mixed pixel

Seediscussions,stats,andauthorprofilesforthispublicationat:http://www.researchgate.net/publication/215904887

SynthesisofmixedpixelhyperspectralsignaturesARTICLEinINTERNATIONALJOURNALOFREMOTESENSINGJANUARY2011ImpactFactor:1.65DOI:10.1080/01431161.2011.610378

CITATIONS2

READS31

1AUTHOR:

BarunRaychaudhuriPresidencyUniversity16PUBLICATIONS154CITATIONS

SEEPROFILE

Availablefrom:BarunRaychaudhuriRetrievedon:13October2015

This article was downloaded by: [Barun Raychaudhuri]On: 07 December 2011, At: 05:25Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Journal of RemoteSensingPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/tres20

Synthesis of mixed pixel hyperspectralsignaturesBarun Raychaudhuri aa Department of Physics, Presidency University, Kolkata, 700073,India

Available online: 07 Oct 2011

To cite this article: Barun Raychaudhuri (2012): Synthesis of mixed pixel hyperspectral signatures,International Journal of Remote Sensing, 33:6, 1954-1966

To link to this article: http://dx.doi.org/10.1080/01431161.2011.610378

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representationthat the contents will be complete or accurate or up to date. The accuracy of anyinstructions, formulae, and drug doses should be independently verified with primarysources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand, or costs or damages whatsoever or howsoever caused arising directly orindirectly in connection with or arising out of the use of this material.

International Journal of Remote SensingVol. 33, No. 6, 20 March 2012, 19541966

Synthesis of mixed pixel hyperspectral signatures

BARUN RAYCHAUDHURI*Department of Physics, Presidency University, Kolkata 700073, India

(Received 5 November 2009; in final form 14 May 2011)

The general method of analysing mixed pixel spectral response is to decomposethe actual spectra into several pure spectral components representing the signa-tures of the endmembers. This work suggests a reverse engineering of standardizingthe mixed pixel spectrum for a certain spatial distribution of endmembers by syn-thesizing spectral signatures with varying proportions of standard spectral librarydata and matching them with the experimentally obtained mixed pixel signa-ture. The idea is demonstrated with hyperspectral ultravioletvisiblenear-infrared(UVvisNIR) reflectance measurements on laboratory-generated model mixedpixels consisting of different endmember surfaces: concrete, soil, brick and veg-etation and hyperspectral signatures derived from Hyperion satellite images con-sisting of concrete, soil and vegetation in different proportions. The experimentalreflectance values were compared with the computationally generated spectral vari-ations assuming linear mixing of pure spectral signatures. Good matching in thenature of spectral variation was obtained in most cases. It is hoped that usingthe present concept, hyperspectral signatures of mixed pixels can be synthesizedfrom the available spectral libraries and matched with those obtained from satelliteimages, even with fewer bands. Thus enhancing the computational job in the labo-ratory can moderate the keen requirement of high accuracy of remote-sensor andband resolution, thereby reducing data volume and transmission bandwidth.

1. Introduction

Hyperspectral sensing, that is, remote sensing and imaging in hundreds of contiguousnarrow spectral bands, can detect, discriminate and classify many subtle features ofground objects. The obvious advantage over conventional multiband sensing is higherspectral discrimination of reflectance, which has made hyperspectral sensing popu-lar for precision agriculture (Zarco-Tejada et al. 2005), vegetation canopy modelling(Panferov et al. 2001), vegetation species identification (Alberotanza et al. 1999), min-eral exploration (Howari et al. 2002), snow measurements (Nolin and Dozier 2000),coastal analysis (Brando and Dekker 2003), environmental studies (Salem et al. 2005),target detection (Kwon and Nasrabadi 2005), military appliances (Moorhead et al.2001) and many other applications.Hyperspectral imaging is not a mere extension to the conventional multiband pro-

cesses, with the number of bands enhanced. It emphasizes the spectral domain andcalls for new approaches in the analysis techniques, as may be found in the liter-ature (Varshney and Arora 2004, Chang 2007). One of the subjects of interest in

