ERDAS IMAGINE

Exercise 1.
by: Sonya Remington

1.  Open ERDAS IMAGINE from your desktop by going to Start --> Programs --> ERDAS
IMAGINE 8.4.  When ERDAS opens, a Menu Bar running across the top of the screen and a
Viewer window below the Menu Bar appear.  All available functions for ERDAS can be
activated from the Menu Bar; however, not all functions can be activated from the Viewer.

2.  In the Viewer window go to File --> Open --> Raster Layer.....   and find the bbcgis folder. 
Open the "pnw98_bbg.img" file from the folder.  Make sure that the file type is ".img".

3.  On the toolbar inside the the Viewer window, click on the Zoom In button  .  Zoom in
on the image a few times.  Click on the image with the right mouse button and choose "Fit
Image to Window" from the menu that appears.  This command is analogous to the "Zoom to
Theme Extent or Full Extent" buttons in ArcView.  This part of the exercise is to illustrate the
point that ***Clicking on the right mouse button gives you access to many of the functions in
ERDAS***  So.....if you cannot find a function in the menu or on the tool bar, try clicking once
on the right mouse button.

4.  Change back to the arrow button  and click on  .  The following menu will appear

and white cross-hairs will appear on the image.  Use your pointer to move the cross-hairs by
clicking once in the intersection of the two lines and dragging the center of the cross around.  As
you do this, notice that the X and Y values in the table change.  Zoom in on the image a few
times so that you can see each individual pixel.
Move the cross-hairs again, but this time stay within one pixel.  Notice that the X and Y values
in the table still change, but that the values in the File Pixel and LUT (Look Up Table) Value
columns do not change.  The columns assign to each pixel a value from 0 to 255 (called a DN
number).  The File Pixel columns are the actual pixel values of your data.  The LUT Value
column assigns a number to each pixel based on a combination of red, green, and blue used to
make the colors on your screen.

5.  Changing Band Combinations.  Zoom back out to fit the image to the window and close the
table that is now open to get rid of the cross-hairs.  From the Viewer menu choose Raster -->
Bands Combinations and a table will appear:

Change the values on the right-hand side of the table to 5,2,1

 and notice that
the forested and grassy areas now appear green.  You can play around with this band
combination to give the image some really funky colors, but this feature is also useful.  For
example, by knowing that soils shows up really well with a certain band combination you can
use this combination to see them more clearly than you would with another combination of
bands.  Forests might show up more clearly in a different band combination than soils.

6.  The purpose of the next part of the exercise is for you to see how ERDAS can help you see
that different features (i.e. water, urban areas, forested areas, etc.) reflect different wavelengths. 

Change The band combination back to 4,3,2.   Now click on the graph button in the toolbar 
and a menu with appear:

Choose Spectral from this menu and hit OK.  A new

screen will appear:
Click on the cross button on this NEW Spectral Profile menu  and then click somewhere in
the water on the image.
A line will appear on your graph that describes the spectral profile at given point.  Now choose a
point over an urban area and see that you will get a different spectral profile on your graph.
Finally, choose a forested area and repeat the procedure.
As you have probably figured out by now, the forest has a different spectral profile.  With your
new knowledge about the spectral profile for a given feature, you can go back and change your
band combinations to make that feature show up more clearly on the image.

8.  This part of the exercise shows you how ERDAS can help you visualize how different
wavelength reflect along a transect of the land surface.  Close the Spectral Profile window to get
rid of your profile points.  Click again on the graph button in the toolbar  in the Viewer
window, but this time choose Spatial and hit OK.  A similar looking window as for the Spectral
Profile will appear; however, notice the the axes are different.

Click on the

line button  .  On the image, click once with the left mouse button to start drawing a straight
line.  Each time you want to change the direction of the line click on the left mouse button.
 To stop drawing the
line and to display the results on your graph, double click with the left mouse button.

9.  The next exercise will use the Spectral Profile Surface, which is similar to the Spatial Profile,
except that it is in 3D.  Close the Spatial Profile table and click again on the graph button in the
Viewer toolbar.  Choose surface and hit OK.  You guessed it.....another graph will appear:
 
 Click on the square button in

the graph window  and choose and area of interest (AOI - get use to this term) on the
image.  To do this, click once on the image with the left mouse button and a little x will
appear  .  Click and hold with the left mouse button on this x and drag the resulting rectangle
to the areal extent you desire.  Release the mouse button and a graph will appear.
 

