with a few relatively broad wavelength bands. Hy-narrow, adjacent spe - PDF document

with a few relatively broad wavelength bands. Hy-narrow, adjacent spectral bands. These measure-tration below. After adjustments for sensor,atmospheric, and terrain effects are applied, thesesurements actually made by the hyperspectral sensor. many narrow, adjacent wavelength Plot Wavelength (micrometers) 0.00.20.40.62.21.71.20.7 A plot of the brightness valuesversus wavelength shows thecontinuous spectrum for theimage cell, which can be usedto identify surface materials. spectral mixing. Factorson pages 11-13, followed Welcome to Hyperspectral Imaging page 4Introduction to Hyperspectral Imaging The Imaging Spectrometerimaging spectrometersThe development of these complex sensors has involved the convergence of two of Earthand its variation in energy with wavelength. As applied to the field of opticalremote sensing, spectroscopy deals with the spectrum of sunlight that is diffuselyreflected (scattered) by materials at the Earth’s surface. Instruments called spec-measurements of the light reflected from a test material. An optical dispersingnarrow, adjacent wavelength bands and the energy in each band is measured bya separate detector. By using hundreds or even thousands of detectors, spec-adjacent areas on the Earth’s surface. In many digital imagers, sequential mea-image. Until recently, imagers were restricted to one or a few relatively broadstorage, transmission, and processing. Recent advances in these areas have al- Element Imaging



Light froma singleground- trometer. Some sensors use spectral reflectance: the ratio of reflected energy to incident energy as a func-tion of wavelength. Reflectance varies with wavelength for most materials becauseenergy at certain wavelengths is scattered or absorbed to different degrees. These(plots of reflectance versus wavelength) for different materials, as in the illustra-tion below. Pronounced downward deflections of the spectral curves mark thewavelength ranges for which the material selectively absorbs the incident energy. in a multispectral or hyperspectral image). The overallmany cases can be used to identify and discriminate different materials. Forterials over the visible light to reflected infrared spectral range. The spectral bandsor it can be expressed as a percentage, as in this graph. When spectral measure-ments of a test material are made in the field or laboratory, values of incidentenergy are also required to calculate the material’s reflectance. These values arethe same illumination conditions as the test material) from a standard reference Vegetation Red Near InfraredMiddle Infrared 1234 7 123 Wavelength (micrometers)1.02.00.61.21.41.61.82.22.4 Wet soil Turbid river water 060 In inorganic materials such as minerals, chemical composition and crystallinespecific absorption bands. Wavelength-specific absorption may be caused by thepresence of particular chemical elements or ions, the ionic charge of certain ele-examples of these effects. In the spectrum of hematite (an iron-oxide mineral), the). Inin montmorillonite is caused by bound water molecules in this hydrous clay. Incontrast to these examples, orthoclase feldspar, a dominant mineral in granite, Wavelength (micrometers) VisibleNear InfraredMiddle Infrared 1.02.00.61.21.41.61.82.22.4 Reflectance spectra of different types of green vegetation compared to a spectralcurve for senescent (dry, yellowed) leaves. Different portions of the spectral curvesfor green vegetation are shaped by different plant components, as shown at the top.shape that is dictated by various plant attributes. In the visible portion of thespectrum, the curve shape is governed by absorption effects from chlorophylland other leaf pigments. Chlorophyll absorbs visible light very effectively butteristic small reflectance peak within the green wavelength range. As aconsequence, healthy plants appear to us as green in color. Reflectance risesred edgenal cellular structure of leaves. Most of the remaining energy is transmitted, andcan interact with other leaves lower in the canopy. Leaf structure varies signifi-cantly between plant species, and can also change as a result of plant stress. Thusspecies type, plant stress, and canopy state all can affect near infrared reflectancemeasurements. Beyond 1.3 At the end of the growing season leaves lose water and chlorophyll. Near infra-red reflectance decreases and red reflectance increases, creating the familiar yellow,brown, and red leaf colors of autumn.Wavelength (micrometers) Reflectance (%)Grass Walnut tree canopy Dry, yellowed VisibleChlorophyllCell StructureWaterWater 1.02.00.61.21.41.61.82.22.4 Spectral Library.NASA’s Earthavailable for public use. These libraries provide a source of reference spectrathat can aid the interpretation of hyperspectral and multispectral images. This library has been made available by NASA as partTER) imaging instrument