*Email: [email protected]

International Journal of Remote SensingISSN 0143-1161 print/ISSN 1366-5901 online 2012 Taylor & Francis

http://www.tandf.co.uk/journalshttp://dx.doi.org/10.1080/01431161.2011.610378

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1

Synthesis of mixed pixel hyperspectral signatures 1955

hyperspectral remote sensing is the spectral analysis of mixed pixels, when severalsmall size targets get embedded in a single imagery pixel, each one contributing tothe resultant spectral reflectance of that pixel. A popular method of analysing suchhyperspectral data is linear mixing (Mazer et al. 1988, Settle 1996, Manolakis et al.2001), which assumes the measured spectra to be a linear superposition of severalpure spectra produced by different targets present in a small spatial extent. For mul-tiple scattering of radiation between elements in the scene, nonlinear mixing occurs(Borel and Gerstl 1994).Whatever may be the process of analysis, the general approach is to somehow

decompose the experimentally obtained hyperspectral data into a number of purespectra of the constituent materials applying various computational techniques on thedigital number (DN) values. The determination of the endmember spectra is basedsolely on the information contained within the image. Therefore, the satellite sensorfor image acquisition must be of the best possible spectral resolution and reflectancequality, thereby increasing data volume and redundancy due to correlation betweenadjacent spectral channels.This work suggests a reverse engineering of synthesizing hyperspectral signatures

with standard library data for different proportions of endmembers and comparingthem with the experimentally obtained mixed pixel signatures. The best match shouldbe considered as the actual condition of endmember distribution in the pixel. The ideais demonstrated with different proportions of reflectance data on both laboratory-measured objects and endmembers selected from satellite images. The experimentalreflectance values are compared with the computationally generated spectral signa-tures using linear mixing of previously known spectral signatures and good matchingis obtained in the nature of spectral variation. Thus, this work proposes a new modelfor spectral matching with the synthesized spectra demonstrated with simple lin-ear mixing of some common objects. It requires more computational developmentsas further scope for the study. In fact, that is the claimed merit of this techniquebecause precise spectral libraries are available on different varieties of pure endmem-bers. Spectral synthesis by random selection, matching with the signature obtainedfrom satellite image and conclusion on the relative weights of endmembers are theworkloads shared by the computer. Enhancing the computational job in laboratorycan moderate the keen requirement of high accuracy of remote-sensor and bandresolution.

2. Methodology

The work comprises reflectance measurement in the laboratory for a few sampleobjects and sampling of endmembers from satellite images. In one phase of this work,model mixed pixels and their hyperspectral signatures were generated in the labo-ratory involving natural and man-made objects: loam-type soil (3335% moisture),vegetation, brick and concrete. Fresh banana leaves were made to serve the purposeof vegetation. The hyperspectral reflectance for each object was measured through-out the ultravioletvisiblenear-infrared (UVvisNIR) region (3001000 nm) with1 nm resolution using a FieldSpec spectroradiometer (ASD Inc., Boulder, CO, USA),having an angular field of view (FOV) fixed at 25. The reflectance values were cali-brated with the Spectralon white reference panel acting as the Lambertian surface andthe measurements were carried out in open sunshine around solar noon. The sensorwas held vertically downwards 1 m above the surface so that the ground FOV was

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1

1956 B. Raychaudhuri

a circular area of approximately 22 cm radius, simulating a single pixel. The charac-teristic reflectancewavelength variation for such a pixel of each object was obtainedseparately as shown in figure 1(a). The surface area of each object under the FOVwas partly covered with fresh banana leaves, as indicated in figure 1(b), to simulatepartial vegetation cover of different extent for each object. Thus the signatures forthe mixtures of three different endmembers, concrete vegetation, brick vegetation andsoil vegetation, were generated at different proportions, as shown in figure 2(a)(c).The percentage of surface coverage for each endmember is indicated against thecurves. The pure pixels are indicated by 100, meaning 100% coverage with singleendmember.As an extension of the above work, 50 cm 50 cm square surface areas of four

endmembers, concrete, brick, loam soil (3335% moisture) and vegetation, were tiledtogether. The spectroradiometer sensor was held vertically downwards at a heightof 2 m just above the junction point (P) of the four different surfaces, as indicated

