Lab 8

Multi-band (Spectral) Enhancements 

Geog 418/518, Fundamentals of Remote Sensing

 

Name 1                                                       Name 2                                            

 

Lab Goals

 

This lab continues your introduction to a suite of techniques known as image enhancements that are used to highlight particular features or processes.  You have already explored: 1) point operator or radiometric enhancements through your work with highlighting areas-of-interest, color changes, and contrast stretches; 2) multi-image enhancements when you created a 3-D drape; and 3) local operators or spatial filter enhancements when you used edge detection algorithms to try to highlight trails on a black and white air photo.  Today you will focus on a suite of multi-band or spectral enhancements that allow you to highlight features by looking at the spectral reflectance values across a range of bands for a given pixel.  Spectral enhancements you will learn about today are:

 

·        Intensity/Hue/Saturation (HIS) transformations, which alter the image colors based on brightness, dominant wavelength, and spectral purity of a pixel across different bands.

 

·        Band ratios, which highlight specific features that have high reflectance in one band and relatively low reflectance in another.

 

·        The tasseled cap transformation, which is used  to separate brightness, greenery and water.

 

·        Principal component transformations, which change the values of pixels based on a multivariate statistical relation derived from the spectral reflectance values for all pixels in all bands.

 

In addition, you will be introduced to some hyperspectral imagery, which will provide you an opportunity to compare visualization results from multi-band and hyper-band imagery.

 

 

Background:

 

The enhancement techniques you will explore today are techniques commonly used by remote sensing professionals. All the techniques you will use today are described in your textbook and in the online ERDAS Field Manual, 1999, as well as being covered in class.  The first part of this lab on mineral indices and IHS transformations is loosely adapted from the online ERDAS tutorial on the topic.

 

 

Lab Procedures:

 

As with all labs, work in pairs on this lab and turn in only one lab per team. Please provide your answers in complete sentences, except where the question specifies otherwise. 


Lab Activities and Questions:

 

Part 1, Band Ratioing and IHS Transformations

 

Sometimes it can be useful to use more than just one enhancement to highlight a specific feature.  The following steps lead you through the process of: a) highlighting mineral types on a TM image using a band ratio procedure; and b) using an IHS transformation to enhance contrast on the band ratio image.

 

Open tmAtlanta in a Viewer and choose the bands that create a true color composite.  You will notice that the colors seem washed out and different features cannot be easily distinguished.  This is largely due to the similar reflectance values for the blue, green and red bands, which create similar reflectance values in adjacent pixels.  The landscape therefore takes on a muted, pastel appearance.

 

One way to get past the pastel washout effect is to create an image made up of band ratios which will highlight certain features, especially those with a high reflectance in one band and a relatively low reflectance in another.  You could do band ratioing by selecting the bands and conducting “band algebra,” where you divide the values for one band into another.  Alternatively, Imagine provides some of the most common band ratio indices and applies them for you, which speeds up the process.  Today, you will use the default indices that Imagine provides

 

Click on the Interpreter icon on the main menu bar, select Spectral Enhancement, then select Indices.  Enter tmAtlanta as the input file and tmAtlanta_min as the output file. Select Mineral Composite in the Select Function window.

 

The mineral indices combine three band ratios that highlight specific mineral types:

 

Band 5/Band 7 highlights clays (Georgia is famous for its “Georgia clay”)

Band 5/Band 4 highlights ferrous oxides

Band 3/Band 1 highlights iron oxide

 

Click on OK to create the new image.

 

Open the image in a new Viewer window using the gray scale and open layer 1 (which is now a layer that represents the ratio of Band 5/Band 7 for each pixel and indicates the presence of clays).

 

1.      In what kind of areas are the clay features highlighted (urban, rural forest, agricultural, etc.)?  Why do you see clays at these locations?  Link the two viewers if you are having trouble comparing sites on the two images.

 

 

 

 

 

 

 

 

Use the Raster/Band Combinations tool to look at the different bands. 

 

2.      Where do the highest concentrations of ferrous and iron oxides occur?  Speculate on why clays do not occur in these areas.

