STIMULA.

between

Linear

wavelength

grid

points. 2.

Atmospheric

Transmittance

The transmittance near

infrared

of the atmosphere

wavelength

regions

scattering,

aerosol

and ozone.

The atmospheric

T

o

and over

=e

extinction

to solar radiation

(.2 - 2 Um) is reduced

and absorption

transmittance

over

by water

in the visible

primarily vapor,

the incidence

by Rayleigh

carbon path

and

dioxide

is given

-Tl_o

the exitance

path by

T = e-'_/IJ

where

T is the optical

viewing

zenith

angles

thickness

and U, U o are the cosines

respectively.

The optical

thickness

of the solar T is given

and

by

n

T(l) =

_.el(X) x. i=1 l

•(3)

by

with _.(%)

being

I

constituent

the spectral

and x. the associated

and ozone.

from the AFGL LOWTRAN To simulate the attenuator and standard

the effects

it is convenient

simulate

random

ator

produces

with

zero mean

attenuator

i

content, for these

variables

enabling

it is not necessary

use of an on-line

in the absorber numbers

and unit variance. the random

water

referred

vapor,

car-

absorbers were

it is assumed With

that each of the attenuator

pseudo-random

+q

The constituents

variability,

(3) are random

Although

variations

amount,

x.l =x.

of atmospheric

o.. i

to assume

thereby

coefficients

the i-th atmospheric

taken

(Ref. 9).

in Eq.

deviation

amount.

total molecular

The spectral

amounts

distribution,

above:

5 model

coefficient_of

attenuator

i

to here are those mentioned bon dioxide

attenuation

having

To simulate

variable

amounts number

(Gaussian)

a particular

x. is computed i

x.I

has a Gaussian generator

The random

a normal

mean

for the simulation,

random

amounts.

a known

that

number

to

gener-

distribution

value

of the

by

0i m

where

q is the random

are inputs 3.

to the program

Multiple

To include decided

The mean values

and are given

x. I and standard

in Table

deviations

O.I

i.

Scattering

the effects

to use the program

2 - 4) and which Igan.

number.

of multiple RADMOD,

was obtained

The program

which

optical

lation.

RADMOD

provides

the capability

by the atmosphere

was developed

from the Environmental

and its auxiliary

required

scattering

depth properties) forms the basis

program

VlSTAU

were adapted

the diffuse

Research

et. al.,

Institute

(used to calculate

to fit the needs

for the radiative

to compute

by Turner,

it was

transfer

radiance

(Refs.

of Michthe

of this simu-

calculations.

It

and the path radiance

(Ld and Lp in Eq.

(2), respectively)

which

result

from multiple

scattering

pro-

cess.

The single scatteringalbedo (which determinesthe relative amount of attenuation due to scatteringalone) is related to the relative humidity and visual range. Variability in the multiple scatteringsimulationthereforewas accomplished by fixing the v_sual range and then varying the relative humidity in the same manner as describedabove for the atmosphericabsorbers. The scatteringphase function used here is one Of several supplied with the RADMOD program by ERIM. All of the simulationsrun thus far have employed a phase functionwith propertiestypical of a weakly absorbing continental aerosol. Data is available to simulate both continentaland marine aerosols with a range of absorbing propertiesfrom no absorptionto "strong" absorption. Table 1 summarizesthe parametersused in the simulation. 4.

Surface Reflectance

Since targets on the surface are to be identifiedand classifiedaccording to their spectral reflectancecharacteristics,it is clearly importantfor the simulated spectral reflectancesand their variabilitiesto be realistic. The reflectancepropertiesof natural surfaces are not only a function of wavelength but also of many other factors such as the direction of both the incident and reflectedrays, the moisture content for soils and the stage of growth for vegetation. These factors plus the difficultiesinvolved in making measurements of natural surfacesmake it very difficult to assemble a representative collectionof spectral reflectancedata. It is particularlydifficult to find informationabout the typicalvariabilityof spectral reflectances (exceptfor some vegetation)and about their spatial distributionor probabilityof occurrence. Furthermore,it is awkward to deal with spectral reflectancedata that

often cover

only part of the wavelength

The spectral

reflectance

(a) one for vegetation, region,

and

bare land,

(b) an expanded

0.4 to 1.0 pm region. To simulate

where

= po(%.)

implemented water,

set for vegetation

the effects

of surface

target

e -x°13°(%.)

surface

variance

For each

from empirical

surface

reflectance

and bare

in the 0.4 to 2.0 pm

land

for the more

that all reflectances

is modeled

to two sets:

variability,

limited

are Lambertian. the reflec-

by

,

(4)

and x o is the standard

= i.

thus far is limited

reflectance

po(%) and Bo(%) are deterministic

the surface,

of interest.

snow and clouds

It is also assumed

tance of a particular

.p(%.)

data

region

functions

normal

random

the parameters

data using

which

are characteristic

variable

0o(%)

with

mean

of

= 0. and

and Bo(%.) are estimated

the relationships

(Ref. i0).

L2 and

13o(%.) :

in

--2

+

(6)

where

0 (%.) = < [p(%.)

The model

given by Eq.

the reflectance produce

8

-

a random

]

.

(5) has been

variability family

2>

shown

of vegetation

of reflectances

to be approximately canopies

whichagrees

representative

(Ref. ii) and does qualitatively

of

indeed

with

selected

data

sets.

This model

targets

simply

because

another

model.

