ENVI Tutorial
Description
ENVI Tutorial: AdvancedHyperspectral AnalysisAdvanced Hyperspectral Analysis 2Files Used in this Tutorial 2Background: MNF Transforms 3Open EFFORT-Corrected Data 3Open and Load MNF Image 4Compare MNF Images 4Examine MNF Scatter Plots 4Use Scatterplots to Select Endmembers 5Pixel Purity Index 8Display and Analyze the Pixel Purity Index 8Threshold PPI to Regions of Interest 9The n-D Visualizer 11Compare n-D Data Visualization with a 2D Scatter Plot 11Use the n-D Visualizer 13Select Endmembers 14Use the n-D Class Controls 14Link the n-D Visualizer to Spectral Profiles 15Link the Spectral Analyst to the n-D Visualizer Spectra 15Load Individual Spectra into the n-D Visualizer 16Collapse Classes in the n-D Visualizer 16Export Your Own ROIs 17Save Your n-D Visualizer State 17Restore n-D Visualizer Saved State 17Spectral Mapping 19What Causes Spectral Mixing 20Modeling Mixed Spectra 20Practical Unmixing Methods 21Linear Spectral Unmixing Results 23Open and Display Linear Spectral Unmixing Results 23Determine Abundance 23Display a Color Composite 23Mixture Tuned Matched Filtering 24Display and Compare EFFORT-Corrected and MNF Data 24Collect EFFORT and MNF Endmember Spectra 24Calculate MTMF Images 25Dispay MTMF Results 25Display Scatter Plots of MF Score versus Infeasibility 26References 281ENVI Tutorial: Advanced Hyperspectral AnalysisAdvanced Hyperspectral AnalysisThis tutorial introduces you to advanced concepts and procedures for ...
Informations
| Publié par | Phion |
| Nombre de lectures | 171 |
| Langue | English |
Extrait
ENVI Tutorial: Advanced Hyperspectral Analysis
Advanced Hyperspectral Analysis Files Used in this Tutorial Background: MNF Transforms Open EFFORT-Corrected Data Open and Load MNF Image Compare MNF Images Examine MNF Scatter Plots Use Scatterplots to Select Endmembers Pixel Purity Index Display and Analyze the Pixel Purity Index Threshold PPI to Regions of Interest The n-D Visualizer Compare n-D Data Visualization with a 2D Scatter Plot Use the n-D Visualizer Select Endmembers Use the n-D Class Controls Link the n-D Visualizer to Spectral Profiles Link the Spectral Analyst to the n-D Visualizer Spectra Load Individual Spectra into the n-D Visualizer Collapse Classes in the n-D Visualizer Export Your Own ROIs Save Your n-D Visualizer State Restore n-D Visualizer Saved State Spectral Mapping What Causes Spectral Mixing Modeling Mixed Spectra Practical Unmixing Methods Linear Spectral Unmixing Results Open and Display Linear Spectral Unmixing Results Determine Abundance Display a Color Composite Mixture Tuned Matched Filtering Display and Compare EFFORT-Corrected and MNF Data Collect EFFORT and MNF Endmember Spectra Calculate MTMF Images Dispay MTMF Results Display Scatter Plots of MF Score versus Infeasibility References
1
2 2 3 3 4 4 4 5 8 8 9 11 11 13 14 14 15 15 16 16 17 17 17 19 20 20 21 23 23 23 23 24 24 24 25 25 26 28
ENVI Tutorial: Advanced Hyperspectral Analysis
Advanced Hyperspectral Analysis This tutorial introduces you to advanced concepts and procedures for analyzing imaging spectrometer data or hyperspectral images. You will use Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) EFFORT-corrected, atmospherically corrected apparent reflectance data from Cuprite, Nevada, USA, to investigate sub-pixel properties of hyperspectral data and advanced techniques for identifying and quantifying mineralogy. You will also review Matched Filtering and Linear Spectral Unmixing results. This tutorial is designed to be completed in two to four hours. Files Used in this Tutorial ENVI Resource DVD:atDc9a\a5svbu File Description cup95eff.int (.hdr)AVIRIS EFFORT-polished, atmospherically corrected apparent reflectance data cup95mnf.dat (.hdr)First 25 Minimum Noise Fraction (MNF) bands cup95mnf.ascMNF eigenvalue spectrum cup95mnf.staMNF statistics cup95ppi.dat (.hdr)Pixel Purity Index (PPI) image cup95ppi.roiRegion of interest (ROI) for PPI values greater than 1750 cup95ppi.ndvn-D Visualizer saved state file cup95ndv.roiROI endmembers corresponding to the n-D Visualizer saved state file _ selected using the PPIpectr bers cup95 em.asc endmem alEFFORT ASCII file of 11 s threshold, MNF images, and n-D Visualization cup95_mnfem.ascMNF ASCII file of 11 spectral endmembers selected using the PPI threshold, MNF images, and n-D Visualization cup95unm.datUnmixing results—fractional abundance images usgs_min.sli (.hdr)USGS spectral library in ENVI format