 

 

 

 

 

 

 

 

 

Compare the data size of your two images.

 

3.      What is the size in kb of your 2 images?  Why is the tmAtlanta_min image so much larger than the original tmAtlanta image? (check out the Layer Info or use the Cursor Inquiry to get a clue)

 

 

 

 

 

 

 

Now bring up the tmAtlanta_min image with all three bands showing.  To me this is a tough image to interpret because it has so much in the blue range of colors (indicating the dominance of clay in the landscape) and so much in the yellow range, indicating many ferrous (green band) and ferric oxides (red band).  The  contrast between features is not well demarcated, with many features having approximately the same hues.

 

One way to try to enhance contrast in an image where hues seem washed out is to use an Inten-sity/Hue/Saturation color transformation.  This requires creation of an intermediate IHS image and a final RGB image.  The equations use to create this transformation are detailed on pages 169-173 of the online ERDAS Imagine Field Guide, 1999.

 

To do an IHS transformation, click on the Interpreter icon in the main menu, select Spectral Enhancement, then select RGB to IHS.  Enter tmAtlanta_min as your input image and tmAtlanta_min_IHS as your output image.

 

Open up tmAtlanta_min_IHS in a new Viewer window.  This is the intermediate image used to create the final enhanced image and is not intended to be the final product for visual interpretation, although it is intriguing to examine it.  Now select Interpreter/Spectral Enhancements/IHS to RGB and enter  tmAtlanta_min_IHS as the input image and tmAtlanta_min_RGB as the output image.  On the IHS to RGB window, click Stretch I_S, then click OK to create the image

 

In a new Viewer window, open up the tmAtlanta_min_RGB image, but be sure to assign Layer 1 to the red gun, Layer 2 to the green gun, and Layer 3 to the blue gun.  This is because the RGB-IHS-RGB transformation has flipped the order of the bands (and my quick answer to your question is “No, I don’t know why the algorithm does not do the obvious step of reassigning the bands to the original color scheme”).  If you assign the colors in this manner, blue still represents clay, green represents ferrous oxides, and red represents iron oxides. 

 

The tmAtlanta_min_RGB image is similar to the tmAtlanta_min image, but with the differences between bands (i.e., minerals in this case) enhanced.

 

4.      Go to the green patch at the x,y location of 140,-175 (you can use the inquire cursor to locate this position).  In terms of mineral composition, what does this area represent?

 

 

 

 

 

 

 

5.      At 380, -351 there is a bright yellow patch near a road.  Immediately adjacent to it are areas that are much darker orange or reddish near 382, -346.  The adjacent road surface is black.  What do these colors represent in terms of mineral composition?

 

 

 

 

 

 

 

 

6.      If you were concerned with non-point source pollution in the form of sediment runoff, how might you use the results of your mineral composite image?

 

 

 

 

 

 

 

 

 

7.      How accurate do you think these results might be?  What could you do to evaluate the accuracy?

 

 

 

 

 

 

 

Part II, Tasseled Cap Transformation
 
The tasseled cap transformation “rotates” spectral values to enhance differences between brightness, greenness, and wetness.  The rotation process is akin to looking at a piece of paper with writing on it.  If you look at the paper edge-on, you will only see a thin sliver of paper and will not see any of the writing.  If you rotate the paper, however, you can now read the writing and get more information from the “image.”  By highlighting brightness, greenness and wetness, the tasseled cap rotation in turn enhances your ability to distinguish between forest, agriculture, urban, and differing water bodies.  A brief introduction is provided on pages 168-169 of the ERDAS Imagine Field Guide, 1999.

 

Close down all your viewer windows.  Then open Lamar_TM5 as a true color composite. 

 

This is a simulated TM5 image, created by resampling 128-band, 1-m spatial resolution Probe1 hyperspectral imagery.  The imagery was collected as part of a NASA EOCAP project on which I worked.  To collect the imagery, an across track, Probe1 hyperspectral scanner was mounted on a helicopter and flown approximately 600 m above the ground surface.  The Lamar River is located in Yellowstone National Park and has been the focus of much research on use of remote sensing imagery for mapping river and riparian features.