program. It includes spectral compilations from NASA’sJet Propulsion Laboratory, Johns Hopkins University, and the United States Geo-logical Survey (Reston). The ASTER spectral library currently contains nearly2000 spectra, including minerals, rocks, soils, man-made materials, water, andsnow. Many of the spectra cover the entire wavelength region from 0.4 to 14 The library is accessible interactively via the Worldwide Web at http://speclib.jpl.nasa.gov. You can search for spectra by category, view a spectral plota text file. These spectra can be imported into a TNTmips spectral library. You canalso order the ASTER spectral library on CD-ROM at no charge from the above The United States Geological Survey SpectroscopyLab in Denver, Colorado has compiled a library of about 500 reflectance spectram. Thishttp://speclab.cr.usgs.gov/spectral.lib04/spectral-lib04.htmlYou can browse individual spectra online, or download the entire library. The Wavelength (micrometers)VisibleNear InfraredMiddle Infrared 1.02.00.61.21.41.61.82.22.4 the differences in spectral properties between different materials, especially whenwe are comparing only a few spectra. Spectral plots are an important tool to usewhen you explore a hyperspectral image. But to understand how a computercompares and discriminates among a large number of spectra, it is useful to con-tral channel (band). Each of these channels can be considered as one dimensionchannels. If we plot the measured reflectance value for each spectral channel onspectrum. A simple two-band example isshown in the illustration. The designated pointthe coordinate system. Spectra with the sameshape but differing overall reflectance (al-but with endpoints at different distances fromthe origin. Shorter spectral vectors representIt may be difficult to visualize such a plot forlength bands, but it is mathematically possible to construct a hyperdimensionalnate axes. Each spectrum being considered occupies a position in thisn-dimensional spectral space. Similarity between spectra can be judged by theals represent “averages” or “typical examples”. All natural materials exhibit Reflectance in Band 1Reflectance in Band2 Point representingspectrum (0.8, 0.7) Vector representing Spatial Resolution and Mixed Spectrathe Earth’s surface, each of which is represented as a pixel (raster cell) in theon the sensor design and the height of the sensor above the surface. NASA’sAirborne Visible/Infrared Imaging Spectrometer (AVIRIS), for example, has aresolution cell is large, it issured by the sensor. The or spectrum, and themacroscopic or intimate. In a macroscopic mixture each reflected photon inter-acts with only one surface material. The energy reflected from the materialscombines additively, so that each material’s contribution to the composite spec-trum is directly proportional to its area within the pixel. An example of such aof vegetation and bare soil. In spectral spacethe corner of a mixing space (for greater num-bers of endmembers). Later we will discusslated for each pixel. In an intimate mixture,teracts with more than one material. Such Wavelength (micrometers)0.40.60.81.01.21.41.61.82.02.22.4 2-band case. All spectra that are Reflectance in Band 2Spectrum AReflectance in Band 1 Spectrum C Spectrum BMixing Radiance and Reflectancesure precisely and accurately using an airborne or satellite hyperspectral sensor.But look at the brightness spectrum in the illustration below. This is the averageof 25 image spectra measured by the AVIRIS sensor over a bright dry lake bedsensor effects using on-board calibration data, but no other transformations havetrated previously. This is because the sensor has simply measured the amount ofcase from an altitude of 20 kilometers. The spectral reflectance of the surfacematerials is only one of the factors affecting these measured values. The spectralIn addition to surface reflectance, the spectral radiance measured by a remotesensor depends on the spectrum of the input solar energy, interactions of thisenergy during its downward and upward passages through the atmosphere, theof the sensor system. These additional factors not only affect our ability to re-image cells. These factors are discussed in more detail on the next two pages. Wavelength, (micrometers)0.51.01.52.02.5 Averaged measured Source Illumination The figure below shows a typical solar irradiance curvefor the top of the Earth’s atmosphere. The incoming solar energy varies greatlywith wavelength, peaking in the range of visible light. The spectrum of incomingsolar energy at the time an image was acquired must be known, assumed, orIllumination Geometry The amount of energy reflected by an area on the grounddepends on the amount of solar energy illuminating the area, which in turn de-energy and a line perpendicular to the ground surface. Specifically, the energy cos , where Eo is the amount of incoming energy. The energyreceived by any ground area therefore