This tutorial is designed to show you how ENVI's advanced hyperspectral tools can be used for analysis of multispectral data. To gain a better understanding of the hyperspectral concepts and tools, please see the ENVI hyperspectral tutorials. For additional details, please see the ENVI User's Guide or the ENVI On-Line help.
You must have the ENVI TUTORIALS & DATA CD-ROM mounted on your system to access the files used by this tutorial, or copy the files to your disk.
The files used in this tutorial are contained in the BH_TMSUB subdirectory of the ENVIDATA directory on the ENVI TUTORIALS & DATA CD-ROM.
The files listed below are required to run this exercise.
BHTMREF.IMG Bighorn Basin, Wyoming Landsat TM Reflectance
BHTMREF.HDR ENVI Header for Above
BH_RATS.IMG Band Ratio Image 5/7, 3/1, 3/4 9RGB)
BH_RATS.HDR ENVI Header for Above
BHTMISO.IMG Isodata Classification
BHTMISO.HDR ENVI Header for Above
BHISIEV.IMG Sieve Image of Isodata Classification
BHISIEV.HDR ENVI Header for Above
BHICLMP.IMG Clump Image of the Sieved Classification
BHICLMP.HDR ENVI Header for Above
BHTM.GRD Saved Grid for Bighorn TM data
BHTMISO.ANN Map Annotation for Bighorn TM data
BHTM_MNF.ASC ACII Eigenvalue data for MNF Transform
BHTM_MNF.IMG MNF Transform Data
BHTM_MNF.HDR ENVI Header for Above
BHTM_MNF.STA ENVI Statistics File for above
BHTM_NS.STA ENVI Noise statistics for above
BHTM_PPI.IMG Pixel Purity Index Image
BHTM_PPI.HDR ENVI Header for Above
BHTM_PPI.CNT Counter for PPI analysis
BHTM_PPI.roi Regions of Interest threshold from the PPI
BHTM_PP.NDV n-D Visualizer Save State file
BHTM_ND.ROI ROIs from n-D Visualizer Analysis
BHTM_EM.ASC ASCII Spectral Endmembers from n-D
BHTM_SAM.IMG SAM Classification
BHTM_SAM.HDR ENVI Header for Above
BHTM_SAM.ANN Map Annotation for the SAM images
BHTM_RUL.IMG ENVI SAM Rule Images
BHTM_RUL.HDR ENVI Header for Above
BHTM_UNM.IMG Linear Spectral Unmixing Result
BHTM_UNM.HDR ENVI Header for Above
BHUNM_EM.ASC Endmembers used for Spectral Unmixing
ENVI was not designed solely as a hyperspectral image processing system. The decision was made in 1992 to develop a general purpose image processing software package with a full suite of standard tools in response to the general lack of powerful yet flexible commercial products capable of handling a wide variety of scientific image data formats. This included support for panchromatic, multispectral, hyperspectral, and both basic and advanced radar systems. ENVI presently contains most of the same basic capabilities as other major image processing systems such as ERDAS, ERMapper, and PCI. Where ENVI differs is in the many advanced, state-of-the-art algorithms resulting from active leading-edge remote sensing research. While many of these features were developed specifically to deal with imaging spectrometer data or "hyperspectral" data having up to hundreds of spectral bands, many of these techniques are applicable to multispectral data and other standard data types. This tutorial presents a scenario for use of some of these methods for analysis of Landsat Thematic Mapper data.
This example is broken into two portions: 1) a typical multispectral analysis of TM data using "standard" or classical multispectral analysis techniques, and 2) analysis of the same dataset using ENVI's hyperspectral tools.
A typical Landsat TM analysis scenario might consist of the following (though many other variations are available within ENVI. See Sabins, 1987 for other examples):
Before attempting to start the program, ensure that ENVI is properly installed as described in the installation guide.
The ENVI Main Menu appears when the program has successfully loaded and executed.
ENVI provides the tools to read standard Landsat Thematic Mapper data from both tape and CD/disk.
To open an image file:
Note that on some platforms you must hold the left mouse button down to display the submenus from the Main Menu.
An Enter Input Data File file selection dialog appears.
The Available Bands List dialog will appear on your screen. This list allows you to select spectral bands for display and processing.
Note that you have the choice of loading either a grayscale or an RGB color image.
Figure 1: A Color Composite Image showing Landsat TM Bands 4, 3, 2 (RGB)
Here you will create Color-Ratio-Composite (CRC) Images using standard TM band-ratio images. This method tries to get around the limitations of relatively broad spectral bands in Landsat TM data by using ratios of bands to determine relative spectral slope between bands and thus the approximate shape of the spectral signature for each pixel. Common band-ratios include: Band-Ratio 5/7 for Clays, Carbonates, Vegetation; Band-Ratio 3/1 for Iron Oxide; Band-Ratios 2/4 or 3/4 for Vegetation; and Band-Ratio 5/4 also for vegetation.
Figure 2: A Color-Ratio Composite Image of ratios 5/7, 3/1, and 2/4 (RGB)
To create a band-ratio image:
The combination of 5/7, 3/1, 2/4 (RGB) results in an image in which clays/carbonates are magenta, iron oxides are green, and vegetation is red. Other ratio combinations and color schemes can be designed to highlight specific materials.