400 500 600 700 800 900 1000

0.0

0.2

0.4

0.6

0.8(a)

(b)

Conc

Conc-JHU

BrickBrick-JHU

Veg

Veg-JHU

Ref

lect

ance

Wavelength (nm)

Object surface

Banana leaves

Figure 1. (a) Comparison of present experimental reflectancewavelength plots for concrete(Conc), brick (Brick) and vegetation (Veg) with those obtained from the spectral library ofJHU and (b) layout of simulating partial vegetation cover.

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


300 400 500 600 700 800 900 10000.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7(a)

Conc 100 Veg 0

Conc 75 Veg 25

Conc 50 Veg 50

Conc 25 Veg 75

Conc 0 Veg 100

Ref

lect

ance

Wavelength (nm)

300 400 500 600 700 800 900 1000

Wavelength (nm)

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8(b)

Brick 100 Veg 0

Brick 75 Veg 25

Brick 50 Veg 50

Brick 25 Veg 75

Brick 0 Veg100

Ref

lect

ance

0.0

Brick 25 Conc 25 Veg 50

Conc 25 Soil 25 Veg 50


Brick25 Conc25 Soil25 Veg25

Conc

Brick SoilP

Veg

300 400 500 600 700 800 900 1000

Wavelength (nm)

0.0

0.2

0.4

0.6

0.8

1.0(c) (d)

Soil 0 Veg 100Soil 25 Veg 75

Soil 100 Veg 0

Soil 75 Veg 25

Soil 50 Veg 50

Ref

lect

ance

Ref

lect

ance

300 400 500 600 700 800 900 1000

Wavelength (nm)

0.5

0.4

0.3

0.2

0.1

0.0

Figure 2. Experimental reflectancewavelength plots obtained with the mixtures of differentendmembers under the FOV: (a) concrete and vegetation; (b) brick and vegetation; (c) soil andvegetation; and (d) these four objects tiled in different proportions of surface coverage; thepercentage of each endmember being indicated against the curves.

in figure 2(d) so that it could find a mixed pixel of circular area of approximately44 cm radius comprising equal proportions of four endmembers. Figure 2(d) showsthe resultant reflectancewavelength plots obtained for different proportions of thesefour endmembers.The solar irradiance spectrum was measured at 1 nm resolution throughout the

same UVvisNIR range with the same ASD instrument fitted with a remote cosinereceptor on the 25 FOV fibre and kept facing vertically upwards irrespective of thesolar elevation. Data were collected at different seasons and different atmospheric con-ditions placing the instrument at the same place and height in open air. Some of theirradiance data have been used as reference later.In the other phase of the work, satellite images of different spatial and spectral res-

olutions were analysed using ENVI 4.5 image-processing software (ITT Corporation,White Plains, NY, USA). A multispectral image procured by the Ocean ColourMonitor (OCM) sensor (ground resolution 360 m 236 m) of the Indian RemoteSensing Satellite IRS-P4 for a region around Chilika Lagoon (85 15 E85 30 Eand 19 30 N19 45 N) at the eastern coast of India was analysed. The pure end-member regions were selected by minimum noise fraction (MNF) transform and pixel

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


Table 1. Specifications of sensor OCM of satellite IRS-P4.

Bands Wavelength range (nm) Central wavelength (nm)

Band-1 404424 414.2Band-2 432452 441.4Band-3 479499 485.7Band-4 502522 510.6Band-5 547567 556.4Band-6 660680 669.0Band-7 748788 768.6Band-8 847887 865.1

Note: Ground resolution 360 m 236 m (Mishra et al. 2008).

