CCSM: Cross correlogram spectral matching F. Van Der Meer & W. Bakker Published online: 25 Nov 2010.

Similar documents
Gilles Bourgeois a, Richard A. Cunjak a, Daniel Caissie a & Nassir El-Jabi b a Science Brunch, Department of Fisheries and Oceans, Box

Published online: 05 Oct 2006.

Use and Abuse of Regression

The American Statistician Publication details, including instructions for authors and subscription information:

George L. Fischer a, Thomas R. Moore b c & Robert W. Boyd b a Department of Physics and The Institute of Optics,

Open problems. Christian Berg a a Department of Mathematical Sciences, University of. Copenhagen, Copenhagen, Denmark Published online: 07 Nov 2014.

Ankara, Turkey Published online: 20 Sep 2013.

Dissipation Function in Hyperbolic Thermoelasticity

University, Tempe, Arizona, USA b Department of Mathematics and Statistics, University of New. Mexico, Albuquerque, New Mexico, USA

Nacional de La Pampa, Santa Rosa, La Pampa, Argentina b Instituto de Matemática Aplicada San Luis, Consejo Nacional de Investigaciones Científicas

To cite this article: Edward E. Roskam & Jules Ellis (1992) Reaction to Other Commentaries, Multivariate Behavioral Research, 27:2,

Guangzhou, P.R. China

Testing Goodness-of-Fit for Exponential Distribution Based on Cumulative Residual Entropy

G. S. Denisov a, G. V. Gusakova b & A. L. Smolyansky b a Institute of Physics, Leningrad State University, Leningrad, B-

Online publication date: 30 March 2011

Diatom Research Publication details, including instructions for authors and subscription information:

FB 4, University of Osnabrück, Osnabrück

Published online: 17 May 2012.

Communications in Algebra Publication details, including instructions for authors and subscription information:

OF SCIENCE AND TECHNOLOGY, TAEJON, KOREA

Erciyes University, Kayseri, Turkey

Dresden, Dresden, Germany Published online: 09 Jan 2009.

Precise Large Deviations for Sums of Negatively Dependent Random Variables with Common Long-Tailed Distributions

Full terms and conditions of use:

The Fourier transform of the unit step function B. L. Burrows a ; D. J. Colwell a a

University, Wuhan, China c College of Physical Science and Technology, Central China Normal. University, Wuhan, China Published online: 25 Apr 2014.

Characterizations of Student's t-distribution via regressions of order statistics George P. Yanev a ; M. Ahsanullah b a

Tong University, Shanghai , China Published online: 27 May 2014.

Discussion on Change-Points: From Sequential Detection to Biology and Back by David Siegmund

The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features

Full terms and conditions of use:

Online publication date: 01 March 2010 PLEASE SCROLL DOWN FOR ARTICLE

PLEASE SCROLL DOWN FOR ARTICLE

Acyclic, Cyclic and Polycyclic P n

Geometric View of Measurement Errors

Park, Pennsylvania, USA. Full terms and conditions of use:

PLEASE SCROLL DOWN FOR ARTICLE

Online publication date: 22 March 2010

Version of record first published: 01 Sep 2006.

Derivation of SPDEs for Correlated Random Walk Transport Models in One and Two Dimensions

T. Runka a, M. Kozielski a, M. Drozdowski a & L. Szczepańska b a Institute of Physics, Poznan University of

Geometrical optics and blackbody radiation Pablo BenÍTez ab ; Roland Winston a ;Juan C. Miñano b a

Tore Henriksen a & Geir Ulfstein b a Faculty of Law, University of Tromsø, Tromsø, Norway. Available online: 18 Feb 2011

To link to this article:

France. Published online: 23 Apr 2007.

The Homogeneous Markov System (HMS) as an Elastic Medium. The Three-Dimensional Case

Xiaojun Yang a a Department of Geography, Florida State University, Tallahassee, FL32306, USA Available online: 22 Feb 2007

Published online: 27 Aug 2014.