2
ENVI Tutorial: Advanced Hyperspectral Analysis
Background: MNF Transforms The Minimum Noise Fraction (MNF) transform is used to determine the inherent dimensionality of image data, to segregate noise in the data, and to reduce the computational requirements for subsequent processing (Boardman and Kruse, 1994). The MNF transform as modified from Green et al. (1988) and implemented in ENVI, is essentially two cascaded Principal Components transformations. The first transformation, based on an estimated noise covariance matrix, decorrelates and rescales the noise in the data. This first step results in transformed data in which the noise has unit variance and no band-to-band correlations. The second step is a standard Principal Components transformation of the noise-whitened data. For the purposes of further spectral processing, the inherent dimensionality of the data is determined by examination of the final eigenvalues and the associated images. The data space can be divided into two parts: one part associated with large eigenvalues and coherent eigenimages, and a complementary part with near-unity eigenvalues and noise-dominated images. By using only the coherent portions, the noise is separated from the data, thus improving spectral processing results. The figure below summarizes the MNF procedure in ENVI. The noise estimate can come from one of three sources; from the dark current image acquired with the data (for example, AVIRIS), from noise statistics calculated from the data, or from statistics saved from a previous transform. Both the eigenvalues and the MNF images (eigenimages) are used to evaluate the dimensionality of the data. Eigenvalues for bands that contain information will be an order of magnitude larger than those that contain only noise. The corresponding images will be spatially coherent, while the noise images will not contain any spatial information.
Open EFFORT-Corrected Data Empirical Flat Field Optimized Reflectance Transformation (EFFORT) is a correction method used to remove residual “saw-tooth” instrument (or calibration-introduced) noise and atmospheric effects from AVIRIS data. It is a custom correction designed to improve the overall quality of spectra, and it provides the best reflectance spectra available from AVIRIS data.
3
ENVI Tutorial: Advanced Hyperspectral Analysis
1. From the ENVI main menu bar, selectFile > Open Image File. A file selection dialog appears. 2. Navigate tova59busaDc\atand selectp9cuff5ei.tn. ClickOpen. 3. In the Available Bands List, selectBand 193rdeune9f5fu.pinct. Select theGray Scale radio button, and clickLoad Band. Open and Load MNF Image 1. From the ENVI main menu bar, selectFile > Open Image File. A file selection dialog appears. 2. Navigate tobc\59vausaDatand selectcup95mtad.fn. ClickOpen. This dataset contains the first 25 MNF bands (floating-point) from the Cuprite EFFORT-corrected data. 3. In the Available Bands List, selectMNF Band 1rednuucatd.fnm59p. Select theGray Scale radio button. 4. In the Available Bands List, clickDisplay #1and selectNew Display. ClickLoad Band. Compare MNF Images 1. From a Display group menu bar, selectTools > Link > Link Displays. ClickOKto link the two display groups. 2. Click in an Image window to use dynamic overlay to compare the two images. 3. From a Display group menu bar, selectTools > Link > Dynamic Overlay Off. 4. From both Display group menu bars, selectTools > Profiles > Z Profile (Spectrum). Compare the MNF spectra with the apparent reflectance spectra from the EFFORT-corrected data. 5. Do you see a pattern or relationship between the MNF image and the apparent reflectance image? Relate the MNF band number to MNF image quality. Examine MNF Scatter Plots 1. From the Display #2 menu bar, selectTools > 2D Scatter Plots. A Scatter Plot Band Choice dialog appears. 2. Choose two bands to scatter plot and clickOK. Try different band combinations. Once you plot the data, you can change the bands to plot by selectingOptions > Change Bandsfrom the Scatter Plot window menu bar. Be sure to choose a high-variance (low band number) MNF band. Also, examine at least one scatter plot of a low-variance (high band number) MNF band. Notice the corners (pointed edges) on some MNF scatter plots, as the following figure shows.