purity index (PPI) techniques; the software having built-in facilities was used for theseanalysis techniques.The OCM consists of only eight bands within the visible and near-infrared (VNIR)

regions, as mentioned in table 1. To illustrate the spectral signature more accu-rately, the Hyperion hyperspectral images (ground resolution 30 m) for Kolkata cityand nearby area (centred around 22 35 N, 88 24 E) were downloaded from theUnited States Geological Survey (USGS) website. The DN values were convertedto reflectance (R) for 50 VNIR waveband channels (No. 857) using the followingformula:

R = Ld2

(ESUN) cos , (1)

where L =DN/40 is the radiance (Wm2 sr1 m1) as a function of wavelength, dis the EarthSun distance in astronomical units (=1), (ESUN) is the Hyperion meansolar exoatmospheric irradiance (Wm2 m1) as a function of wavelength and is thesolar zenith angle.

3. Results and discussion

Figure 1(a) compares the present experimental reflectancewavelength plots for con-crete (Conc), brick (Brick) and fresh banana leaf (Veg) with standard signatures ofconcrete (Conc-JHU), brick (Brick-JHU) and green grass (Veg-JHU), respectively,obtained from the spectral library of Johns Hopkins University (JHU).Figure 2(a)(d) represents the experimental reflectancewavelength plots obtained

with the mixtures of different endmembers under the FOV: concrete and vegetation;brick and vegetation; soil and vegetation; and these four objects in different propor-tions, respectively. The pure pixels are indicated by 100% and different proportionsof their mixtures are indicated against the curves. The general trend of the resultantsignature of the region under the FOV is found to approach that of vegetation as thevegetation-covered portion increases.Such mixed pixel signatures with different endmembers were also computationally

generated with the assumption of linear mixing using the individual experimentalspectra of pure endmembers. The general algorithm presented in figure 3 was usedto compute data for the mixed spectra for different combinations of two and four

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


Assumptions:

1=mi = 1

iw

where wi = weight of fractional abundance of the ith endmember such that 0 wi 1 m = number of endmembers, 2 or 4 in the present case n = number of wavelength channels, 700 in the present case (3001000 nm with 1 nm resolution)

Input data file: for i = 1 to m for j = 1 to n

Ei,j = measured reflectance of the ith endmember in the jth channel

Computation: for j = 1 to n for i = 1 to m

Rj = m

i = 1i,ji,j Ew

where Rj = resultant mixed pixel reflectance of the jth channel wi,j = weight of fractional abundance of the ith endmember in the jth channel Ei,j = measured reflectance of the ith endmember in the jth channel

Output data file: Mixed pixel reflectance (Rj, j = 1 to n) generated with m endmembers for n wavelength channels

Figure 3. A linear mixing model for generating mixed pixel spectral signatures. A generalalgorithm for computationally generating mixed pixel spectra for different possible linearcombinations of endmembers using their experimentally obtained individual spectral data.

endmembers. The results are illustrated in figure 4(a)(d) and may be compared withthe experimental results of figure 2(a)(d), respectively.Good agreement is obtained in most cases on comparing the original spectra with

those simulated with linear mixing of the individual spectra. Thus, it is possible tosynthesize the mixed pixel spectra and to predict the relative weights of differentendmembers in a mixed pixel spectrum provided that the spectra of the individualmembers are properly selected from the library.The next part of this work is the comparison of the spectral signature (wavelength

reflectance plot) extracted from the satellite image with that established by spectrora-diometry in laboratory as above. Let us start with the vegetation signature because ithas a typical nature and it is easily obtained in pure pixel form in the images.Figure 5(a) shows a portion of the OCM image consisting of Chilika Lagoon (A),

deep sea of the Bay of Bengal (B), vegetation/forest (C) and bare/cultivated soil (D)sorted out by MNF transform and identified by Google Earth imaging. The averagenormalized DN values for vegetation (C) and soil (D) zones were calculated for the

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


300 400 500 600 700 800 900 10000.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7(a)

Conc 100 Veg 0

Conc 25 Veg 75

Conc 50 Veg 50

Conc 25 Veg 75

Conc 0 Veg 100

Ref

lect

ance

Wavelength (nm)

(b)

300 400 500 600 700 800 900 1000

Wavelength (nm)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Brick 100 Veg 0

