Hyperspectral Processing II Adapted from ENVI Tutorials #14 & 15

Transcription

1 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p. 1 Hyperspectral Processing II Adapted from ENVI Tutorials #14 & 15 In this lab we consider various types of spectral processing: Spectral Angle Mapper, Spectral Feature Finder, and Binary Encoding. We will again use the AVIRIS image of Cuprite, NV. Images needed for this exercise: File Description cup95eff_invmnf-15.img &.hdr Cuprite EFFORT-Corrected ATREM calibrated apparent reflectance data. cup95_em. &.asc Endmember collection in ASCII format. jpl1sli.dat &.hdr JPL Spectral Library in ENVI format. usgs_min.sli &.hdr USGS Spectral Library in ENVI format. Images to be created: cup95_sam.img & hdr cup95_sam-rule.img & hdr cup95_cr.dat & hdr cup95_sff.img Cuprite image classified with SAM SAM rule image Continuum-removed data (floating point) Spectral Feature Fitting Results For this lab we will use the cupu95eff image which has been processed by EFFORT (Empirical Flat Field Optimized Reflectance Transformation) to remove residual saw-tooth instrument (or calibrationintroduced) noise and atmospheric effects from ATREM-calibrated AVIRIS data. It is a custom correction for AVIRIS data designed to improve the overall quality of the spectra and provides the best reflectance spectra available from AVIRIS data. EFFORT is a relatively automated improvement on the Flat-Field Calibration method (Boardman and Huntington, 1996) and can be run from ENVI by selecting Spectral > Effort Polishing. We will not run this function during this exercise, but will simply use the corrected data. For this lab we will use the cupu95eff image which has been processed by EFFORT (Empirical Flat Field Optimized Reflectance Transformation) to remove residual saw-tooth instrument (or calibrationintroduced) noise and atmospheric effects from ATREM-calibrated AVIRIS data. It is a custom correction for AVIRIS data designed to improve the overall quality of the spectra and provides the best reflectance spectra available from AVIRIS data. EFFORT is a relatively automated improvement on the Flat-Field Calibration method (Boardman and Huntington, 1996) and can be run from ENVI by selecting Spectral > Effort Polishing. We will not run this function during this exercise, but will simply use the corrected data. Spectral Angle Mapper Classification The Spectral Angle Mapper (SAM) is an automated method for comparing image spectra to individual spectra or a spectral library (Kruse et al., 1993a). We will use both image and laboratory spectra to classify the AVIRIS data using the Spectral Angle Mapper (SAM). We will go through the endmember selection process for SAM, but will not actually run the algorithm. We will examine previously calculated classification results to answer specific questions about the strengths and weaknesses of the SAM classification.