Table

work

as well

bility.

as the

Figure

signatures

is also used

for the reflectance

there are insufficient 2 summarizes

(assumed)

4 illustrates

and Figure

data

the

and substances

deviation

(expected)

5 shows the simulated

of other

in the literature

the categories

standard

variability

G 0 of their

mean

values

variability

to suggest

used

in this

reflectance



vari"

of the spectral

of two of these

signa-

tures. The mean measurements.

spectral

reflectances

The curves

for crops

(Ref. 12) and Suits and Safir, (Ref. 14). abilities

The associated reported

correction

well

for simulation

suited

reflectance

variability

that is only ±5 percent variability

reported

The spectral Condit,

curves

tance curves were obtained

reflectance,

in Figure

The variability for varying

an average

of vari-

Rao,

for bare

land were

results

is much

and the variables

an

are not on the

of reflectance lower

than

the

(Ref. 16). obtained

(Ref. 19).

4 are the averages

et. al.,

incorporate

variability

which

mostly

The mean

from

spectral

re-

of the wet and dry reflec-

of the spectral

the wet and dry reflectance

i standard

of crops,

(Ref. 15) and Duggin,

Handbook,

the range

(Ref. 16)i.

et. al.,

from Vlcek,

(Ref. 14) who reports

of the mean

given by Condit, by using

and Vlcek,

curves

for forests

so that their

gives

by Collins,

shown

effects,

of forests,

reflectance

tance plus or minus

ations

studies;

variability

in situ

from Leeman,

within

(Ref. 15) and Duggin,

(Ref. 18) and the Infrared

flectance

fall roughly

for atmospheric

represent

mostly

(Ref. 13) and the curves

on the reflectance

unspecified

for vegetation

were obtained

variables

by Collins,

(Ref. 17) who report

used

curves

as

reflectances

the mean

reflec-

deviation.

for water was obtained amounts of chlorophyll,

from the water

as given

reflectance

in the Infrared

vari-

Handbook,

9

(Ref. i0).

The spectral

by O'Brien

and Munis,

curves

19.

from samples

The variability

for ice clouds with

different

atmospheric

(Ref. 22).

for clouds

based

of Novosel'tsev, A standard

employs

random

subset

include

access

of the data.

the ability

rently

stored,

5.

Signal

values

are varied;

in Figure

spectral

6.

Three

from

results

of 0.i was chosen

quick

in a data base which access

allow several

the titles

to any partic-

options

which

of all the data cur-

data sets.

visual

L(%)

ranges

and bare moist

constant,

7(b) both

comparison

the atmospheric

and model

with

dX + nj,

are

The variability

of

and background

permitted s with

different

this conversion

the signal

angles

absorber

L into the signa ! vector obtained

by the above

zenith

7(a) target

and atmosphere:are

to normalize

S.(X) dXl_Sj(X) 3

and two solar

In Figure

and only

of signals

of the sensor

as produced

soil as background.

7.

surface

the radiance

it is convenient

response

s. 3 = _

provides

at the top of the atmosphere

are kept

converts

To facilitate shapes,

and maintained

to list

particular

radiance

reflectance

sponse

obtained

(Ref. 23).

programs

field is shown in Figure

The sensor

reflectance

Computation

oats as target

this radiance

feature

to add new data,

is shown fn Figure

shown with

is stored

This

were

in reflectivity

region

of reflectances

(Ref. 21) and the experimental

The data base

and to update

The spectral

i0

files.

spectral

The spectral

thicknesses

in Kondratyev,

data

the range

histories.

deviation

on data presented

All of this reflectance

s.. 3

represents

thermal

results

from data reported

for the 0.4 to 0.6 _m

different

of Zander,

model

data

with

the analytical

Ular

for snow was obtained

(ref. 20) with

added from Reference obtained

reflectance

amounts

to vary. components

spectral

re-

by the integrated

component

sj as

(7)

where

S(%) is the spectral

the subscript as a normal

random

Three the U.S.

j denotes

d'Observation scientists responses

variable

with mean

channels

Thematic

Mapper

de la Terre

Kondratyev,

Vasilyev

arbitrarily

measurements, selected

to these channels,

8.

whereas

noise

ratios

flectances

the TM has two other

The basic

tialize

Structure

structure

of a driver parameters,

subroutine

SIMULA

Subroutines cients),

SOLARI

the surface functions

angles

program, assemble

CALPHA

of the TM.

(Ref.

on sen-

are somewhat In addition

at 2.24 and

11.5 _m)

The TM signal-to-

24) for specifiedsurface

e o ; however, arbitrarily

STIMULA

MAIN,

which

is shown

invokes

the atmospheric

re-

state was

for our simulation.

data and perform

the actual

the calculations.

the atmospheric irradiance)

reflectance

data) all handle

data which

it was decided

9.

The program which

ini-

The principal

simulation.

(which assembles

_ To access

in Figure

the subroutines

sets up the solar

by the simulation,

are based

0.5 to 0.7 _m).

(which

of wavelength.

spectral

for STIMULA

of program

drives

Probatoire

Their

by Kondratyev.

(centered

(panchromatic,

and thus had to be chosen

Program

consists

channels

(SNR) are given by Salomonson,

not specified, 6.

recommended

characteristics

p and solar incidence

(Ref. 26).

include

by the Russian

the shapes of the KON responses

for the intervals

3 lists sensitivity

(KON),

They

System

proposed

The TM and SPOT responses

and the SPOT has one other channel Table

n

(Ref. 24), the French

and Ivanyan

and

2.

in this simulation.

(Ref. 25) and those

noise,

noise n is characterized

= 0 and Variance

(TM),

electronic

The electronic

are used

(SPOT),

are shown in Figure

sor response

n is the normalized

the jth channel.

sets of sensor

Landsat

response,

data values

to use average

attenuation

and SREFL are stored

at spectral values

coeffi-

(which reads in tables

grid

points

of the spectral

as required functions II

over

the appropriate

was performed

in subroutine

Subroutine supplied

spectral

with

PFINPT

Subroutine

SETRAN

Subroutine

NCVRCND

simulation.