 

Click on the Interpreter icon on the main menu bar, select Spectral Enhancement, then select Tasseled Cap.  Enter Lamar_TM5 as the input file and Lamar_tasseled_cap as the output file.  Click on the Set Coefficients button in the tasseled cap window.  These coefficients represent the numbers used to rotate the different bands.  The default values are ones that previous research has shown to be optimal for separating brightness, greenness and wetness for given sensors. Do not change the coefficients, but do make sure that the sensor is set to Landsat 5 TM.  When all the entries are correct, click OK to create the image.

 

Open three new viewers (leave the true color composite open in Viewer 1).  In Viewer 2, open up layer 1 of Lamar_tasseled_cap as a gray image.  In Viewer 3, open up layer 2 as a gray image.  In viewer 4, open up layer 3 as a gray image.

 

  1. What features are highlighted (i.e., are the brightest) in layers 1, 2 and 3 of the tasseled cap transformation?

 

 

 

 

 

 

Take some time to look closely at the three layers and the results.  Focus on the areas near the trees.

 

  1. Which image does the best job of separating trees from tree shadow?

 

 

 

 

 

Now close down your gray image viewers and bring up the Lamar_tasseled_cap image in a new viewer, assigning band 1 to red, band 2 to green, and band 3 to blue.  Compare the 3 band composite tasseled cap image to the true color composite.

 

  1. Which features, if any, shown up more clearly on the tasseled cap image?  Are there features which are confused (e.g., different features showing up as the same color) on the tasseled cap image?

 

 

 

 

 

 

 

Finally, remember that there may multiple ways to highlight a certain type of features.  Using Lamar_TM5 as the input file, create a Vegetation Index image called Lamar_veg (see if you can work out how to do this on your own – ask if you get stuck).  The Vegetation Index in Imagine is a ratio of band 4/band3 or NIR/Red.  Open up Lamar_veg and layer 2 of Lamar_tasseled_cap as gray images in two viewers.

 

  1. Can you detect differences in how these two images highlight vegetation?  What about other features shown on the image?
 
 
 
 
 

 

 

 

Part III, Hyperspectral Imagery and Principal Components

 

Hyperspectral imagery collects reflection over a wide range of very narrow band widths compared to multispectral imagery.  For example, the Probe1 sensor collects reflectances in 12 to 16 nm bandwidths ranging from 400 nm (0.4 um) to 2500 nm (2.5 um). 

 

The hyperspectral image you will work with today only has 26 of the 128 bands sampled by the Probe1 sensor (specifically, your resampled image includes bands 1, 3, 9, 15 , 19, 22, 25, 32, 40, 42, 44, 48, 51, 56, 60, 64, 75, 79, 82, 90, 96, 100,106, 111, and 125 from the original Probe1 image, with bands 1,3, 9, 15 being in the visible spectrum and all the others in the SWIR.  In the process of resampling, Imagine renumbered these bands so that they now have labels of layers 1-26.  The bands I selected capture the major peaks and valleys in the spectral signals for the image.

 

Also, I resampled the original 16 bit image to make it an 8-bit image.  Reducing the number of bands and the radiometric resolution significantly reduces the size of the image in terms of data storage and avoids clogging the lab network, which cannot handle the amount of data in an actual high radiometric resolution hyperspectral image.

 

Open up Lamar_TM5 (bands 3,2,1) and Lamar_hyper (bands 4,3,2) in two separate viewers.  These band combinations should give you a true color composite for each image.  At face value, when looking at the visible wavelengths these images appear identical.  Open up a spectral viewer for each image and compare the spectral plots for the same vegetative feature. 

 

  1. In very general terms, describe how do the spectral plots differ between the TM5 image and the hyperspectral image.

 

 

 

 

 

 

 

 

 

  1. Why does the hyperspectral image consistently show such low reflectances in layers 8,12, 16 and 21?

 

 

 

 

 

 

 

 

 

 

  1. Open band 1 of the hyperspectral image as a gray scale image.  Why does this image show so little variations across the landscape?