4
ENVI Tutorial: Advanced Hyperspectral Analysis
3. Use linked display groups, dynamic overlays, and Z Profiles to understand the reflectance spectra of the MNF corner pixels. Look for areas where the MNF data transition from “pointy” to “fuzzy.” Also notice the relationship between scatter plot pixel location and spectral mixing as determined from image color and individual reflectance spectra. How do you explain these patterns? How can you exploit them? Use Scatterplots to Select Endmembers You will now investigate the possibilities of deriving unmixing endmembers from the data using MNF images and 2D scatter plots. 1. From the Scatter Plot menu bar, selectOptions > Change Bands. A Scatter Plot Band Choice dialog appears. 2. UnderChoose Band X, selectMNF Band 1. UnderChoose Band Y, selectMNF Band 2. Click OK. 3. From the Scatter Plot menu bar, selectOptions > Image: ROI. Draw a polygon ROI around a few extreme data points in a corner or arm of the data cloud. The following figure shows an example:
5
ENVI Tutorial: Advanced Hyperspectral Analysis
Right-click to close the polygon. These data points are mapped in the corresponding image as colored pixels. 4. From the Scatter Plot menu bar, selectClass > New. Draw another polygon ROI around a few extreme data points in a different corner or arm of the data cloud. 5. From the Scatter Plot menu bar, selectOptions > Image: Dance. Click-and-drag inside the MNF Image window to view the corresponding pixels in the Scatter Plot, shown as "dancing pixels." Or, click-and-drag the middle mouse button inside the Scatter Plot to highlight the corresponding pixels in the MNF Image window. 6. From the Scatter Plot menu bar, selectOptions > Export All. An ROI Tool dialog appears with a list of the ROIs you defined. 7. Repeat Steps 1-6, using different combinations of the first several MNF bands. It is important to use different band combinations to identify the most spectrally unique materials. Corner pixels generally make good endmember estimates, however you will see several overlapping or repeating ROIs. This is a limitation of examining the data in a pairwise (2D) fashion. 8. Load your ROIs into the apparent reflectance image by selectingOverlay > Region of Interest from the Display #1 menu bar. 9. In the ROI Tool dialog, clickSelect All, followed byStats, to extract the mean apparent reflectance spectra of the ROIs. An ROI Statistics Results dialog appears. 10. In the ROI Statistics Results dialog, clickPlotand selectMeanfor all ROIs to extract the mean apparent reflectance spectra of the ROIs. 11. Use the linked display groups and Z Profiles to examine the relationship between the MNF and reflectance spectra.
6
12.
From Keep
the the
Scatter display
Plot menu bar, selectFile > Cancel. groups open for the next exercise.
7
ENVI Tutorial: Advanced Hyperspectral Analysis
Close
the
ROI
Statistics
Results
dialog.
ENVI Tutorial: Advanced Hyperspectral Analysis
Pixel Purity Index Separating purer pixels from more mixed pixels reduces the number of pixels to analyze for determining endmembers, and it makes separation and identification of endmembers easier. The Pixel Purity Index (PPI) is a means of finding the most “spectrally pure,” or extreme, pixels in multispectral and hyperspectral images (Boardman et al., 1995). The most spectrally pure pixels typically correspond to mixing endmembers. You compute the PPI by repeatedly projecting n-dimensional (n-D) scatter plots onto a random unit vector. ENVI records the extreme pixels in each projection—those pixels that fall onto the ends of the unit vector—and it notes the total number of times each pixel is marked as extreme. A PPI image is created where each pixel value corresponds to the number of times that pixel was recorded as extreme. The following diagram summarizes the use of PPI in ENVI:
Display and Analyze the Pixel Purity Index In this exercise, you will examine the role of convex geometry in determining the relative purity of pixels. 1. From the ENVI main menu bar, selectFile > Open Image File. A file selection dialog appears. 2. Navigate toaD\ctaav95bsuand select9pucipp5tad.. ClickOpen. 3. In the Available Bands List, clickDisplay #2and selectNew Display. 4. Select theGray Scaleradio button. SelectPPI Resultand clickLoad Band. Following is a summary of what each display group should contain at this point. Displays #1 and #2 should still be open from the previous exercise. Display #1: EFFORT-corrected apparent reflectance data (.int5effcup9) Display #2: MNF data (tdaf.mn95upc)
8
ENVI Tutorial: Advanced Hyperspectral Analysis