AEROSOL

range read

in NCVRCND.

Once

PERTRB

Subroutine

CTAU

(3).

are then stored described

scattering

which

integration

phase

method.

function

data

is then passed

generator

amount

based

SIMULA

is called.

First

Then the target reflectance SREFL

the atmospheric to evaluate

RADMOD

GETRAN.

upon visual

and RFLPRM

optical

the single

to CSIGNL

which

calculates

the signal vectors.

to the signal

albedo

to calculate

above.

given

according

scattering

is then called

and the

described

thichness

the reflectance

ZERSTS

respectively.

all Of the random variables

to provide

and used as input

number

data to set up a particular

subroutine

to calculate

factors.

system

the input

are read using

is used

is used

the random

arrays.

to calculate

RFLRND

of the sensor

an integration

rule

of the aerosol

to generate

geometry

at the sensor which

by

to Eq. and sun-

the radiance

the signal in each These

signal

vectors

processing

programs

which

are

computed,

a number

of options

below.

Signal Processing Once the simulated

are available

large number so that

2).

of the signals

of times

the stochastic

tions which

signal

(see Figure

are the values

12

a value

are assembled,

is used

GENCND

surface-viewing

B.

to read

parameters

is used

Subroutine

channel

in the single

is used

assigns

Subroutine

a Simpson's

is used to initialize

variability

Subroutine

(4).

using

to zero out the necessary

reflectance

Eq.

To do so required

software.

all the arrays

is called

INTGRT

reads

the RADMOD

bins.

have been

vectors

have been

The J-dimensional in each channel

(i00), thus, allowing nature

of the system,

the parameters

of the simulation

implemented

signal vectors,

are programs

elements

are generated

to vary

may be studied. to display

whose

a

sufficiently

Among

the op-

two-dimensional

scatter

plots of the signals ize the variation the distribution means

by which

ances between

of signal values

about

one-sigma

indicate

Analysis

The spectral tic process variables ation

whose

E{e} denote radiances

autocovariance



covariance

the number

radiance value

channels

standard

These

to show

all provide

deviations

conditions

change.

and covariExamples

i0 - 12. ELLIP

are two-dimensional

ellipsoids

onto

the plane

of the signal

of measurements

that reaches

contained

the sensor

% depends

the atmosphere

the expectation associated

to character-

projections

shown.

scatter, within

These

of ellip-

but they do

their

areas.

Processes

L(%)

both

plots

(HISTPLT).

as viewing

at each wavelength

with

ellipse

(ELLIP) and histograms

the mean

size and orientation

of Stochastic

associated

possible

in Figures

(SCATPLT),

(mean values,

plots drawn by program

the relative

not actually

statistics

can be studied

are shown

J-dimensional,

i.

between

channels)

The ellipse

space

and covariance

the signal

of these plots

ses depict

in the signal

(average)

with

upon

taken

as a stochas-

a number

and surface.

a particular

C(%,% _) of the radiance

is modeled

of random

Letting

over the ensemble

surface,

can be expressed

the mean

the operof all



and

as

= E{L(%)}

(8)

and

CL(%,%_)

Likewise, treated

= E{[L(%)

the signal

- ]

vector

as a multivariate

components

denoted

r° = E{s.} 3 J

• [L(% _) - P(c:/s) _

IF AND ONLY IF P(s/c::)

m = i, 2, ..., M,

i = i, 2, ..., Im .

The relative

occurrences

are usually

unequal

information, targets

identical.

It is common distribution density

in which

The most

variate

or Gaussian

normal

m

1

P(si/ci)

= (27)J/2

m

tasks

of this lack of

of occurrence

of the

on the conclusions

imposes

constraints

of classification processes

(or training)

frequently

PDF which

is given

1 Icml ½ exp [-_

in order

probability

to reduce

storage

is the J-dimensional

multi-

by the expression

m t (s-

are

the statistical

data by an analytical

used function

rules

accuracy.

to characterize

only a few parameters

(cm')-l--i(£

--1-rm')_ ]

(15)

r i)

m

where --1 r. and _ i' respectively,

target

Because

sensing

obviously

in MLH decision

requirements.

probabilities

remote

decision

from predictions

(PDF) with

in typical

P(c_.)

case the MLH and simple Bayes

of the reference

function

> P(s/c:)

also unknown.

that the a priori

This assumption

that can be drawn

of interest

and, unfortunately,

we assume

are equal,

of targets

I

P(ci,)

m

c.1 and are given

are the mean

vector

and covariance

matrix

for

by J

__ _ _ m_ = f -s P(slC_)

j_l dsj

(16)

and J CTM = -i

-

f (s -

m

) (s -

-

.

m t e(s/C_)

dsj

(17)

17

To avoid an equivalent

the computational classification

This procedure assigns

is realized

m s to c. if and only 1

(_i' m')-i

reduced

mean-square

matrix

acquired

data

classifier

conceivable

that the added

18

r)+ :m

(s-

log e I_

required

assigns

for the MLH classifier

distribution matrix.

of the signal

is so

The resulting

s to c_1 if and only

--

depends

targets;

of the signal expense

to examine

simulates

accuracy

if

(19)

data

is representative

for each target

of performing

by the MLH

the extent

and the fit of the assumed

over the simpler

these

that can be attained

on two factors:

of the reference

in accuracy

•Implementation

STIMULA

(18)

to the identity

in classification

for the selected

has been designed

which

; i = i, 2, ..., Im .

distribution

3.