 

 

 

 

 

 

 

One of the joys (and mind traps) of hyperspectral imagery are the many band combinations you can come up with to highlight different features.  Try some different band combinations, thinking about which features might be best highlighted with certain bands.  At a minimum, you should get some great color combinations.

 

Because there are so many possible combinations, it can be useful to “condense” the information in hyperspectral bands.  Principal components provide one mechanism for doing this.  In simple terms, principal components identify the trends in the spectral data and create images that shows these trends.  The first principal component always captures the biggest trend, the second principal component captures the second biggest trend, as so on.  You can have as many different principal components as there are bands, but in reality, the first few principal components generally capture all the signal (trend) and the later principal components only show the noise.  Your text book and pages 164-168 of the ERDAS Imagine Field Guide, 19999 provide overviews of principal components analysis.

 

In a new viewer, open layer1of Lamar_hyper_pc as a gray scale image. Lamar_hyper_pc is a principal component image that has been rescaled to 8 bit imagery and had some of the bands removed in order to avoid taking up too much data storage.  The pc bands that I have left in the image are the original pc bands 1-11, 14,15,19, although Imagine has relabeled them as layers 1-14.

 

Please note!: In the process of rescaling from 16-bit to 8-bit resolution, a great deal of information is lost, so the pc layers appear much more noisy than they did in they original form.  Reducing the radiometric resolution of the images therefore badly undersells the true potential of pc imagery.

 

Using the band combinations tool, look at how the image changes as you go from pc1 (layer 1) through pc 19 (layer 14).

 

  1.  Which layers appear to best distinguish between big categories of land cover (vegetation, water, wood, gravel, etc.)?  Alternatively, at what point in terms of layer number do the images become noisy and make it difficult to distinguish categories of land cover?

 

 

 

 

 

 

 

 

  1. Focus on the variations within the stream.  Do layers 1 and 2, or do layers 3 through 5 seem to depict more of the in-stream variations?  Combined with your answer to question 15, what does this indicate about the types of information depicted in different pc layers as the principal component number gets higher?

 

 

 

 

 

 

 

 

 

You too can easily create a principal components image.  Close all your viewers (in order to avoid clogging your computer memory). Click on the Interpreter icon on the main menu bar, select Spectral Enhancement, then select Principal Comp.  Enter Lamar_TM5 as the input file and Lamar_TM5_pc as the output file.  State that the Number of Principal Components Desired is 6, which is the maximum you can have with 6 input bands, and state that the output should be Signed 16 bit resoltuio  Click OK to create the image.

 

In separate viewers, open up layer 1 of Lamar_TM5_pc and Lamar_tasseled_cap as gray images.

 

  1. Compare layer 1 of the Lamar_TM5_pc image to layer 1 of the Lamar_tasseled_cap image.  You will note that they are very similar.  Why do you think this is the case?

 

 

 

 

 

 

 

 

  1. What is the size of your Lamar_TM5_pc image?

 

 

 

 

 

Your 16 bit image is relatively large compared to images with lower radiometric resolution. So let’s generate an 8-bit pc image to and see if it makes much difference in what we can detect on the imagery.  Repeat the process above, except stipulate that the output should be unsigned 8-bit resolution and that the output file name is Lamar_TM5_pc_8bit.  Make sure to select layers 1:6 so that all the layers are converted.  Click OK.

 

In separate viewers, bring up each image as a gray scale image.  Compare layer 6 in the 16 and the 8 bit images.

 

19.  How do the two images vary?  Specifically, which one provides more information and detail?

 

 

 

 

 

 

 

Hopefully your conclusion is that radiometric rescaling may save space, but at a significant cost in terms of degrading the imagery. 

 

As a final challenge to this lab, close all images except for your Lamar_tm5_pc image (the 16 bit version).  Look once again at the gray scale images of layers 1 through 6, taking special note of where you can detect within stream variations.  Based on your analysis, create a color composite image that enhances the in-stream variations to the maximum extent possible.  Remember, you can also apply contrast stretches to the image, or other enhancements we have examined.

 

20.  State the layers you chose and the enhancements you applied to create an image that maximized in-stream variations.