Brick 75 Veg 25

Brick 50 Veg 50

Brick 25 Veg 75

Brick 0 Veg 100

Ref

lect

ance

(c)

300 400 500 600 700 800 900 1000

Wavelength (nm)

0.0

0.2

0.4

0.6

0.8

1.0

Soil 100 Veg 0

Soil 75 Veg 25

Soil 50 Veg 50

Soil 25 Veg 75

Soil 0 Veg 100

Ref

lect

ance

(d)

300 400 500 600 700 800 900 1000

Wavelength (nm)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Brick25 Conc25 Soil25 Veg25

Brick 25 Soil 25 Veg 50Conc 25 Soil 25 Veg 50


Ref

lect

ance

Figure 4. The computationally synthesized reflectancewavelength plots for mixed pixels withdifferent proportions of endmembers: (a) concrete and vegetation; (b) brick and vegetation; (c)soil and vegetation; and (d) different proportions of these four simulated by linear mixing oftheir individual experimental spectra. The percentage of surface coverage for each endmemberis indicated against the curves.

400 500 600 700 800 9000.0

0.2

0.4

0.6

0.8

1.0

(b)

Veg 60 Soil 40Veg 40 Soil 60Soil 100

Veg 100

Nor

mal

ized

DN

val

ue

Wavelength (nm)

0 80 km

(a)

Figure 5. (a) the IRS P4 OCM image (85 15 E85 30 E and 19 30 N19 45 N) composedof bands 5, 6 and 7, dated 31 March 2007, showing Chilika Lagoon (A), deep sea (B), forestry(C) and bare/cultivated soil (D) and (b) average normalized DN values for vegetation (C) andsoil (D) zones.

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


(a) (b)

Figure 6. The Hyperion image composed of bands 42, 32 and 21 for Kolkata (centred on 2235 N, 88 24 E): (a) rainy season (27 July 2002) and (b) winter season (6 January 2010).

eight available wavebands and their fractional combinations were generated by linearsummation as plotted in figure 5(b). No definite result is apparent from the graph.However, careful observation reveals a regular change with the vegetationsoil ratio,especially at the redNIR transition region.With a hope for more accurate reflectance variation with wavelength, vegetation

spectral signatures were derived from the Hyperion hyperspectral satellite images.These are shown in figure 6(a) (rainy season: 27 July 2002) and figure 6(b) (winterseason: 6 January 2010). Obviously, there is also a wide span of time between thesetwo. A famous river named Ganga is visible there making it convenient to locate thecity. The spectral signatures of vegetation were derived from a number of randomlychosen pure pixel vegetated zones. Such regions are denoted as A in figure 6(a) and(b). A few such vegetation spectra derived from some randomly chosen zones of figure6(a) are shown in figure 7(a). The different pure vegetated zones yielded almost thesame spectral variation.Apparently, these are looking quite different from the spectral variations obtained

in the laboratory. The reasons are the absorptions due to oxygen (around 760 nm)and water vapour (around 810 and 900 nm). Figure 7(b) further clarifies it. The watervapour absorption is much reduced in winter as noted from the vegetation spectralsignature obtained from figure 6(b). Figure 7(c) clearly denotes the actual positionsof these absorption bands in the solar irradiance spectrum measured with this work.In the next step, the channel numbers 40, 41, 4548 and 5457 corresponding to theoxygen and water vapour absorptions were eliminated. The signatures were generatedwith the remaining channels and compared with the laboratory-derived signature, asshown in figure 7(d). It is apparent that although there is difference in the absolute val-ues of reflectance, the nature of spectral variation is the same in both laboratory- andsatellite-derived data. It indicates that it is possible to imitate the signature obtainedfrom the satellite image with a rigorous and methodical establishment of the spectrallibrary. Also the library can help to predetermine the suitable channels of the satelliteand reduce the number accordingly.