PLEASE SCROLL DOWN FOR ARTICLE

C.K. Li a a Department of Mathematics and

University of Thessaloniki, Thessaloniki, Greece

Sports Technology Publication details, including instructions for authors and subscription information:

Projective spaces in flag varieties

MSE Performance and Minimax Regret Significance Points for a HPT Estimator when each Individual Regression Coefficient is Estimated

Published online: 04 Oct 2006.

PLEASE SCROLL DOWN FOR ARTICLE

PLEASE SCROLL DOWN FOR ARTICLE. Full terms and conditions of use:

PLEASE SCROLL DOWN FOR ARTICLE

Astrophysical Observatory, Smithsonian Institution, PLEASE SCROLL DOWN FOR ARTICLE

Quan Wang a, Yuanxin Xi a, Zhiliang Wan a, Yanqing Lu a, Yongyuan Zhu a, Yanfeng Chen a & Naiben Ming a a National Laboratory of Solid State

William Henderson a & J. Scott McIndoe b a Department of Chemistry, The University of Waikato, Hamilton,

PLEASE SCROLL DOWN FOR ARTICLE

Melbourne, Victoria, 3010, Australia. To link to this article:

A note on adaptation in garch models Gloria González-Rivera a a

Global Existence of Large BV Solutions in a Model of Granular Flow

PLEASE SCROLL DOWN FOR ARTICLE

To link to this article:

Published online: 04 Jun 2015.

Published online: 10 Apr 2012.

HYPERSPECTRAL IMAGING

P. C. Wason a & P. N. Johnson-Laird a a Psycholinguistics Research Unit and Department of

Avenue G. Pompidou, BP 56, La Valette du Var cédex, 83162, France

Online publication date: 12 January 2010

A Simple Approximate Procedure for Constructing Binomial and Poisson Tolerance Intervals

Marko A. A. Boon a, John H. J. Einmahl b & Ian W. McKeague c a Department of Mathematics and Computer Science, Eindhoven

Morris J. Robins a, Sanchita Sarker a & Stanislaw F. Wnuk a a Department of Chemistry, Biochemistry, Brigham Young

A Strongly Convergent Method for Nonsmooth Convex Minimization in Hilbert Spaces

35-959, Rzeszów, Poland b Institute of Computer Science, Jagiellonian University,

Communications in Algebra Publication details, including instructions for authors and subscription information:

Silvio Franz a, Claudio Donati b c, Giorgio Parisi c & Sharon C. Glotzer b a The Abdus Salam International Centre for Theoretical

Raman Research Institute, Bangalore, India

PLEASE SCROLL DOWN FOR ARTICLE. Full terms and conditions of use:

Yan-Qing Lu a, Quan Wang a, Yuan-Xin Xi a, Zhi- Liang Wan a, Xue-Jing Zhang a & Nai-Ben Ming a a National Laboratory of Solid State

Hyperspectral Data as a Tool for Mineral Exploration

Osnabrück, Germany. Nanotechnology, Münster, Germany

Fractal Dimension of Turbulent Premixed Flames Alan R. Kerstein a a

100084, People's Republic of China Published online: 07 May 2010.

PLEASE SCROLL DOWN FOR ARTICLE

PLEASE SCROLL DOWN FOR ARTICLE

Móstoles, Madrid. Spain. Universidad de Málaga. Málaga. Spain

ECONOMET RICS P RELIM EXAM August 24, 2010 Department of Economics, Michigan State University

Analysis of Leakage Current Mechanisms in BiFeO 3. Thin Films P. Pipinys a ; A. Rimeika a ; V. Lapeika a a

András István Fazekas a b & Éva V. Nagy c a Hungarian Power Companies Ltd., Budapest, Hungary. Available online: 29 Jun 2011

Adsorption of pyridine on dealuminated zeolite HY H. -C. Wang a ; H. Paul Wang b ; Kuen-Song Lin b a

A. H. Abuzaid a, A. G. Hussin b & I. B. Mohamed a a Institute of Mathematical Sciences, University of Malaya, 50603,

Online publication date: 29 April 2011

. (Al 2 SiO 5 ) Field Spec 3.