2 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p. 2 SAM assumes that the data have been reduced to apparent reflectance (i.e., data that have been atmospherically corrected, but without correcting for the effects of topography and shadows). The algorithm determines the similarity between two spectra by calculating the spectral angle between them, treating them as vectors in a space with dimensionality equal to the number of bands (nb). Because it uses only the direction of the spectra, and not their length, the method is insensitive to the unknown gain factor, and all possible illuminations are treated equally. Poorly illuminated pixels will fall closer to the origin. The color of a material is defined by the direction of its unit vector. The angle between the vectors is the same regardless of the length. The length of the vector relates only to how fully the pixel is illuminated. Two-dimensional example of the Spectral Angle Mapper The SAM algorithm generalizes this geometric interpretation Figure 1 to nb-dimensional space. SAM determines the similarity of an unknown spectrum t to a reference spectrum r, by applying the following equation (CSES, 1992): For each reference spectrum chosen in the analysis of a hyperspectral image, the spectral angle is determined for every image spectrum (pixel). This value, in radians, is assigned to the corresponding pixel in the output SAM image one output image for each reference spectrum. The derived spectral angle maps form a new data cube with the number of "bands" corresponding to the number of reference spectra used in the mapping. Gray-level thresholding is typically used to empirically determine those areas that most closely match the reference spectrum while retaining spatial coherence. The SAM algorithm implemented in ENVI takes as input a number of training classes or reference spectra from ASCII files, ROIs, or spectral libraries. It calculates the angular distance between each spectrum in the image and the reference spectra or endmembers in n-dimensions. The result is a classification image showing the best SAM match at each pixel and a rule image for each endmember showing the actual angular distance in radians between each spectrum in the image and the reference spectrum. Darker pixels in the rule images represent smaller spectral angles, and thus spectra that are more similar to the reference spectrum. The rule images can be used for subsequent classifications using different thresholds to decide which pixels are included in the SAM classification image. Select Image Endmembers and Display SAM results (or Execute SAM) 1. Load the cup95eff_invmnf-15.img image. It should automatically display as a color image using bands 183, 193, and 207 as RGB, respectively. This is a 1995 AVIRIS ATREM-calibrated apparent reflectance data with the EFFORT correction. Display band 193 (2.20 ) as a grayscale image. 2. From the ENVI main menu, select Classification > Supervised > Spectral Angle Mapper to start the SAM endmember selection process. 3. In the Classification Input File dialog, select the file cup95eff_invmnf-15.img as the input file and click OK. 4. In the Endmember Collection: SAM dialog, select Import > from ASCII file. 5. Select the file cup95_em.asc and click Open.

3 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p. 3 This is a set of laboratory reflectance (endmember) spectra that have been identified as probable mineral types in this scen 6. In the Input ASCII File dialog, hold down the Ctrl key and, in the Select Y Axis Columns list, deselect the spectra Dark/black, Bright/playa, Silica? (Dark), and Alunite (2.18). This will leave you with the mean spectra for Zeolite, Calcite, Alunite (2.16), Kaolinite, Illite/Muscovite, Silica (Bright), and Buddingtonite. 7. Click OK to load all of the selected endmember spectra into the Endmember Collection:SAM dialog. 8. Reset the colors for the spectra. (Some of the default colors are startling and others are hard to see making viewing the spectra and classification unnecessarily difficult.) Adjust the colors according as in (below by right-clicking on the cell in the color column and selecting the appropriate color. 9. From the Endmember Collection:SAM dialog menu, Select All and Plot to plot the endmember spectra, then right click in the plot window and select Plot Key from the shortcut menu to display the legend. The result is shown in Error! Reference source not found.. Note: You may increase the area available for the legend by selecting Edit > Plot parameters window and increasing the right margin size. Code Mineral Color 1 Zeolites White 2 Calcite Green 3 Alunite Purple 4 Kaolinite Red 5 Illite/Muscovite Coral 6 Silica Cyan3 7 Buddingtonite Cyan Figure Select Options > Stack Plots. This will spread the spectra out, making them easier to view. 11. Run the spectral angle mapper by clicking Apply in the Endmember Collection:SAM dialog, entering output file names in the Spectral Angle Mapper Parameters dialog. Leave the default maximum angle setting (0.1 radians) select memory for the output file and turn off the output rule image. 12. Display the SAM classification image. With the 0.1 radian (~ 5.7 ) setting there are no unclassified pixels. This is a rather large acceptance angle. Repeat steps 11 & 12 using a series of lower threshold angles (e.g., 0.07, 0.05) and compare the classifications. As you restrict the acceptance angle, fewer pixels will be classified.