varies as the sun’s height changes withtime of day and season. If the terrain is not flat, the energy received also variesinstantaneously across a scene becauseof differences in slope angle and direc-reduced by shadows. Shadows cast byimage cells. Trees, crop rows, rock out-image cell. Both types of shadows have the effect of lowering the measuredbrightness across all wavelengths for the affected pixels.Illumination FactorsIllumination differences can arise fromdiffering incidence angles ( ABC Wavelength (micrometers, 1.01.5 Even a relatively clear atmosphere interacts with incom-ing and reflected solar energy. For certain wavelengths these interactions reduceof reflected energy reaching an airborne or satellite sensor. The transmittance ofmolecules and particulates. These effects combine to produce the transmittancecurve illustrated below. The pronounced absorption features near 1.4 and 1.9energy almost completely, so little useful information can be obtained from im-age bands in these regions. Not shown by this curve is the effect of light scatteredupward by the atmosphere. This scattered light adds to the radiance measured by. Atmospheric effects may also differ between areas in a single scene ifelevation differences that vary the path length of radiation through the atmo- A sensor converts detected radiance in each wavelength channelthat represent “encoded” radiance values. Variations between detectors withinan array, as well as temporal changes in detectors, may require that raw measure-ments be scaled and/or offset to produce comparable values.Plot of atmospheric transmittance versus wavelength for typical atmospheric con-ditions. Transmittance is the proportion of the incident solar energy that reaches theground surface. Absorption by the labeled gases causes pronounced lows in thecurve, while scattering is responsible for the smooth decrease in transmittance with Wavelength (micrometers)Transmittance0.51.01.52.02.5 2O2 O3 VisibleNear InfraredMiddle Infrared CO2CO2O2H22 reflectance. A comprehensive conversion must account for the solar source spec-trum, lighting effects due to sun angle and topography, atmospheric transmission,and sensor gain. In mathematical terms, the ground reflectance spectrum is mul-tiplied (on a wavelength per wavelength basis) by these effects to produce themeasured radiance spectrum. Two other effects contribute in an additive fashionto the radiance spectrum: sensor offset (internal instrument noise) and path radi-ance due to atmospheric scattering. Several commonly used reflectance conversionstrategies are discussed below and on the following page. Some strategies useclude a uniform area that has a relatively flat spectral reflectance curve. Themean spectrum of such an area would be dominated by the combined effects ofsolar irradiance and atmospheric scattering and absorption The scene is con-mean spectrum. The selected flat field should be bright in order to reduce theeffects of image noise on the conversion. Since few if any materials in naturalfield” is difficult for most scenes. For desert scenes, salt-encrusted dry lake bedscrete may serve in urban scenes. Any significant spectral absorption features inreflectance spectra. If there is significant elevation variation within the scene,the converted spectra will also incorporate residual effects of topographic shad-ing and atmospheric path differences.Average Relative Reflectance Conversion entire image. Before computing the mean spectrum, the radiance values in eachThis adjustment largely removes topographic shading and other overall bright-ing a mean spectrum similar to the flat field spectrum described above. ThisReflectance Conversion I Reflectance Conversion IIcount for multiplicative contributions to the image spectra. Most studies thatm) for which the additive effect of atmospheric pathradiance is minimal. If the spectra to be analyzed include the visible and nearinfrared ranges, however, path radiance effects should not be neglected. If the Field researchers using hyperspectral imagery typicallyto reflectance. Field reflectance spectra must be acquired from two or more uni-form ground target areas. Target areas should have widely different brightnessand be large enough to recognize in the image. Using the image radiance andadditive component (offset). These values are thentance. The final values should be considered Radiative-transfer computerand atmospheric scattering and absorption. In the absence of measurements ofsuch as amount and distribution of scattering agents. Absorption by well-mixedto water vapor is often variable. Water vapor absorption effects can be estimatedinclude water absorption bands. The final apparent reflectance values may stillincorporate the effects of topographic shading, however. Reflectance Bright targetDark targetSlope = gainIntercept = offset Reflectance conversionimage band using knowntarget reflectance values. Strategies for Image Analysisfor