Display #3: PPI results (9pp5uctaipd.) Brighter pixels in the PPI image represent more spectrally extreme finds (hits) and indicate pixels that are more spectrally pure. Darker pixels are less spectrally pure. 5. From the Display #3 menu bar, selectahnEecnand try various interactive stretches to understand the PPI image’s histogram and data distribution. Why is the histogram skewed to the low values? What does this mean from a mixing point of view? The PPI image is the result of several thousand iterations of the PPI algorithm on the MNF data. The values in the PPI image indicate the number of times each pixel was discovered as extreme in some projection. These numbers then indicate the degree of local convexity of the data cloud near each pixel and the proximity of each pixel to the convex hull of the data. In short, the higher values indicate pixels that are nearer to corners of the n-D data cloud, and are thus relatively purer than pixels with lower values. Pixels with values of 0 were never found to be extreme. 6. From a Display group menu bar, selectTools > Link > Link Displaysand clickOKto link all three display groups. 7. From each Display group menu bar, selectTools > Profiles > Z Profile (Spectrum). Now you can examine the spectral profiles of selected pixels in the PPI display group. 8. From the Display #3 menu bar, selectTools > Cursor Location/Valueand examine the range of data values in the PPI image. 9. Move around the PPI image, and use the Spectral Profile window and dynamic overlay to examine the purest pixels, both spatially and spectrally. Do any of the high PPI values fall in the regions of the image corresponding to the 2D plot corners you selected in the previous exercise? Why? Threshold PPI to Regions of Interest 1. From the Display #3 menu bar, selectTools > Region of Interest > ROI Tool. The ROI Tool dialog appears. 2. From the ROI Tool menu bar, selectFile > Restore ROIs. A file selection dialog appears. 3. Select5pp9cuoi.rpiand clickOpen. An ENVI Message dialog appears with information about the ROI. ClickOK. This ROI represents a collection of pixels where the PPI value is over 1750. How many high PPI pixels are there? Next, you will create your own thresholded PPI ROIs. 4. From the Display #3 menu bar, selectEnhance > Interactive Stretching. 5. To determine a threshold to use for choosing only the purest pixels, read and understand the data values from the histogram. Click the middle mouse button in the histogram to zoom to the lower end of the distribution. Click-and-hold the left mouse button as you browse the histogram. 6. Select a value on the high tail of the histogram as the minimum threshold (if this seems too difficult, try a value of 2000 as a starting point). 7. From the ROI Tool menu bar, selectOptions > Band Threshold to ROIto create an ROI containing only the pixels with high PPI values. A file selection dialog appears. 8. SelectPPI Resultuerndp9cupi5pd.taand clickOpen. A Band Threshold to ROI Parameters dialog appears.
9
ENVI Tutorial: Advanced Hyperspectral Analysis
9. In theMin Thresh Valuefield, enter the value you determined in Step 6. ClickOK. ENVI determines the number of pixels that meet the selected criteria and issues an ENVI Question dialog. For this exercise, if your threshold results in more than 2000 pixels being selected, you should select a higher minimum threshold.
10. ClickYesin the ENVI Question dialog. A new ROI called "Thresh" appears near the bottom of the table in the ROI Tool dialog. This ROI contains the pixel locations of the purest pixels in the image, regardless of the endmember to which they correspond.
In the next exercise, you will use the n-D Visualizer to isolate the specific pure endmembers.
10
ENVI Tutorial: Advanced Hyperspectral Analysis
The n-D Visualizer The n D Visualizer is an interactive tool to use for selecting the endmembers in n-D space. You can think of spectra as points in an n-D scatter plot, where n is the number of bands. The coordinates of the points in n-D space consist of n spectral radiance or reflectance values in each band for a given pixel. You can use the distribution of these points in n-D space to estimate the number of spectral endmembers and their pure spectral signatures. When using the n-D Visualizer, you can interactively rotate data in n-D space, select groups of pixels into classes, and collapse classes to make additional class selections easier. You can export the selected classes to ROIs and use them as input into classification, Linear Spectral Unmixing, or Matched Filtering techniques. The following figure summarizes the steps involved in using the n-D Visualizer to select endmember spectra.
Compare n-D Data Visualization with a 2D Scatter Plot 1. From the ENVI main menu bar, selectSpectral > n-Dimensional Visualizer > Visualize with New Data. A file selection dialog appears. To visualize pixels in the n-D Visualizer scatter plot, you must define an ROI from a PPI image. You performed this step in the previous exercise. 2. Selectcuptf.da95mn. ClickSpectral SubsetFile Spectral Subset dialog appears with all. A bands selected. 3. SelectMNF Band 1, hold down theShiftkey, and selectMNF Band 10. ClickOK. The first several bands of the MNF file encompass most of the variance in the original dataset. Limiting the number of bands improves the performance of the n-D Visualizer. 4. ClickOKn-D Visualizer Input File dialog. An n-D Visualizer Input ROI dialog appears. Ifin the only one valid ROI was listed in the ROI Tools dialog, those ROI data would be automatically loaded into the n-D Visualizer. If more than one ROI is listed, choose the ROI derived using the PPI threshold when queried. 5. In the n-D Visualizer Input ROI dialog, selectsThhreand clickOK. An n-D Visualizer plot window and n-D Controls dialog appear. Each number in the n-D Controls dialog represents a spectral band.
11
© 2010-2024 YouScribe