(C_)-i

_ reduces

distribution

the increase

classifier

m t (s - !i m ), < (s - ri)

statistical

weigh

- !imt)

(MSD) classifier

over the MSD

the statistical

- in P_/c_).

I_i' m'I

the statistical

The improvement

(15),

if

by disregarding

m = !, 2, ..., M

minimizes

(or Gaussian)

and storage

distance

in Eq.

i = i, 2, ..., Im .

m' t (_ - _i,) m' (! - !i,)

classifier

_H

of computations

that the covariance

the exponent

can be used which

(! - _im'') + l°ge

m = i, 2, ..., M,

The number

of evaluating

by the so-called

< (s

further

procedure

--

(_ - _i' m_)t

expense

(Ref. 27).

MLH classification

to which

of all (Gaussian) It is also may out-

MSD classification.

This model

types of questions.

Of Decision •Rules an orbiting

multispectral

sensor

system which

generates

pseudo-random

observations

(i.e.,_ signal vectors)

at the top of the atmosphere. ing sets for the reference decision

algorithms.

sification

with

library

of training

are both based

larger

significantly

change

grams BNDARY,

RPLPROG

numbers

observations

or classified

The assembly

accuracy

Computations

These

sensor

are either

as train-

using

assembled

the BAM, MSD,

upon a total of i00 observations

(typically

and CLASIFY

which

and MLH

sets and the assessment

of observations

the results

from a nadir-looking

of clas-

per target.

(e.g., 400) per target

only by about

implement

these

i percent).

tasks

did not The pro-

are described

below. a.

BNDARY

Program

BNDARY

from STIMULA regions

meters

to perform

by the boundary

of planes

specifying

is used

approximation

in the signal

the boundaries

are variables

feature

space

categorization

method

(BAM).

are straight

lines,

are the slopes• and intercepts

in BNDARY

of signals

output

The boundaries

between

and so the parameters

of the lines.

and can be •adjusted to achieve

These

para-

the maximum

accu-

racy. The boundaries

shown by dashed

the signals

into their

and clouds)

with

ing conditions

shows

considered

which

b.

imaging

vegetation,

consideration

conditions accuracy

routinely

selected

bare land,

for the two most

to divide water, extreme

snow imag-

experiment.

A tradeoff

implicit

is illustrated

in Figure

13; which

with

land and water

can be accomplished

were

(i.e., vegetation,

in discrimination

divides

type of analysis

equal

in Figure ii

in this computational

between

the variation

boundary

five categories

approximately

in this compromise

lines

changes

in the•threshold

from snow and clouds.

using

This

BNDARY.

RPLPROG

As discussed computation

above,

classification

of discriminant

functions

by the MSD or MLH schemes which

govern

requires

the classification

the

process. 19

To compute vectors

the discriminant

_ and the covariance

is called

the"reference

in practice, conditions

the reference

signals

RPLPROG

uses i00 signals

and store reference the covariance

matrices,

would

the reference

of the surfaces

their

reasonable

be generated

the clearest

comprising

can then be accessed

under

sky

were

namely

the mean

that

clear

patters

in the library

and their

in what

to assume

conditions,

the following:

inverses

are stored

signal

cal-

V=55

km.

to generate signal vectors

determinants,

_i' C--_ "I'

by the classification

and

program

CLASIFY

Program

CLASIFY

by both

nal vectors

correctly,

is used

by STIMULA

Classification is the output.

before

to perform

feature

the MSD and MLH algorithms.

generated

by RPLPROG.

known

These

quantities

It seems

to date,

use of the mean

for use in the MSD a_dl MLH algorithms.

c.

signals

under

repeated

These

library

done

for each

patterns

ICi[ respectively. CLASIFY

pattern

simulated

C.

library".

and so in simulations

using

requires

matrices

pattern

culated

h,

functions

and'the

accuracy, Since

classification

identification

Inputs

reference

the fraction

the "true"

takes place,

of multispectral

to the program pattern

are the sig-

libraries

generated

of all classifications

identification

the calculation

done

of the signal

is

of the accuracy

is

trivial. Classification

by MSD amounts

between

the point

signals

in the reference

for which

the distance

Classification function

for each

sification

20

specified

to calculating

by the signal

library

being

and assigning

the "distance" classified

in signal

and each

the "unknown"

signal

space

of the mean to the class

is a minimum.

by MLH requires of the surfaces

based on the surface

the computation in the reference

which

yields

of a probability library

the largest

and makes

probability

density the clasdensity.

III.

The model smart sensor

described

design

fies

specific

display variable

This

looking

patterns

under

ual range

of 55 km and solar

Figure BAM using

sensor

zenith angle

The boundaries Figures

employed

are those

15 to 17 present

classification

schemes.

SPOT, and KON and classification of an additional

channel;

For further

discussion

are typical

examples

and Figure

of all these

it classishown here

as a function

of the

All of the simula-

simulated,

accuracies

Reference i.e., vis-

the three

themselves;

results

of the type of studies

with

the

ii.

results

17 shows

attained

0.84, and 1.68 _m respectively.

in_Figure

15 compares schemes

accuracy

of

of 30 ° .

classification

Figure

A measure

the examples

of conditions

at 0.67,

shown

a few examples

with which

conditions.

categorization

centered

in studying

at the top of the atmosphere.

the clearest

14 shows the feature

three TM channels

of different

gives

this model.

Therefore,

or classification

tions are for a downward calculated

with

be the accuracy

or types of targets.

for a number

to be used

of the report

which may be performed

categorization

of interest

were

section

of a smart sensor would targets

either

APPLICATIONS

in this report was deisgned

concepts.

of the types of studies the performance

SAMPLE

using

sets of channels, Figure

the effects

see Huck, which

the MSD and MLH

16 shows

the effects

of simulated

et. al.,

are possible

TM,

noise.