4 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p Select the image with a classification that appears capture the most features without wholesale classification of large areas and run the SAM classifier one last time using that angle setting. Name the output file cup95_sam.img, and create rule images using the name cup95_sam-rule.img. 14. Load and display the SAM classification cup95_sam.img. Be sure that the Gray Scale radio button is selected and load the image into a new display window. Each class is color coded and numbered as in the table above. Verify this by selecting Tools > Color Mapping > Class Color Mapping in the SAM classification image window. Note: The number of pixels displayed as a specific class is a function of the threshold used to generate the classification. The classification is the best guess given the endmember spectra that were available for the classification. SAM is a similarity measure, not an identifier.. Examine the Classification 1. Start a spectral profile by selecting Tools > Profiles > Z-Profile in the original scene (the color image window). 2. Link the original image and the classified images, and use the classified image to guide the examination of the spectra. a. Move the zoom box to the area of interest (i.e., a region of uniform color in the classified image). You may make adjustments in the position using the arrow keys. b. Compare the image spectrum to the corresponding endmember spectrum. Do they match? c. Compare the SAM classification results with the distributions shown by the color composite image. Keep in mind that all pixels are classified to one of the 7 classes represented by the endmembers, and that there are more than 7 types of material in this image. Can you spot any obvious mineral types (distinctive colors in the original color image) that are not represented? If so, does the spectrum of this material match any of the expected Most of the unclassified areas will be in outwash planes and may be mixtures of nearby materials in the uplands. Is there any calcite in your classified image? 3. Compare the Rule images for each mineral to the SAM classification. a. Load cup95_sam-rule.img. The rule image has one band for each endmember classified, with the pixel values representing the spectral angle in radians. Since the smaller spectral angles (darker pixels) represent better spectral matches to the endmember spectrum. b. In the Available Bands List dialog, ensure that the Gray Scale radio button is selected. Select Display>New Display, Select a Rule image (e.g., Zeolite), and then click Load Band. 4. Evaluate each rule image with respect to the color composite and the SAM classification image as well as the individual spectra extracted using the Z Profiler. a. Verify that the darkest areas in each of the rule image correspond to the classified areas in your SAM classification. b. Try a color composite of any three categories. ( I like calcite (R), Sicica (G), and Illite (B) )

5 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p. 5 Modify the classification using Rule Classifier Generate a new classified images based on different thresholds in the rule images. 1. Select Classification > Post Classification > Rule Classifier and choose cup95_sam-rule.img. OK. 2. In the Rule Image Classifier Tool dialog, select Minimum Value in the Classify by box. This is necessary because the smaller angles indicate that a pixel is closer to a particular class. 3. Next select Options > Change Class Colors/Names. Change the color map to match the classification colors. 4. Enter the threshold that you used to create the cup95_sam.img into the box at the top and select Set All Thresholds, then Quick Apply. The rule image classification should be identical to the SAM classification. 5. Adjust the threshold for each class. One convenient way to do this is to use ENVI s interactive density slice tool. Select Tools > Color Mapping > Density Slice in the main image window. Click Apply. The tool automatically splits the gray range into eight uniform ranges and assigns colors to each range. You can edit the ranges, but it's often possible to make an estimate of a reasonable threshold using the default values. In the example below, the lowest range (red; rad, ~2.2 ) indicates the best match and corresponds to the most localized regions. The next range (green; rad ~ 3.6 ) seems to capture a general distribution that is reasonable when comparing to the original color image. A reasonable starting value for a threshold for zeolite might be about Repeat the procedure for each of the 7 classes, noting your threshold selection in the threshold box in the Rule Image Classifier Tool for each class. You can use Quick Apply to update the classification at any time. You may also turn classes on and off using the check boxes on the left. Figure 3: Zeolite rule image (left) density sliced (center) using the default settings (right). An example of a rule classification is shown in Figure 4

6 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p. 6 Figure 4: Colors and thresholds used for the rule classification (left) and the classification result (right).