Cool! Now change

the "PLOT LAYER" value in the Surface Profile window to display the surface of each of the
bands of reflectence within the profile AOI.
10.  Finally, we will learn about geo-linking image with ERDAS.  This means that when two
images have been geo-registered in a common real world projection and coordinate system they
therefore have both spatial information (the X,Y locations for each pixel or raster grid cell) and
they have descriptive data (the value of reflected energy for a given number of wavelengths).
This means that the geo-registered image data may be considered "geographic" data. Geographic
data my be "linked" through their common locational data and quired about their variation or
differences. In image processing, this is most often done to aid in understanding the nature of the
data by linking "high" resolution data with "coarse" resolution data.

Lets start by fitting the existing image to the window of viewer #1. Open a new viewer and load
the iconisbbc.img image file. Your desktop may look like this

Now use you RIGHT MOUSE BUTTON in the viewer #2 (the iconisbbc.img) to select the
GeoLink/Unlink option. You'll see the following window.
The instructions tell you to click the left mouse button in the viewer with the image you wish to
"link" with. The mouse will change symbols to inform you what action will take place in each
window. Select viewer #1 with you left mouse. An "index box" (or some call it the "zoom box" )
will appear. Now zoom in and out in BOTH viewers and see what happens. Change the size of

the "index box" (use the tool).

That's it for now. Close both viewers, and select EXIT IMAGINE from the session menu on the
top tool bar.

IMPORTANT:  When you close ERDAS, you will be asked if

you want to print the Log File.  Say NO, unless you really have something
against trees.

ERDAS IMAGINE
Exercise 2.
Unsupervised

by: Miles Logsdon

1.  Open ERDAS IMAGINE from your desktop by going to Start --> Programs --> ERDAS
IMAGINE 8.4. 

2.  In the Viewer window go to File --> Open --> Raster Layer.....   and find the bbcgis folder. 
Open the "pnw98_bbg.img" file from the folder. 

3.  Click on the image with the right mouse button and choose "Fit Image to Window" from the
menu that appears. 
4. Click on the classifier icon from the icon pane Select Unsupervised classification,
the dialog opens

5. Complete the Input window option.

NOTE: Input raster file (*.img) and provide an output name. You can output a signature file.
Initial clusters can be generated either arbitrarily or from an existing signature mean. You can
select to initial from statistics

 Enter the number of different NUMBER OF CLASSES (ie. Try 9)

 Set the maximum iterations up to 15 (this insure the process will not run on forever)
 Set the convergence threshold to .95 (ie. Stop processing when 95% of the pixels stay in
the same cluster between interations)

Click OK on the dialog and on the status dialog when the process is completed.

6. When the process is completed open a second viewer and load the new image.
7. Remember; the image is now a single dimension where each pixel has a value between 1-9
which assigns that pixel to a cluster of similar pixels. This image is a RASTER data layer. It
therefore has "attributes". From the viewer #2 select RASTER > ATTRIBUTE and click on one
of the shaded cells to change the color attribute for that attribute.

8. Make a color selection for the other pixel values. Remember that you can geo-link to the other
view to help you understand what the clustered values might mean.
9. One very nice function that helps evaluate clusters and/or classes is to visually overlay images.
Make sure you have a view with the spectral image file opened (file >open > raster layer and
display bands 4,5,3 respectively

10. from that viewer tool bar open a classified raster layer, make sure the "clear display" option
is turned off from the "raster options tab" found from that dialog box select Raster > Attributes
from the viewer, the editor will be displayed

11. NOTE: you can edit the way this editor appears by selecting Edit > Column Properties

12. Start by setting opacity for all classes to "0". In the Raster Attribute Editor, click on the word
OPACITY at the top of the Opacity column, then right-click-hold on the word Opacity and
select formula from the column option menu, click on the "0" (the zero on the number pad);
and click on Apply.

13. If you have not already change the color for class 1 do so now - to something like Yellow or
red so it’s easier to see, … click on the color patch and change the color.

Change the opacity for class 1 in the Cell Array to 1 and press return, this class will be shown in
the viewer. Also, change the opacity for class 7.
14. From the viewer menu bar select Utility |> Flicker … the dialog opens … turn on the auto
mode. The flashing pixels are the pixels in this class, click on the Class_Name in the editor and
give it meaningful name and assign it a meaningful color . Repeat for the other classes.