(Ref. 28).

These

by employing

this model.

21

IV.

The computational used

as a tool

in evaluating

Such evaluations requirements. attenuation ities

are useful The program

processes

examined

in developing

the performance

smart

sensors

remote to reduce

and surface

reflectance

and the effects

of multispectral

sensor

of the variable

of occurrence variability

presented

of potential

here

remote

sensor

elements

and cloud

of their

reflectance.

a useful

concepts.

data processing

Data processing may also be

of the model.

of the spatial

of surface

provides

concepts.

of these variabil-

signals

models

sensor

to be

of atmospheric

systems.

simulated

by a lack of representative

and temporal model

remote

in classifying

as a function

or probability

the computational

of potential

designed

of the variability

is limited

angular

has been

for studies

systematically

distribution

in the report

the performance

and their performance

The model

spectral,

described

REMARKS

allows

in the performance

algorithms

22

model

CONCLUDING

targets

and the

Nevertheless,

tool for assessing

REFERENCES

i.

Malila, W. A., Crane, R. B., Omarzu, C. A., and Turner, of Spectral Discrimination", NASA CR-134181, 1971.

2.

Turner, R. E., "Atmospheric Effects in Remote Sensing", Remote of Earth Resources, Shahroki, F., ed., University of Tennessee Institute, Vol. II, 1973.

3.

Turner, R. E., "Atmospheric NASA CR-141863, 1975.

4.

Turner, R. E., "Elimination of Atmospheric Effects from Remote Data", Proc. Twelfth International Symposium on Remote Sensing Environment, II, 1978.

5.

Landgrebe, of Scanner 1977.

6.

Kondratyev, K. Ya., Grigoryev, A. A., and Pokrovskiy, O. M., "Information Content of the Data Obtained by Remote Sensing of the Parameters of the Environment and the Earth's Resources from Space", Izdatel'stvo Leningradskogo Universiteta 17, NASA TT F-16435, 1975.

7.

Kondratyev, K. Ya., Beliavsky, A. I., Pokrovsky, O. M., "Possibilities Optimal Planning of Multipurpose Survey from Space", Proc. Thirteenth International Symposium on Remote Sensing of the Environment, 1979.

8.

Smith, E. V. P., and Gottlieb, D. M., "Solar Space Science Reviews, 16, 1974.

9.

Effects

in Multispectral

R. E., "Studies

Remote

Sensin$ Space

Sensor

Data",

Sensor Of the

D. A., Biehl, L. L., and Simmons, W. R., "An Empirical Study System Parameters", IEEE Trans. on Geoscience Electronics, GE-15,

of

Flux and It's Variations", •

r

Kneizys, F. X., et al., "Atmospheric Transmittance/Radiance LOWTRAN 5", AFGL-TR-80-0067, Environmental Research Papers,

Computer Code No. 697, 1980.

i0.

Huck, F. O., et al., "ComputationalModeling for the Study spectral Sensor System and Concepts", Optical Engineering,

of Multi21, 1982.

ii.

Park, J. K., and Deering, D. W., "Relationships between Diffuse Reflectance and Vegetation Canopy Variables based on the Radiative Transfer Theory", NASA TM-82067, 1981.

12.

Leeman, V., Earing, D., Vincent, R. K., and Ladd, S., "The NASA Earth Resources Spectral Information System: A Data Compilation", NASA CR-I15757, 1971.

13.

Suits, G. H., and Safir, G. R., "Verification of a Reflectance Model for Mature Corn with Applications to Corn Blight Detection", Remote Sensing of the Environment, 2, 1972.

23

14.

Vleck, J., "Difficulties in Determining Meaningful Spectral of Forest Tree Canopies", Proc. Symposium on Remote Sensing Interpretation, Vol. II, 1974.

15.

Collins, W., "Remote Sensing of Crop Type and Maturity", Engineering and Remote Sensing, 1978.

16.

Duggin, M. J., "On the Natural Limitations of Target Differentiation by Means of Spectral Discrimination Techniques", Proc. Ninth International Symposium on Remote Sensing of the Environment, 1974.

17.

Rao, Vi R., Branch, E. J., and Mack, A. R., "Crop Discriminability in the Visible and Near Infrared Regions", Photogram_etric Engineering and Remote Sensing, 1978.

18.

Condit, H. R., "The metric Engineering,

19.

Wolfe, W. L., and Zissis, G. J., The Infrared Research Institute of Michigan, 1978.

20.

O'Brien, H. W., and Munis, R. H., "U. S. Army Cold Regions Research and Engineering Laboratory Research", Report 332, 1975. Also in The Infrared Handbook, by Wolfe and Zissis.

21.

Novosel'tsev, 1965.

22.

Zander, Journal

23.

Kondratyev, the Earth's

24.

Salomonson, V. V., "Landsat D, A Systems Overview',, Proc. Twelfth national Sympsoium on Remote Sensing of the Environment, 1978.

25.

Begni, G., "Selection of the Optimum Spectral Bands Colloquim on Spectral Signatures of Ground Objects, Translated by P. N. Slater.

26.

Kondratyev, K. Ya., Vasilyev, O. G., and Ivanyan, C. A., "On the Optimum Choice of Spectral Intervals for Remote Sensing of Environment from Space", Remote Sensing of Earth Resources, Shahroki, F., ed., University of Tennes" see Space Institute, Vol. II, 1973.

27.

Haralick, R. M.,i_"Automatic Remote Sensor Image Processing" ture Analysis, Rosenfeld, A., ed., Springer-Verlag, 1976.