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


400 500 600 700 800 900 10000.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45(a)

Signatures from figure 6(a)R

efle

ctan

ce

Wavelength (nm)

(b)

400 500 600 700 800 900 1000

Wavelength (nm)

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

From figure 6(a) (rainy)

From figure 6(b) (winter)

Ref

lect

ance

(c)

400 500 600 700 800 900 1000

Wavelength (nm)

0.0

0.2

0.4

0.6

0.8

1.0

Water vapour

Oxygen

Water vapour

Rain

Winter

Nor

mal

ized

sol

ar ir

radi

ance

(d)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

400 500 600 700 800 9000.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

Ref

lect

ance

from

spe

ctro

radi

omet

ry

Wavelength (nm)

From spectroradiometry

From figure 6(a)

From figure 6(b)

Ref

lect

ance

from

Hyp

erio

n im

age

Figure 7. Vegetation signatures extracted from the Hyperion images: (a) from figure 6(a), (b)from figure 6(a) and (b) compared, (c) normalized solar irradiance in rainy and winter seasonsshowing the change in oxygen and water vapour absorption and (d) laboratory signature ofvegetation is compared with vegetation signatures extracted from figure 6(a) and (b) eliminatingthe oxygen and water vapour absorption bands.

The zones B and C of figure 6(a) are not as pure as vegetated zones like A.It has been verified by ground survey that zone B is a highly populated area con-sisting of concrete and asphalt and very less amount of vegetation. Zone C is partof a well-planned township consisting of uniformly distributed concrete and asphaltand containing larger quantities of vegetation in garden, on roadsides and so on. Thespectral signatures derived from several randomly chosen pixels of zone B are shownin figure 8(a). The oxygen and water absorption bands are similar to those of figure7(a). Eliminating these bands, as mentioned earlier, the sample signatures of zonesA, B and C are generated and plotted in figure 8(b). The change in the quantityof vegetation is obvious in the NIR region. Figure 8(c) shows the curves generatedby linear combination of concrete (Conc), asphalt (Asph) and vegetation (Veg) sig-natures in different proportions indicated against the curves. The asphalt signaturewas taken from the JHU spectral library and the rest were generated from this work.Comparing figure 8(b) and (c), it is understood that the relative weights of endmem-bers in a satellite image can be imitated by laboratory spectra and their permutationcombination.Figure 9(a) shows the spectral signatures derived from several randomly chosen pix-

els of zone D of figure 6(b). Such zones consist of slum housings with terracotta tile

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


400 500 600 700 800 900 1000

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45(a)

Zone-A

Zone-B

Zone-C

Ref

lect

ance

Wavelength (nm)

(b)

400 500 600 700 800 9000.10

0.15

0.20

0.25

0.30

0.35

0.40Zone-A

Zone-C

Zone-BRef

lect

ance

Wavelength (nm)

(c)

400 500 600 700 800 900

Wavelength (nm)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Conc 0 Asph 0 Veg 100



Ref

lect

ance

Figure 8. (a) The spectral signatures derived from several randomly chosen pixels of zonesA, B and C of figure 6(a); (b) the same signatures after eliminating the oxygen and watervapour absorption bands and (c) signatures generated by linear combination of concrete (Conc),asphalt (Asph) and vegetation (Veg) reflectances in different proportions. The percentage ofsurface coverage for each endmember is indicated against the curves.

roofs, as noted from ground observation. Figure 9(b) compares one such signaturewith that obtained from zone C, the absorption bands being removed from both. ZoneC is the counterpart of that in figure 6(a) with a time gap of almost 8 years. The tem-poral change during this span is not studied here. However, it is clear that zones Dand C are distinguishable signatures. Zone D mainly consists of terracotta whereaszone C consists of concrete, both mixed with some amount of soil and vegetation.The spectral signature of terracotta was verified in the laboratory and was found toexhibit the same nature as that of brick within the UVvisNIR range. Therefore, it isnot separately shown. Pure terracotta has a lower value of reflectance than pure con-crete within the visible region, as may be understood from the comparison of brickand concrete in figure 1(a). A similar feature is noted in figure 9(b) too. However, theNIR reflectance of zone D has exceeded that of zone C, which might be due to thelarger quantity of vegetation involved in the former. In order to guess the situation offigure 9(b), figure 9(c) generates some laboratory signatures with possible mixtures ofconcrete (Conc), asphalt (Asph), terracotta (Tc), soil and vegetation (Veg). It is notedthat a mixture somewhere between pure Tc (100%) and Tc (75%) plus Veg (25%) mayproduce spectral variation similar to that of zone D in figure 9(b), having lower visible