research or commercial purposes. The hyperspectral images produced bythese sensors present a challenge for the analyst. They provide the fine spectralthe volume of data in a single scene can seem overwhelming. The difference insmall and their grayscale images therefore appear nearly identical. Much of theals. Finding appropriate tools and approaches for visualizing and analyzing thevidual bands or groups of bands. The statistical classification (clustering) methodspress). More sophisticated methods combine both spectral and spatial analysis. rosneS oitazinagrO rtnuoC ebmuNsdnaBfo tgnelevaW(egnaRµµµµµ)m IRIVA SAN etinUsetatS 22 .2-4.0 SIA dtLgnigamIlartcepS nalniF 82 .0-54.0 SAC craeseRsertI danaC 82 8.0-34.0 IAD5112 proCREG etinUsetatS 12 .21-4.0 AMYH cinortcepSdetargetnIdtLytP ilartsuA 21 4.2-4.0 -EBORP ecneicShcraeShtraE.cnI etinUsetatS 21 4.2-4.0 Match Each Image Spectrumtral library. This approach requires an accurate conversion of image spectra toreflectance. It works best if the scene includes extensive areas of essentially purematerials that have corresponding reflectance spectra in the reference library. Ansome measure of goodness of fit, with the best match designated the “winner.”materials (see page 10). The re-present. If the best-matching reference spectrum has a sufficient fit to the imagepixel is assigned to this material. If no reference spectrum achieves a sufficientfor most of the image cells, such as the example shown below. Sample mixederence library.Cuprite AVIRIS scene,Spectral Library. Whitesufficient match to any of Alunite + KaoliniteMontmorillonite 2.12.22.3Wavelength (micrometers) Spectral Matching Methodsseveral absorption features. Curve B shows the 0.51.52.5Wavelength (1.02.0spectrum and narrow, trough-like absorption features. This distinction leads totwo different approaches to matching image spectra with reference spectra.(depth), and shape of their absorption features. One common matching strategytrum and ignores other parts of the spectrum. A unique set of wavelength regionsits absorption features. The local position and slope of the spectrum can affect: the upper limit of the spectrum’s generalshape. The continuum is computed for each wavelength subset and removed byvalue. Absorption features can then be matched using a set of derived valuesthe feature. These typessorption features. Thesewavelength subset for all candidate materials. One approach to matching seeks page 19Introduction to Hyperspectral ImagingLinear Unmixing Portion of an AVIRISand a river, shown withis red). Fraction imagesshown below.Vegetation fractionWater / shade fractionspectral matching. Its underlying premise is that a scenetions, of these common endmember components. Ifematically “unmix” each pixel’s spectrum to identifyfractions should sum to 1.0. The best-fitting set of frac-procedure described on the previous page. A fractionimage for each endmember distills the abundance in-manipulated. An image showing the residual error forcal components on the surface. Endmembers can beoutlined on the next page can be used. Alternatively,reference library, but this approach requires that theVariations in lighting can be included directly in thecan mix with the actual material spectra. A shade spec-portion of the image. In the absence of deep shadows, Defining Image Endmembersrepresents. (Ideally, each endmember would be a single pure material, but “pure”pixels of each endmember may not be present in the image). If image spectra areeral steps. Because of the high degree of correlation between adjacent spectralnoise-free components. The MNF transform (Green et al. 1988) is a noise-ad-their amount of image content. Second, an automated procedure is applied to theMNF components to find the extreme spectra around the margins of the n-dimen-sional data cloud. One such procedure is the Pixel Purity Index (PPI). It examinescoordinate space. For each test direction, all spectral points are projected ontothe test vector, and the extreme spectra (low and high) are noted. As directionsextreme. Pixels with high values in the resulting PPI raster should correspondprimarily to the edges of the MNF data cloud. In the third step, the PPI raster isn-Dimensional Visualizer in the TNTmipsHyperspectral Analysis process). Bypoints that are extreme in those directions,Finally, the marked cell image is overlaiddex procedure. All spectral points areprojected to each test vector, and ex- MNF Component 1MNF Component 2 Extreme spectra page 21Introduction to Hyperspectral Imaging Partial Unmixingdance of all endmember components in the scene. Instead the objective may beto detect the presence and abundance of a single target material. In this case acomplete spectral unmixing is unnecessary. Each pixel can be treated as a poten-all other materials in the scene. Finding the abundance of the target componentprocessing. Various matched filtering algorithms have been developed, includ-ing