28.

Huck, F. O., Davis, R. E., Fales, C. L., Aherron, R. M., Arduini, R. F., and Samms, R. W., "Study of Remote Sensor Spectral Responses and Data Processing Algorithms for Autonomous Feature Classification", to be published, Optical Engineering, 1984.

24

Spectral 1970.

Reflectance

Ye. P., "Spectral

R., "Spectral Scattering of Geophysical Research,

of American

Photogrammetric

Soils",

Handbook,

Signatures arid Photo

Photogram-

Environmental

Reflectivity

of Clouds",

Properties 1966.

of Ice Clouds

NASA TT-F-328,

and Hoarfrost",

K. Ya., ed., Radiation Characteristics of the Atmosphere and Surface, Amerind Publishing Co. Pvt. Ltd., New Delhi, 1973. Inter-

for the SPOT Satellite", Avignon, September 1981.

Digital

Pic-

TABLE

i. SUMMARY

OF PARAMETERS

PARAMETER GEOMETRY SOLAR ZENITH ANGLE NADIR VIEWING ANGLE

SYMBOL

USED

IN SIMULATION

VALUES

@o @

30 ° 0°

_

i00 °

---

0_ 0b

see Fig. 4 see Fig. 4

see Table 2 and F_g. 5

MOLECULAR OXYGEN BURDEN WATER VAPOR BURDEN CARBON DIOXIDE BURDEN OZONE RELATIVE HUMIDITY

X02 XH20 XCO 2 X03 RH

1.71 1.14 8.01 0.34 0.40

0.13 0.36 0.24 0.12 0.20

SURFACE

Po

1013 mb

---

55 km 33 km 14 km

-------

RELATIVE SURFACE

AZIMUTH

40 °

VARIABILITY

-----

PROPERTIES

TARGET REFLECTANCE BACKGROUND REFLECTANCE ATMOSPHERIC

PROPERTIES

PRESSURE

VISUAL RANGE "CLEAR" "INTERMEDIATE" "HAZY"

km STP cm -I arm cm atm cm

km STP cm -I atm cm atm cm

V

25

TABLE

2.

TARGETS

AND THE ASSUMED

(a)

WAVELENGTH

CATEGORY

VEGETATION

COTTON TOBACCO BEAN OATS PINE

II.

BARE LAND

BARE MOIST DRY SAND

0.4 - 2.0 _m

LI L2

0.13 0.i

LOAM, 1% WATER GNEISS

L3 L4

0.ii 0.i

WATER

SEA WATER

W

0.06

SNOW

SNOW, 14 HOURS SNOW, 44 HOURS SNOW, 70 HOURS

SI $2 $3

0.08 0.08 0.08

CI C2 C3 C4

0.i 0.I 0.I 0.i

T is optical

ICE ICE ICS ICE

thickness

CLOUD, CLOUD, CLOUD, CLOUD,

SOIL

• T T T

= = = :=

128 16 8 4

REFLECTANCE

STANDARD DEVIATION OF REFLECTANCE

0.i 0.I 0.i 0.I 0.i

V. CLOUD

26

REGION:

CODE

OF THEIR

Vl V2 V3 V4 V5



IV.

DEVIATION

SUBSTANCE

I.

III.

STANDARD

0

TABLE

2.

TARGETS

AND THE ASSUMED STANDARD (CONCLUDED) •

DEVIATION

(b)

0.4 - 1.0 _m

WAVELENGTH

CATEGORY

I.

II.

VEGETATION

BARE LAND

REGION:

BARE

REFLECTANCE

STANDARD DEVIATION OF REFLECTANCE

SUBSTANCE

WHEAT BEAN BARLEY OATS CORN RED SPRUCE BALSAM FIR COTTONWOOD ASPEN PINE WHITE PINE

OF THEIR

• _

SOIL

PEDOCAL, OHIO PEDOCAL, NEBRASKA PEDOCAL, OKLAHOMA CLAY, MISSOURI QUARTZ SAND, OREGON CHERNOZEM, NEBRASKA PEDALFER SILT, ARKANSAS •• RED QUARTZ AND CALCITE SAND, UTAH LOAM 20% WATER

0

0.i 0.i 0.i 0.i 0.i 0.i 0.i 0.i 0.i 0.i 0.13 0.i! 0.02 0.I 0.06 0.14 0.13 0.ii 0.ii 0.ii i

27

TABLE

3.

SPECTRAL CHANNEL

WIDTH

SENSITIVITY

CHARACTERISTICS

SNR FOR LOW LEVEL

INPUT a

OF THEMATIC

NORMALIZED LOW LEVEL IMPUTFROM SIMULATION b (W m-2 sr-I _m -I)

NORMALIZED NOISE,

1

0.45-0.52

32

1.58

.049

2

0.52-0.60

35

1.00

.029

3

0.63-0.69

26

.60

.023

4

0.76-0.90

32

.28

.009

5

1.55-1.75

13

.16

.012

aspecified in Ref. 24 as P = 0.01 and 0o = 70° for bands P = 0.02 and eo = i0 ° for band 5. bMean

28

MAPPER

value

for visual

range V = 55 km.

rms On

i to 4, and _

SIGNAL GENERATION

STOCHASTIC ELEMENTS :

ABSORBER

,

AMOUNTS

i REL. HUMIDITY IVISUAL RANGE i

AEROSOLS

l DETERMINSTIC ELEMENTS :

SOLAR IRRADIANCE AT TOP OF ATMOSPHERE

t

SENSOR

VARIABILITY

NOISE i

l

ATMOSPHERIC ABSORPTION

I

i

REFLECTANCE

ATMOSPHERIC SCATTERING

J SURFACE REFLECTANCE

I

l RADIANCE AT SENSOR

I RESPONSE SENSOR

FIGURE [,o

i.