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


400 500 600 700 800 9000.05

0.10

0.15

0.20

0.25(a)

Ref

lect

ance

Wavelength (nm)

(b)

400 500 600 700 800 900

Wavelength (nm)

0.05

0.10

0.15

0.20Zone-D

Zone-C

Ref

lect

ance

(c)

400 500 600 700 800 900

Wavelength (nm)

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

Conc 30 Asph 30 Soil 25 Veg 15

Tc 75 Veg 25

Tc 100

Ref

lect

ance

Figure 9. (a) The spectral signatures derived from several randomly chosen pixels of zone D offigure 6(b); (b) a sample signature of zone D is compared with that of zone C after eliminatingthe oxygen and water vapour absorption bands in both and (c) some laboratory signatures ofmixtures of concrete (Conc), asphalt (Asph), terracotta (Tc), soil and vegetation (Veg). Thepercentage of surface coverage for each endmember is indicated against the curves.

and higher NIR reflectances than those of a mixture of Conc (30%) plus Asph (30%)plus Soil (25%) plus Veg (15%), which is a possible mixture of zone C. Thus it is under-stood that with certain combination of the endmembers, the laboratory signature canimitate the satellite-derived signature.

4. Conclusion

This work suggests a reverse method of standardizing the mixed pixel spectralresponse for a certain distribution of endmembers by synthesizing spectra with varyingproportions of spectral library data and matching them with the experimental result.This idea is demonstrated with a simple model having a linear combination of end-members. In one phase of this work, model pixels (pure and mixed) were generatedin the laboratory with different proportions of endmembers, concrete, brick, moistsoil and vegetation, and hyperspectral reflectance measurements were carried out withthem. The results were compared with the computationally generated data synthe-sized by linear mixing of pure spectral signatures. Good matching in spectral variationwas obtained in most cases. In the other phase of this work, hyperspectral signaturesderived from the Hyperion images were compared with laboratory signatures. Thetrends of spectral variation were found to be the same in both cases.

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


This work is a model to support the idea of spectral synthesis and it needs furtherdevelopment both in computational techniques and in generating spectral libraries.The constituent original spectra used for synthesizing the mixed spectra should bevery accurate, incorporating all possible variations. This work was demonstrated withonly one set of data for each endmember and simple linear mixing of them.However, the merit of such a technique is that the image need not be the sole

resource of information. The basic idea is to compare the image spectrum with thespectral library, where the reflectance quality is assured. Thus the computational jobis increased with the advantage of comparison even with an optimal number of satel-lite spectral bands. In fact, such a sampling of the optimal subset of the complete set ofhyperspectral bands has been suggested earlier in the context of feature classification(Serpico and Bruzzone 1994). Therefore the keen perfection and hence the technolog-ical complicacy of the satellite sensor may be a bit compromised with optimization inband selection, thereby reducing the data volume and transmission bandwidth.

AcknowledgementThe author thankfully acknowledges the financial support of the National ResourcesData Management System (NRDMS), Department of Science and Technology(DST), Government of India, New Delhi, India.

ReferencesALBEROTANZA, L., BRANDO, V.E., RAVAGNAN, G. and ZANDONELLA, A., 1999, Hyperspectral

aerial images. A valuable tool for submerged vegetation recognition in the OrbetelloLagoons, Italy. International Journal of Remote Sensing, 20, pp. 523533.

BOREL, C.C. and GERSTL, S.A., 1994, Nonlinear spectral mixing models for vegetative and soilssurface. Remote Sensing of Environment, 47, pp. 403416.