Unsupervised classification provides a simple way to segment multispectral data using the data statistics. IsoData calculates class means evenly distributed in the data space and then iteratively clusters the remaining pixels using minimum distance techniques. Each iteration recalculates means and reclassifies pixels with respect to the new means. This process continues until the number of pixels in each class changes by less than the selected pixel change threshold or the maximum number of iterations is reached.
Figure 3: An IsoData Image.
Once the classification is complete, because classified images often suffer from a lack of spatial coherency (speckle or holes in classified areas), it is often desirable to generalize the classes to generalize the classification for operational use. Low pass filtering could be used to smooth these images, however, the class information would be contaminated by adjacent class codes. The Sieve Class and Clump Class operators have been designed to avoid this problem by removing isolated pixels and clumping together adjacent similar classified areas respectively using morphological operators.
Figure 4: Sieve and Clump Classification Generalization. Sieve is on the left, Clump of the Sieved image on the right.
The final output from any image processing within ENVI is usually a map-oriented, scaled image-map for presentation or visual analysis and interpretation. In this case, the TM data were already geographically referenced, however, ENVI includes full image-to-image and image-to-map registration capabilities. Please see the Registration Tutorial or the ENVI User's Guide and on-line tutorials.
ENVI also provides all of the tools to produce fully annotated publication-quality maps. This includes pixel, map, and geographic (latitude/longitude) grids; scale-bars; declination diagrams and north arrows; text and symbols; polygons, polylines, and geometric shapes (circles, rectangles); map keys and legends; and image insets. For additional information on map composition, please see the Map Composition Tutorial or the ENVI User's Guide and on-line tutorials.
Figure 5: Isodata Classification annotated image-map.
As described above, ENVI provides the tools to read standard Landsat Thematic Mapper data from both tape and CD/disk.
A reflectance calibration is required for Landsat TM data to compare image spectra to library reflectance spectra and to run some of ENVI's hyperspectral routines. ENVI provides TM calibration through the use of pre-launch gains and offsets calculated for the Landsat Sensors (Markham and Barker, 1986).
The resulting image approximates reflectance.
Extract Z-profiles (reflectance spectra) from the data by selecting Functions->Profiles->Z Profile in the Main Display window and browse around the image by clicking and dragging the red Zoom Window box using the left mouse button.
Figure 7: Landsat TM reflectance spectra.
MNF Transform is a method similar to Principal Components used to segregate noise in the data, determine inherent data dimensionality, and reduce computational requirements for subsequent processing (Green et al., 1988; Boardman and Kruse, 1994). For hyperspectral data (less-so for multispectral data), the MNF divides data space into two parts; one with large eigenvalues and coherent eigenimages and the second with near-unity eigenvalues and noise-dominated images. It is used as a preparatory transformation to put most of the interesting information into just a few spectral bands and to order those bands from most interesting to least interesting.
Figure 8: Landsat TM MNF Bands.
See the Hyperspectral Tutorials for additional background information and examples. To calculate the MNF transformation from the TM reflectance data:
When the MNF transform is completed, an eigenvalue plot will be shown and the MNF-transformed bands will be displayed in the Available Bands List.
Figure 9: MNF Eigenvalue Plot.
The decreasing eigenvalue with increasing MNF band shown in the eigenvalue plot above shows how noise is segregated in the higher number MNF bands.
The Pixel Purity Index (PPI) function finds the most spectrally pure or "extreme" pixels in multispectral and hyperspectral data (Boardman and Kruse, 1994). These correspond to the materials with spectra that combine linearly to produce all of the spectra in the image. The PPI is computed by using projections of n-dimensional scatterplots to 2-D space and marking the extreme pixels in each projection. The output is an image (the PPI Image) in which the digital number (DN) of each pixel in the image corresponds to the number of times that pixel was recorded as extreme. Thus bright pixels in the image show the spatial location of spectral endmembers. Image thresholding is used to select several thousand pixels for further analysis, thus significantly reducing the number of pixels to be examined. See the Hyperspectral Tutorials for additional PPI background information and examples.
To start the PPI analysis:
This calculates the PPI in memory.
The PPI image will appear in the Available Bands List when processing has completed.
Figure 10: Pixel Purity Index Image.
The selected pixels will be entered into ENVI's ROI Controls Dialog.
Though both the MNF and PPI operations above effectively reduce the data volume to be analyzed interactively, the high dimensionality of hyperspectral data requires advanced visualization techniques. ENVI's N-Dimensional Visualizer is an interactive n-dimensional scatterplotting paradigm that allows real-time rotation of scatterplots in n-dimensions (Boardman et al., 1995). This is accomplished by casting the scatterplots from n-d to 2-D to simplify analysis. Animation of the scatterplots then provides the capability to simultaneously use all bands for interactive analysis. The scientist's visual skills and scatterplot geometry are used to locate image spectral endmembers. See the Hyperspectral Tutorials and the ENVI User's Guide and on-line help for additional background information and examples.
Figure 11: The n-Dimensional Visualizer.