SCHEMATIC

I

OF SIGNAL GENERATION

MODEL

J

.SIGNAL PROCE$SI_IG VECTORS SIGNAL

SCATTER

I

)

OR

PATTERN

CLASSIFY

(RPLPROG) LIBRARY

I

1OA APPROXIMATION METHOD (BNDARY)

SQUARE DISTANCE (CLASIFY)

CCURACY

FIGURE

2.

LIKELIHOOD (CLASIFY)

J

SCHEMATIC OF SIGNAL PROCESSING (PROGRAM NAMES IN PARENTHESES)

OPTIONS

.30 B

LLI

ro z,

5

OATS

,4--

w o

_.3

_

.3

W

.2

.1

.1

•3

.5 --

.4

.5

.6

BARE

.7

.8

.g 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1

MOIST SOIL

.4 --

r"

.-_'"

,,-_

\ \

/

1

,,-'"-"--,

\

.3

.4

.5

.6

.7

.8

.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1

.4

.5

.6

.7

.8

.g 1.o 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 WAVELENGTH. pxn

(b)

Typical

"5I'_ .4

\

.3 _.3 F0

/

I \

/ / _-_'-,\\ /

W

uJ fY

/

,/

_,/_

/,..,,

.1 o .,3 .4.

_"

,--'--_

\

x./ //

(a)

.6

.7

.2

\/_ .1

//

.5

/ /

_\x..J\ \ ] .8

Average

FIGURE

.9 1.o 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 WAVELENGTH. pxn reflectance

5.

and

standard

SIMULATED

o .3

deviation.

SPECTRAL REFLECTANCE

VARIABILITY

realization

of

variability.

FOR TWO TARGETS

10 --

10

Oo = 30 ° B--

V. km 55

0 o = 40 ° 8_

"'7, ','5 _E

2 __

2

o IIIiilii ,3

.4

.5

.6

.7

.8

I

.9

1.0 1.1 1.2 1.3 1.4 1.5 1,6 1.7 1.8 1.9 2.0 2.1

.3

.4

WAVELENGTH. p.rn

FIGURE

6.

MEAN

SPECTRAL

ANGLES,

RADIANCES

_o u1

.6

.7

.8

.9

1.0 1.1

1.2 1.3 1.4 1.5 1.6 1.7 1.8

INCIDENT

ON REMOTE

RANGES,

V,

SENSOR FOR TWO SOLAR

USING OATS AS TARGET

I

1.9 2.0 2,1

WAVELENGTH.

e o , AND THREE VISUAL

SOIL AS BACKGROUND

,5

INCIDENCE

AND BARE MOIST

L,_ O_

I

10--

0o = 30°, V = 55 km

-..1_

0 o = 40°, V = 14. km

o__ 0.3 ,4.I .5I .6I .7I .8II .9 1.0I I,II 1.2I 1.3I 1,4. 1.5 1.6 1.7 1,8 1.9 2.0 2.1I

o.3

.4- .5

.6

.7

.8

.9 1,0 1.1 1,2 1,3 1.4. 1.5 1.6 1.7 1.8 1.9 2.0 2.1

WAVELENGTH. /_m (o)

WAVELENGTH.

Accountrng

for

the effects

of atmospherlc

1o

variability

only.

10 --

e o = 30°, V = 55 km

_o = 40°. V = 14. km

_6 _ '_ffl

6

n*"°E 4 < 0

4.

_E

2

0 .3

2

.4

.5

.6

.7

.S

.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.g 2.0 2.1

0 .3

.4. .S

.6

.7

WAVELENGTH. pzn (b)

FIGURE

7.

Accounting

.9

1.0 1.1 1.2 1.3 1.4. t.5 1.6 1.7 1.8 1.9 2.0 2.1

WAVELENGTH, p.m for

TYPICAL

REALIZATIONS

VIEWING

CONDITIONS,

AS BACKGROUND

.8

the effects

of both

OF SPECTRAL USING

atmospheric

RADIANCE

OATS AS TARGET

ond

surface

VARIABILITY

variobility.

FOR TWO

AND BARE MOIST

SOIL

1.o .9

^_- r,,,,_ I I I

.8

1.o ,9 --

.8

.8 --

it II II II

.7 t.=.J O3 Z .6

II

.7 .6

.7 -.6 --

II II II

o .5

'I

.5

.5--

II

.4

.4 --

.3

.3 --

.2

.2 --

.1

.1 --

.4 n."

.3

I I I

.2 .1

o

o

WAVELENGTH, (a)

Thematic

p,m

Mapper

FIGURE

L_

t.O ,9

8.

(b)

SIMULATED

SPECTRAL

'=

o i i1"t

WAVELENGTH, (TM).

"

/a,m

SPOT.

SENSOR RESPONSES

(c)

II II II II II II 11

i i I

WAVELENGTH,

/.m'i

Kondratyev

(KON).

..

TA!_LE

ISIMULA[

FIGURE

9.

STIMULA

PROGRAM

STRUCTURE

10 _

10]_

9--

0o = 30° V = 55 km

8 --

.:".'.

9 _



0o = 40° V = 14 km

8--

i,_i: •.l. -t •

_d

e --

_.v"': "

s-

6_

:.,..

_

'..

_'d s--

4 --

4 _

2 --

2 --

1 --

1 --

o0

I1

I

1 I ,3 4

2

I

5

I

6

o0

s(0.67)

FIGURE

i0.

SIGNAL USING

o',",,;-....

""-.'

."