Online publication date: 22 January 2010 PLEASE SCROLL DOWN FOR ARTICLE

First published on: 11 November 2010

Transcription:

This article was downloaded by: [Universiteit Twente] On: 23 January 2015, At: 06:04 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK International Journal of Remote Sensing Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/tres20 CCSM: Cross correlogram spectral matching F. Van Der Meer & W. Bakker Published online: 25 Nov 2010. To cite this article: F. Van Der Meer & W. Bakker (1997) CCSM: Cross correlogram spectral matching, International Journal of Remote Sensing, 18:5, 1197-1201, DOI: 10.1080/014311697218674 To link to this article: http://dx.doi.org/10.1080/014311697218674 PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the Content ) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sublicensing, systematic supply, or distribution in any form to anyone is expressly

forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions

int. j. remote sensing, 1997, vol. 18, no. 5, 1197± 1201 C CSM: cross correlogram spectral matching F. VAN DER MEER International Institute for Aerospace Survey and Earth Sciences ITC, Department of Earth Resources Surveys, Hengelosestraat 99, P.O. Box 6, 7500 AA Enschede, The Netherlands and W. BAKKER International Institute for Aerospace Survey and Earth Sciences ITC, Department of Geoinformatics, Hengelosestraat 99, P.O. Box 6, 7500 AA Enschede, The Netherlands (Received 12 June 1996; in nal form 30 October 1996) Abstract. Cross correlogram spectral matching (CCSM) is a new approach towards mineral mapping from imaging spectrometer data. A cross correlogram is constructed by calculating the cross correlation at di erent match positions between a test spectrum (i.e., a remotely-sensed spectrum) and a reference spectrum (i.e., a laboratory mineral spectrum). To assess the sensitivity of the cross correlogram as a means of spectral matching, the technique is applied to laboratory spectra. In each experiment, the cross correlogram function was derived, a test of signi cance of the correlations was conducted and a moment of skewness was calculated to characterize the curve shape of the correlogram. The cross correlogram for a perfect spectral match had a parabolic shape with the maximum correlation of one at match position zero and a symmetry around the central match number. The cross correlation is found to be insensitive to di erences in gain and thus allows to compare materials of di erent albedos. The cross correlogram is relatively insensitive to noise which will reduce the overall correlation but does not a ect the shape of the correlogram. Relative small di erences in absorption band position and shape between the test and reference spectrum a ect the shape, signi cance and correlation of the correlogram values. 1. Introduction The advent of imaging spectrometers with typically more than 200 spectral bands requires new analytical tools that can be automated to a large extent to extract relevant information from hyperspectral data sets. Imaging spectrometers are devices that acquire data in many, narrow and contiguous spectral bands over large portions of the electromagnetic spectrum producing laboratory-like re ectance spectra for each pixel. Visual comparison of remote and laboratory spectra may allow positive identi cation of surface mineralogy. This is laborious and subjective. Alternatively, standard remote sensing techniques such as colour compositing and ratioing can be used. However, an enormous amount of possible two or three-band combinations arise from imaging spectrometer data sets. Several techniques have been developed to extract information in a semi-automated manner from imaging spectrometer data sets. It is outside the scope of this Letter to review these, the interested reader is referred to recent review papers by Cloutis (1996) and Ichoku and Karnieli (1996). In this letter, we present a new approach to hyperspectral data analysis that is based on the cross correlogram. A cross correlogram is constructed for each image 0143± 1161/97 $12.00 Ñ 1997 Taylor & Francis Ltd