7 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p. 7 Spectral Feature Fitting and Analysis Spectral Feature Fitting (SFF ) is an absorption-feature-based method for matching image spectra to reference endmembers, similar to methods developed at the U. S. Geological Survey, Denver (Clark et al., 1990, 1991, 1992; Clark and Swayze, 1995). The method requires that data be reduced to reflectance and that a continuum be removed from the reflectance data prior to analysis. A continuum is a mathematical function used to isolate a particular absorption feature for analysis (Clark and Roush, 1984; Kruse et al, 1985; Green and Craig, 1985). It corresponds to a background signal unrelated to specific absorption features of interest. Spectra are normalized to a common reference using a continuum formed by defining high points of the spectrum (local maxima) and fitting straight line segments between these points. The continuum is removed by dividing it into the original spectrum. Figure 5: Example of a fitted continuum and a continuum-removed spectrum for the mineral kaolinite Spectral feature fitting requires that reference endmembers be selected from either the image or a spectral library, that both the reference and unknown spectra have the continuum removed, and that each reference endmember spectrum be scaled to match the unknown spectrum. A Scale image is produced for each endmember selected for analysis by first subtracting the continuum-removed spectra from one, thus inverting them and making the continuum zero. A single multiplicative scaling factor is then determined that makes the reference spectrum match the unknown spectrum. Assuming that a reasonable spectral range has been selected, a large scaling factor is equivalent to a deep spectral feature, while a small scaling factor indicates a weak spectral feature. A least-squares-fit is then calculated band-by-band between each reference endmember and the unknown spectrum. The total root-mean-square (RMS) error is used to form an RMS error image for each endmember. An optional ratio image of Scale/RMS provides a Fit image that is a measure of how well the unknown spectrum matches the reference spectrum on a pixel-by-pixel basis.

8 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p. 8 Open and Load the Continuum-Removed Data 1. Open the file cup95eff_invmnf-15.img and select Spectral > Mapping Methods >Continuum Removal. 2. In the Continuum Removal Input File dialog, select the file cup95eff_invmnf-15.img, perform spectral subsetting if desired to limit the spectral range for continuum removal, and click OK. 3. Enter the continuum-removed output file name cup95_cr.dat in the Continuum Removal Parameters dialog and click OK to create the continuum-removed image. This image will have the same number of spectral bands as the input image if no spectral subsetting was done. 4. Start a spectral profile (z-profile) for both the original image and the continuum-removed (CR) image, and link the images. Stretch the vertical scale of the continuum-removed image. Choose Edit > Plot Parameters in the Spectral Profile window, select the Y-Axis radio button, and enter 0.6 in the Range text box and 1.0 in the To text box and click Apply to apply the new Y- axis range. Close the Plot Parameters dialog. 5. Click the left mouse button in the cup95eff_invmnf-15.img image window to activate the dynamic overlay. [Use the middle button to adjust the size of the overlay window.] 6. Compare the CR image to the color composite image. Note the correspondence between deep red areas on the CR image and red areas on the color composite. These are areas with absorption bands near 2.2 microns. Figure 6: Original spectrum vs. CR spectrum for a feature with strong absorption near 2.2 microns. 7. Similarly, the yellow areas in the CR image correspond to green areas on the original image and mark the location of a strong absorption at 2.34 microns. Figure 7: Original spectrum vs. CR spectrum for a feature with strong absorption near 2.34 microns.