•!,.:_'.:v .. ..._..."'_'i_":':'-" .!:_! .__._,'.' .. .: v:,...._:

I1

I

2

I

_

I

4

I" I 5 6

s(0.67)

SCATTER

PLOT FOR TWO TM CHANNELS,

OATS AS TARGET

AND BARE MOIST

SOIL

AS BACKGROUND

39

35 --

35 O o = 30 °

.30-

V

00 = 40o

= 5.5 km

3o--

V

= 14 km

/ 25 --

/

I

25-

/

_o-

.-

(5

/ /

_,5-/

..

//

C1_1

. •

/ //'C2/_/3

=

._._o-_

Z.,."_

d

10

et_



.

//I_'/'jV5

_

/

/ /

L_- I ll_

VEGETATION

/ J _/

/.

SNOW

_

_///_/j

//.

,I

%_ i / CLOUD.

__.

_

"_",s --

i

_/

S(0.67

_

x

1I

//

_ 05

_o

o'.._ v' i w i i i i o-_""i i i i i 5

10

15

20

25

30

35

0

5

10

s(O.67)

FIGURE

ii.

SIGNAL

COVARIANCE

CONTOURS

LISTED

FOR THREE

IN TABLE

SOIL AS BACKGROUND.

ARE USED

25

TM CHANNELS,

2(a) AS TARGETS

THE BOUNDARIES

TO DISTINGUISII BETWEEN

BARE LAND, WATER, 40

20

30

s(0.67)

THE SUBSTANCES MOIST

15

SNOW AND CLOUD.

AND BARE

(DASHED

THE CATEGORIES

USING

LINES)

VEGETATION,

35

3O"

# -

z_

_ i

--

I

I.II. ]

3.737

30"

ee " _'

_' "

.,,21

;

I II I I I I

eo -

, = .,_7

i

°L J__L_J

_

3.362

I/

L

e, - 30. v -55win

#. a -

FIGURE

o - .3_5

12.

/= o -

GAUSSIAN

HISTOGRAM DISTRIBUTION

AND VARIANCE, MOIST

USING

VARIABILITY

!

I

_ - 2.513

I

I

.819 .076

(SOLID) AND "EQUIVALENT" (DASHED) WITH OATSAS

SOIL AS BACKGROUND.

RADIANCE

I

eo - 30v -55_

6.4452 .622

TM SIGNAL

L

TARGET

EQUAL

MEAN

AND BARE

THE CORRESPONDING

IS ILLUSTRATED

IN FIG.

7(b).

41

FIGURE

12.

TM SIGNAL GAUSSIAN

HISTOGRAM DISTRIBUTION

AND VARIANCE, MOIST

(DASHED) WITH OATS AS TARGET

SOIL AS BACKGROUND.

RADIANCE

VARIABILITY

(CONCLUDED).

42

USING

(SOLID) AND "EQUIVALENT" EQUAL MEAN AND BARE

THE CORRESPONDING

IS ILLUSTRATED

IN FIG.

7(b)

Discrimination between vegetation/land/water and snow/cloud for unratJoedsignal=.

FIGURE

13.

DISCRIMINATION

ACCURACY

BET_CEEN GROUPS

OF CATEGORIES,

LOCATED

VERSUS

THRESHOLD USING

BOUNDARY

THE TM CHANNEL

AT 0.67 _m

43

1.00

>-. _,._. •

t.

Report No.

2. Govecnrnent Accession No.

3. Recipkmt's Ca_log No.

NASA CR-172393 =4. Title _

Subtitle

5'. Report Dlte

A SIMULATION OF REMOTE SENSOR SYSTEMS AND DATA _ PROCESSING ALGORITHMS FOR SPECTRAL FEATURE

JULY 1984 6. Performing Or_nizationCode

CLASSIFICATION 7. Author(s)

8. Performing Organ;zat;on Report No.

FR 683110 ROBERT

F. ARDUINI,

R. MARTIN

AHERRON,

RICHARD

W. SAMMS

10. Work Unit No.

9. Performing Organization Name and Address

INFORMATION & CONTROL 28 RESEARCH DRIVE HAMPTON, VA 23666

SYSTEMS,

INCORPORATED 11. Contract or Grant No.

NASI-16870 13. Type of Report and Period Covered

12. Sponsoring Ager_y Name and Address

CONTRACTOR

NATIONAL AERONAUTICS AND SPACE ADMINISTRATION WASHINGTON DC 20546 '

REPORT

14. Sponsoring AgencyCode 506-58-13-11

15. S4Jpl_ementaryNotes

NASA LANGLEY FINAL REPORT

TECHNICAL

MONITOR:

RICHARD

E. DAVIS

16. Abstract

A computational model of the deterministic and stochastic processes involved in multispectral remote sensing has been designed to evaluate the performance of sensor systems and data processing algorithms for spectral feature classification. Accuracy in distinguishing between categories of surfaces or between specific types is developed as a means to compare sensor systems and data processing algorithms. The model allows studies to be made of the effects of variability of the atmosphere and sensor

17. Key W_

and of surface reflectance, as well as the effects noise. Examples of these effects are shown.

(Suggestedby Auth,(s))

Multispectral

Sensor;

18. Distri_ti_

Remote

(of this report)

Unclassified .-3_

Statement

Sensor;

Smart Sensor; Spectral Responses; Classification Algorithms; Feature Classification; Spectral Reflectances. 19. SeCurity Clmf.

of channel

20. Security Clair.

Unclassified-Unlimited Subject (of this I:_ge)

Unclassified

Category

21. NO. of Pa_t

54

- 35

22. Dice

A04

Fu sale bytheNationalTechnicalInfumationService,Springfield.Vsrginia22161

selection