1198 F. van der Meer and W. Bakker pixel by calculating the linear correlation coe cient between a test (i.e., a remotely sensed) spectrum and a reference (i.e., laboratory) spectrum at di erent match positions (i.e. spectral channel shifts). We demonstrate that the cross correlogram (e.g., the plot of match positions versus cross correlation) provides important statistical information that can be used to develop spectral curve matching algorithms that ultimately allow single-component mineral mapping as well as sub-pixel analysis. Through a set of experiments on laboratory spectra we will show the potential of the cross correlogram. 2. The cross correlogram for spectral matching 2.1. Basic calculations A cross correlogram is constructed by calculating the cross correlation coe cient between a test spectrum, usually a remotely-sensed spectrum, and a reference spectrum, usually a laboratory spectrum at di erent match positions (or lags). By convention, we move the reference spectrum and refer to a negative match position when shifting towards shorter wavelengths and to a positive match position when shifting toward a longer wavelength. Thus match position Õ 1 means that we are calculating the cross correlation between the test spectrum and the reference spectrum in which all channels have been shifted by one channel position number to the lower end of the spectrum. The cross correlation, rm, at each match position, m, is equivalent to the ordinary linear correlation coe cient and is de ned as the product of the covariance and the sum of the standard deviations as rm = COVt,r where COVt,r is the covariance between the overlapped portions of the test spectrum, t, and reference spectrum, r, and st and sr are the corresponding standard deviations. If we denote the test and reference spectrum as lt and lr, respectively, and de ne n as the number of overlapping positions, the cross correlation for match position m can be calculated as rm = stsr n lrlt Õ lr lt Ó [n l 2 r Õ ( lr ) 2 ][n l 2 t Õ ( lt ) 2 ] (1) (2) This is analogous to the approach (in a complete di erent context) of Mahle and Ashley (1979) and McFee et al. (1994). However, Mahle and Ashley (1979) did not use the correlation coe cient as a similarity index while McFee et al. (1994) only exploited the correlation at m=0 rather than calculating the full correlogram function. The signi cance of the cross correlation coe cient can be assessed by the following t-test S nõ 2 t=rm 1Õ r 2 m (3) which has (nõ 2) degrees of freedom and tests the null hypothesis stating that the correlation between the two spectra at a speci c match position is zero.

Remote Sensing L etters 1199 2.2. Sensitivity analysis To test the sensitivity of the cross correlogram as a tool for spectral matching, we applied the method in a number of experiments to re ectance spectra from the NASA-JPL laboratory spectral library (Grove et al. 1992). Spectra in this library were measured on a Beckman UV 5240 spectrophotometer which has a sampling interval of 1 nm in the wavelength range of 0 4 to 0 8 mm, and a sampling interval of 4 nm in the wavelength range of 0 8 to 2 5 mm. Bandwidth ranges from 1 nm at 0 4 mm to 40 nm at 2 5 mm with a spectral resolution (de ned as bandwidth/wavelength) better than 2 per cent at all wavelengths. As a reference spectrum we have selected kaolinite between 2 0 and 2 3 mm and compared it with two other clay minerals; alunite and buddingtonite. Kaolinite has a strong absorption feature at 1 4 mm and a double absorption feature centred at 2 16 mm and 2 2 mm. Due to the lack of H2O, the feature at 1 9 mm is weakly developed or missing. Alunite is characterized by absorption features at 2 16 mm and 2 20 mm. Due to OH frequency stretching and a nearly symmetrical shape in the 2 08± 2 28 mm region. A second broad absorption feature occurs at 2 32 mm. An absorption feature at 2 02 mm and a vibrational absorption feature due to NH4 at 2 11 mm are the main diagnostic features distinguishing buddingtonite spectrally from other clay minerals. Experiment 1 was a cross correlogram with the kaolinite spectrum both as test and reference spectrum in the wavelength region of the characteristic doublet. The correlogram found was symmetric around the match position zero which also has the highest correlation value of one. The symmetry can be expressed in the moment of skewness which we calculate as the correlation at match position minus ten subtracted from the correlation at match position plus ten. The results of the t-test indicated that for all match positions the correlation was found to be signi cant indicating that the two spectra are dependent, non-random series. To demonstrate that the cross correlogram is insensitive to di erences in albedo, in another experiment the test spectrum was o set by 10 per cent re ectance which had no e ect on the correlogram ( gure 1). The e ect of noise on the correlogram was tested by adding a component of 5 per cent random noise to the test spectrum which did not a ect the shape characteristics of the correlogram, but resulted in a decrease of the cross correlation values by 5 per cent ( gure 2). Adding 10 per cent random noise to the test spectrum and 5 per cent random noise to the reference spectrum caused a minute e ect on the shape characteristic of the correlogram (skewness of 0 0094) and a decrease of the cross correlation values of 7 5 per cent. The cross correlogram for the spectral matching of kaolinite (as reference spectrum) versus buddingtonite, alunite and kaolinite (as test spectrum) is shown in gure 3. The function for buddingtonite is highly skewed with the negative moment of skewness indicating that the peak of the correlation coe cients is found when shifting the kaolinite spectrum toward shorter wavelengths. Note that the correlation coe cients were found insigni- cant for all match positions. The cross correlogram for alunite versus kaolinite shows a peak at match position minus six which corresponds to a shift of the kaolinite reference spectrum of 24 nm toward shorter wavelength. 3. Conclusions Cross correlogram spectral matching (CCSM) is a simple and straightforward technique that allows the quanti cation of the spectral similarity existing between remotely-sensed and laboratory spectra. Assessment of the statistical signi cance of the spectral match is an integral part of the matching process. The cross correlogram