BRANDO, V.E. and DEKKER, A.G., 2003, Satellite hyperspectral remote sensing for estimat-ing estuarine and coastal water quality. IEEE Transactions on Geoscience and RemoteSensing, 41, pp. 13781387.

CHANG, C.-I. (Ed.), 2007, Hyperspectral Data Exploitation: Theory and Applications (NewJersey and Canada: John Wiley & Sons Inc.).

HOWARI, F.M., GOODELL, P.C. and MIYAMOTO, S., 2002, Spectroscopy of salts common insaline soils. In From Laboratory Spectroscopy to Remotely Sensed Spectra of TerrestrialEcosystems, Chapter1, R.S. Muttiah (Ed.), pp. 120 (Dordrecht: Kluwer Academic).

KWON, H. and NASRABADI, N.M., 2005, Kernel RX-algorithm: a nonlinear anomaly detectorfor hyperspectral imagery. IEEE Transactions on Geoscience and Remote Sensing, 43,pp. 388397.

MANOLAKIS, D., SIRACUSA, C. and SHAW, G., 2001, Hyperspectral subpixel target detectionusing linear mixing model. IEEE Transactions on Geoscience and Remote Sensing, 39,pp. 13921409.

MAZER, A.S., MARTIN, M., LEE, M. and SOLOMON, J.E., 1988, Image processing software forimaging spectrometry data analysis. Remote Sensing of Environment, 24, pp. 201210.

MISHRA, A.K., DADHWAL, V.K. and DUTT, C.B.S., 2008, Analysis of marine aerosol opticaldepth retrieved from IRS-P4 OCM sensor and comparison with the aerosol derivedfrom SeaWiFS and MODIS sensor. Journal of Earth System Science, 117, pp. 361373.

MOORHEAD, I.R., GILMORE, M.A., HOULBROOK, A.W., OXFORD, D.E., FILBEE, D., STROUD,C., HUTCHINGS, G. and KIRK, A., 2001, CAMEO-SIM: a physics-based broad-band scene simulation tool for assessment of camouflage, concealment, and deceptionmethodologies. Optical Engineering, 40, pp. 18961905.

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1


NOLIN, A.W. and DOZIER, J., 2000, A hyperspectral method for remotely sensing the grain sizeof snow. Remote Sensing of Environment, 74, pp. 207216.

PANFEROV, O., KNYAZIKHIN, Y., MYNENI, R.B., SZARZYNSKI, J., ENGWALD, S., SCHNITZLER,K.G. and GRAVENHORST, G., 2001, The role of canopy structure in the spectral vari-ation of transmission and absorption of solar radiation in vegetation canopies. IEEETransactions on Geoscience and Remote Sensing, 39, pp. 241253.

SALEM, F., KAFATOS, M., EL-GHAZAWI, G.R. and YANG, R.X., 2005, Hyperspectral imageassessment of oil-contaminated wetland. International Journal of Remote Sensing, 26,pp. 811821.

SERPICO, S.B. and BRUZZONE, L., 1994, A new search algorithm for feature selection in hyper-spectral remote sensing images. IEEE Transactions on Geoscience and Remote Sensing:Special Issue on Analysis of Hyperspectral Image Data, 39, pp. 13601367.

SETTLE, J.J., 1996, On the relationship between spectral unmixing and subspace projection.IEEE Transactions on Geoscience and Remote Sensing, 34, pp. 10451046.

VARSHNEY, P.K. and ARORA, M.K. (Eds.), 2004, Advanced Image Processing Techniques forRemotely Sensed Hyperspectral Data (Berlin: Springer).

ZARCO-TEJADA, P.J., USTIN, S.L. andWHITING,M.L., 2005, Temporal and spatial relationshipsbetween within-field yield variability in cotton and high-spatial hyperspectral remotesensing imagery. Agronomy Journal, 97, pp. 641653.

Dow

nloa

ded

by [B

arun R

aych

audh

uri]

at 05

:25 07

Dec

embe

r 201

1

mixed pixel

Documents

mixed pixel signature

pure spectral components

presidency university

different proportions

sisting of concrete

mortimer street

mortimer house

systematic supply