ENVI allows comparison of image spectra to spectra measured in the laboratory and saved in spectral libraries. Several relatively high spectral resolution spectral libraries are provided with ENVI.
The spectra will be plotted in an ENVI plot window.
Compare these spectra to the Landsat TM image spectra.
Figure 13: Library Spectra resampled to TM,
The Spectral Angle Mapper (SAM) measures the similarity of unknown and reference spectra in n-dimensions. The angle between the spectra treated as vectors in n-space is the "spectral angle", this is illustrated in 2 Dimensions in the figure below. This method assumes that the data have been reduced to apparent reflectance and uses only the "direction" of the spectra, and not their "length". Thus the SAM classification is insensitive to illumination effects. See the Hyperspectral Tutorials and the ENVI User's Guide and on-line help for additional background information and examples.
To start SAM:
Figure 14: The Spectral Angle Mapper (SAM) concept.
The results of the classification will be a set of rule images corresponding to the number of endmembers you selected and a SAM Classification Image. The rule images show the best matches in black when first displayed, however, these are typically inverted to better show the matches as bright pixels in the displayed rule images, select Functions->Color Mapping->ENVI Color Tables in the Main Display Window and reverse the Stretch Bottom and Stretch Top slider bars to invert the image.
The figure below shows the best match for each pixel (within the default threshold of 0.10 radians) color coded for each endmember.
Figure 15: SAM classification result.
Image pixels typically represent areas of from 1 to several square meters. Within these pixels, the Earth's surface is composed of mixtures of materials; pure pixels are extremely rare (Boardman). The mixed spectrum received by most imaging systems is a linear combination of the "pure" or "endmember" spectra, each weighted by their fractional abundance of area. Mixed pixels can be analyzed using a mathematical model where the observed spectrum is the result of multiplication of the mixing library of pure endmember spectra by the endmember abundances. Mixing can also be visualized, however, using a geometric model; this is the basis of ENVI's 2-D Projections of n-dimensional scatterplots. See ENVI's Hyperspectral Tutorials and the ENVI User's Guide and on-line tutorials for additional unmixing background information and examples.
Figure 16: Linear Spectral Mixiung
Figure 17: The linear spectral mixing concept.
To perform linear spectral unmixing using ENVI:
When complete, the Spectral Unmixing endmember image will appear in the Available Bands List.
Bright values in the abundance images represent high abundances; the Cursor Value/Location function can be used to examine the actual values.
When the RMS image doesn't have any more high errors, and all of the abundance images are non-negative and sum to less than one, then the unmixing is completed. This iterative method is much more accurate than trying to artificially constrain the mixing, and even after extensive iteration, also effectively reduces the compute time by several orders of magnitude compared to the constrained method.
Figure 19: Linear Spectral Unmixing Results.
As previously described, the final output from any image processing within ENVI is usually a map-oriented, scaled image-map for presentation or visual analysis and interpretation. In this case, the TM data were already geographically referenced, however, ENVI includes full image-to-image and image-to-map registration capabilities. Please see the Registration Tutorial in this volume or the ENVI User's Guide and on-line help. ENVI also provides all of the tools to produce fully annotated publication-quality maps. This includes pixel, map, and geographic (latitude/longitude) grids; scale-bars; declination diagrams and north arrows; text and symbols; polygons, polylines, and geometric shapes (circles, rectangles); map keys and legends; and image insets. For additional information on map composition, please see the Map Composition Tutorial or the ENVI User's Guide or on-line help.
A wide variety of advanced tools have been developed for analysis of imaging spectrometer (hyperspectral data). These tools are mature and are being used operationally for analysis of AVIRIS and other datasets. We don't have hyperspectral data for many of the areas we would like to investigate, however, widely available mulispectral data can be analyzed using some of the "hyperspectral" tools. ENVI allows users to use approaches developed for analysis of hyperspectral data to provide new insight to the use and analysis of multispectral datasets.
Boardman, J. W., Kruse, F. A., and Green, R. O., 1995, Mapping target signatures via partial unmixing of AVIRIS data: in Summaries, Fifth JPL Airborne Earth Science Workshop, JPL Publication 95-1, v. 1, p. 23-26.
Boardman, J. W., 1993, Automated spectral unmixing of AVIRIS data using convex geometry concepts: in Summaries, Fourth JPL Airborne Geoscience Workshop, JPL Publication 93-26, v. 1, p. 11 - 14.
Green, A. A., Berman, M., Switzer, P, and Craig, M. D., 1988, A transformation for ordering multispectral data in terms of image quality with implications for noise removal: IEEE Transactions on Geoscience and Remote Sensing, v. 26, no. 1, p. 65-74.
Markham, B. L., and Barker, J. L.,1986, Landsat MSS and TM post-calibration dynamic ranges, exoatmopspheric reflectances and at-satellite temperatures: EOSAT Landat Technical Notes, No. 1, August 1996.
Research Systems Inc, 1997, ENVI User's Guide, Chapter 10.
Sabins, F. F. Jr., 1986, Remote Sensing Principles and Interpretation: W. H. Freeman and Company, New York, 449 p.
Source: http://www.ltid.inpe.br