9 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p Load continuum-removed band 207 (2.34 m) in the appropriate display. Note by moving the Zoom window to the darkest areas and examining the spectra that these correspond to absorption features near 2.34 m in both the continuum-removed and Effort spectra. Create the SFF Scale and RMS Images 1. Open the file cup95_cr.dat and select Spectral > Mapping Methods >Spectral Feature Fitting. 2. Select Import > from ASCII file, select the cup95_em.asc spectral library and click APPLY. (This is where you might perform spectral subsetting, if desired, to limit the spectral range for fitting.) 3. Choose "Output separate Scale and RMS Images in the Spectral Feature Fitting Parameters dialog, enter an output file name, cup95_sff.img, and click OK to create the Scale and RMS error images. The output image will have two images for each endmember, a Scale image and an RMS error image. 4. Load the RMS (Mean: Kaolinite...) image as a gray scale image into the display that contains the Kaolinite Scale Image. 5. Load the RMS (Mean: Alunite ) image as a gray scale image into the display that contains the Alunite Scale Image. 6. The RMS images represent a "goodness of fit" distribution and the distributions appear to be nearly identical. To examine the subtle differences, display a 2-D Scatter plot from the RMS (Mean: Kaolinite...) image window and plot the RMS Kaolinite vs. RMS Allunite. Note that the off-axis areas represent a better fit to one mineral than the other. In Figure 8, the red area represents a good fit (low RMS error for Kaolinite while there is a consistently higher range of fitting errors for Allunite. 7. Create a second 2-D Scatter Plot contrasting Scale and RMS.images for Kaolinite. Draw a Region of Interest (ROI) on the scatter plot at low RMS values for all ranges of Scale. Note the highlighted pixels in the RMS Kaolinite. 8. If time permits, examine other mineral distributions. 9.. Select Window > Close All Plot Windows. Figure 8: Kaolinite vs. Allunite (RMS)

10 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p. 10 References 1. Boardman, J. W., and Huntington, J. F., 1996, Mineral Mapping with 1995 AVIRIS data: in Summaries of the Sixth Annual JPL Airborne Research Science Workshop, JPL Publication 96-4, Jet Propulsion Laboratory, v. 1, p Kruse, F. A., Lefkoff, A. B., Boardman, J. W., Heidebrecht, K. B., Shapiro, A. T., Barloon, J. P., and Goetz, A. F. H., 1993a, The spectral image processing system (SIPS) - Interactive visualization and analysis of imaging spectrometer data: Remote Sensing of Environment, v. 44, p Center for the Study of Earth from Space (CSES), 1992, SIPS User s Guide, The Spectral Image Processing System, v. 1.1, University of Colorado, Boulder, 74 p. 4. Clark, R. N., and Roush, T. L., 1984, Reflectance spectroscopy: Quantitative analysis techniques for remote sensing applications: Journal of Geophysical Research, v. 89, no. B7, pp Clark, R. N., Gallagher, A. J., and Swayze, G. A., 1990, Material absorption band depth mapping of imaging spectrometer data using the complete band shape leastsquares algorithm simultaneously fit to multiple spectral features from multiple materials: in Proceedings of the Third Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Workshop, JPL Publication 90-54, p Clark, R. N., Swayze, G. A., Gallagher, A., Gorelick, N., and Kruse, F. A., 1991, Mapping with imaging spectrometer data using the complete band shape leastsquares algorithm simultaneously fit to multiple spectral features from multiple materials: in Proceedings, 3rd Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) workshop, JPL Publication 91-28, p Clark, R. N., Swayze, G. A., and Gallagher, A., 1992, Mapping the mineralogy and lithology of Canyonlands, Utah with imaging spectrometer data and the multiple spectral feature mapping algorithm: in Summaries of the Third Annual JPL Airborne Geoscience Workshop, JPL Publication 92-14, v. 1, p Clark, R. N., and Swayze, G. A., 1995, Mapping minerals, amorphous materials, environmental materials, vegetation, water, ice, and snow, and other materials: The USGS Tricorder Algorithm: in Summaries of the Fifth Annual JPL Airborne Earth Science Workshop, JPL Publication 95-1, p Green, A. A., and Craig, M. D., 1985, Analysis of aircraft spectrometer data with logarithmic residuals: in Proceedings, AIS workshop, 8-10 April, 1985, JPL Publication 85-41, Jet Propulsion Laboratory, Pasadena, California, p Kruse, F. A., Raines, G. L., and Watson, K., 1985, Analytical techniques for extracting geologic information from multichannel airborne spectroradiometer and airborne imaging spectrometer data: in Proceedings, International Symposium on Remote Sensing of Environment, Thematic Conference on Remote Sensing for Exploration Geology, 4th Thematic Conference, Environmental Research Institute of Michigan, Ann Arbor, p