1200 F. van der Meer and W. Bakker Figure 1. Library and test spectrum ( left) and corresponding cross correlogram (right) for kaolinite versus kaolinite o set by 10 per cent re ectance. The vertical dashed line on the top of the right-hand chart indicates the match positions for which the correlation coe cients were found to be signi cant according to the t-test of equation (2). Figure 2. Library and test spectrum ( left) and corresponding cross correlogram (right) for kaolinite versus kaolinite with 5 per cent random noise added. The vertical dashed line on the top of the right-hand chart indicates the match positions for which the correlation coe cients were found to be signi cant according to the t-test of equation (2). outlined is insensitive to gain factors and random noise a ects the values of the correlation coe cients equally at each match position without a ecting the shape of the correlogram. The position of the correlation peak in the correlogram is directly related to the relative position of the spectral absorption features that are being compared and the t-test provides a means of testing the signi cance of the correlation coe cients found. Finally, the cross correlogram provides a statistically meaningful comparison between test and reference spectra that can lead to automated mineral

Remote Sensing L etters 1201 Figure 3. Library and test spectrum ( left) and corresponding cross correlogram (right) for kaolinite versus kaolinite, alunite and buddingtonite. The vertical lines in the diagram above the right-hand chart indicate the match positions for which the correlation coe cients were found to be signi cant according to the t-test of equation (2). Note that for the mineral buddingtonite all correlation coe cients were insigni cant and that the peak of the correlogram occurs outside of the mapped area to the left. mapping using imaging spectrometer data. We are currently developing and testing algorithms to apply CCSM to imaging spectrometer data by addressing problems related to the mixed nature of remotely-sensed spectra and problems arising from di erences in bandpasses and spectral response function of eld and imaging spectral devices. Referen ces Cloutis, E. A., 1996, Hyperspectral geological remote sensing: evaluation of analytical techniques. International Journal of Remote Sensing, 17, 2215± 2242. Grove, C. I., Hook, S. J. and Paylor, II, E. D., 1992, L aboratory Re ectance Spectra of 160 minerals, 0 4 to 2 5 Micrometers (Pasadena: Jet Propulsion Laboratory), NASA-JPL Publication 92± 2. Ichoku, C. and Karnieli, A., 1996, A review of mixture modelling techniques for sub-pixel land cover estimation. Remote Sensing Reviews, 13, 161± 186. Mahle, N. H. and Ashley, J. W., 1979, Application of a correlation coe cient pattern recognition technique to low resolution mass spectra. Computers & Chemistry, 3, 19± 23. McFee, J. E., Achal, S. and Anger, C., 1994. Scatterable mine detection using CASI. Proceedings of the First International Airborne Remote Sensing Conference and Exhibition, Strasbourg, France, 12± 15 September 1994 (Ann Arbor: ERIM), pp. 587± 599.