orthogonal subspace projection and constrained energy minimization (Farrandand Harsanyi, 1994). All of these approaches perform a mathematical transfor-mation of the image spectra to accentuate the contribution of the target spectrumwhile minimizing the background. In a geometric sense, matched filter methodsabundance of the target spectrum but “hides” the variability of the background.signature from the image itself. Some methods only work well when the targetthan the spectra themselves, which improves the matching of spectra with differ- Filtering (right) for a portion of the Cuprite AVIRIS scene. The target imagespectrum represents the mineral alunite. Brighter tones indicate pixels withhigher alunite fractions. The image produced by Derivative Matched Filteringareas with differing alunite fractions. Kruse, F.A. (1999). Visible-Infrared Sensors and Case Studies. In Renz, Andrew (3rd ed.), Vol 3. New York: John Wiley & Sons, pp. 567-611.Processing for Remote Sensing. River Edge, NJ: World Scientific Publish-ing Company, pp. 3-38.Vane, Gregg, Duval, J.E., and Wellman, J.B. (1993). Imaging Spectroscopy of theVane, Gregg, and Goetz, A.F.H. (1988). Terrestrial Imaging Spectroscopy. mote Sensing of EnvironmentBen-Dor, E., Irons, J.R., and Epema, G.F. (1999). Soil Reflectance. In Renz, Andrew (3rd ed.), Vol 3. New York: John Wiley & Sons, pp. 111-188.Spectroscopy. In Renz, Andrew N. (ed), (3rd ed.), Vol 3. New York: JohnWiley & Sons, pp. 3-58.Ustin, S.L., Smith, M.O., Jacquemoud, S., Verstraete, M., and Govaerts, Y. (1999).Geobotany: Vegetation Mapping for Earth Sciences. In Renz, Andrew N.(3rd ed.), Vol 3. New York: John Wiley & Sons, pp. 189-248.Reflectance ConversionFarrand, William H., Singer, R.B., and Merenyi, E., 1994, Retrieval of ApparentSurface Reflectance from AVIRIS Data: A Comparison of Empirical Line,Radiative Transfer, and Spectral Mixture Methods. References ReferencesGoetz, Alexander F.H., and Boardman, J.W. (1997). Atmospheric Corrections:On Deriving Surface Reflectance from Hyperspectral Imagers. In Descour,of SPIE, 3118, 14-22.van der Meer, Freek (1994). Calibration of Airborne Visible/Infrared ImagingSpectrometer Data (AVIRIS) to Reflectance and Mineral Mapping inAdams, John B., Smith, M.O., and Gillespie, A.R. (1993). Imaging Spectros-copy: Interpretation Based on Spectral Mixture Analysis. In Pieters, Carle. Cambridge, UK: CambridgeClark, R.N., Gallagher, A.J., and Swayze, G.A. (1990). Material absorption bandProceedings of the Sec-ond Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) WorkshopCloutis, E.A., (1996). Hyperspectral Geological Remote Sensing: Evaluation ofAnalytical Techniques. Farrand, William H., and Harsanyi, J.C. (1994). Mapping Distributed Geologi-cal and Botanical Targets through Constrained Energy Minimization.Proceedings of the Tenth Thematic Conference on Geological RemoteSensing, San Antonio, Texas, 9-12 May 1994, pp. I-419 - I-429.Green, Andrew A., Berman, M., Switzer, P., and Craig, M.D. (1988). A Trans-formation for Ordering Multispectral Data in Terms of Image Quality withIEEE Transactions on Geoscience andMustard, John F., and Sunshine, J.M. (1999). Spectral Analysis for Earth Sci-(3rd ed.), Vol 3. New York: John Wiley & Sons, pp. 251-306. page 1Introduction to Hyperspectral Imaging HyperspectralImaging Introduction to INTROTOHYPRS Introduction to Hyperspectral Imagingpage 24 Advanced Software for Geospatial Analysis INTROTOHYPERS MicroImages,Inc. www.microimages.com absorption by...........................13,18scattering by.................................13irradiance, solar.......................................12spectroscopy.........................................4,5in library.........................................8solar..............................................12water...............................................5analysis, CAD, TIN, desktop cartography, and geospatial database management.TNTmips Freeals with small projects. You can download TNTmips Free from MicroImages’ web site.TNTeditTNTedit provides interactive tools to create, georeference, and edit vector, image,TNTatlasTNTatlas lets you publish and distribute your spatial project materials on CD orDVD at low cost. TNTatlas CDs/DVDs can be used on any popular computing platform. Before Getting Started You can print or read this booklet in color from MicroImages’ web site. Theweb site is also your source for the newest Getting Started booklets on othertopics. You can download an installation guide, sample data, and the latestresearchers. With the recent appearance of commercial airborne hyperspectralVisible/Infrared Imaging Spectrometer (AVIRIS), which is operated by the NASAJet Propulsion Laboratory. The same scene is used in the exercises in the com-. You can download thisanalyzed using the Hyperspectral Analysis process (choose Image / Hyperspectral. Additional back-Pro), the low-cost TNTmips Basic version, and the TNTmips Free version. All