15. Now you go figure out what Utility > Swip and Utility > Blend will do.
IMPORTANT:  When you close ERDAS, you will be asked if
you want to print the Log File.  Say NO, unless you really have something
against trees.

ERDAS IMAGINE
Exercise 3.
Subset and Mosaic

by: Miles Logsdon

There are multiple ways to clip (SUBSET) an image, many of them use the subset command.
The method that I have used and found to work the fastest is the following: . 

1. Open ERDAS IMAGINE from your desktop by going to Start --> Programs --> ERDAS
IMAGINE 8.4. 
2.  In the Viewer window go to File --> Open --> Raster Layer.....   and find the bbcgis folder. 
Open the "pnw98_bbg.img" file from the folder. 

3. Click on UTILITY --> INQUIRE BOX

4. Move the box and enlarge it to include the area you want to clip

5. Click the DATAPREP button, and select SUBSET IMAGE

6. Fill out the options selecting the image input and output, and click on FROM INQUIRE
BOX button. This will make a .img file that contains the area, layers, and data type you selected.
REPEAT those steps until you have the subsets you wish to work with.

NOW here is one method to (MOSAIC) multiple Images (individual blocks) together. Once you
have all your clipped images, you will want to put them back together for your unsupervised
classification. Do the following:

1. Click the DATA PREP button, and select MOSAIC IMAGE

2. Click EDIT and ADD IMAGE or use the button . Add all your clipped images
3. Then click PROCESS and RUN MOSAIC

4. This will make a new .img file with all your clipped images included.

5. You'll need to contrast stretch the resulting image. Raster --> Contrast --> General
Contrast. Try the linear option.
You can play with the contrast to make the visual that you like.

IMPORTANT:  When you close ERDAS, you will be asked if

you want to print the Log File.  Say NO.

ERDAS IMAGINE

                                                                                      Exercise 4.

                                                                                            by:  Sonya Remington

1.  This exercise will show you how to edit the signature file created from an Unsupervised Classification, perform a
Supervised Classification, and check your data for accuracy by using Accuracy Assessment in ERDAS.

2.  Begin by opening ERDAS from your Start Menu:  Start > Programs > ERDAS IMAGINE 8.4.  Click on the

Classifer button  located in the main menu bar.  Select Signature Editor from the menu and a Signature
Editor table will appear.
Go to the File menu in the Signature Editor window and open the .sig file that you named in your unsupervised
classification.  You should now get values in your Signature Editor table.  The classes that you see here are those
that were generated by the unsupervised classification and are based on spectral properties.

3.  Now that your .sig file is open, you may begin to edit it.  This part of the exercise will show you how to merge
two classes.  You may want to do this after you go into the field and decide that two of the classes that were
separately grouped during the unsupervised classification are really the same thing (i.e. both are alder).  To merge
two classes, you first need to select them from the table.  To do this, click on the row of one of the Class #'s that you
want to merge.  Hold down the shift key and then click on the other class(es) being merged.  In this example we will
be merging classes 2 and 3 that were generated from an unsupervised classification into nine total classes.
Now go to Edit > Merge.  Notice that a new class containing the data that you merged has been added to the last row
in your table.  With the two merged classes still selected, go to Edit > Delete to get rid them.  The table should now
look like this:

Save the new Signature File and close the Signature Editor table.

4.  You will now learn how to perform a Supervised Classification using the edited Signature File.  Click on the
Classify button again and select Supervised Classification.  The following  will appear.
Enter fields as they are in the above example table.  Remember, the "Classified File" will be the output file for the
Supervised Classification and you can name it whatever you want.  When you are entering the "Input Signature File"
by sure that you enter the edited Signature file, not the original signature file from the Unsupervised Classification. 
Fill in the Decision Rules section as shown above.  Don't worry about what they mean for right now.  Hit OK and
wait for the image to be classified.  Hit OK when the classification is finished.

5.  Open two Viewer windows.  In one Viewer open the image Unsupervised Classification image and in the other
viewer open the Supervised Classification image.  Open a raster attribute table for each viewer by going to Raster >
Attributes from the menu of each.

One of the first things to notice is that if you change to color to red (or any other color) for the two classes that you
merged together, that they are now "No Data" and the red color does not appear on the image.  This is because they
no longer exist as separate classes.
Now make Class 10 (your new class from the merge) red and notice that it does appear on the image.
Another way to see that Classes 2 and 3 have been merged together is to go to the Raster Attribute table for the
Unsupervised Image and make Class 2 red.  Now you can compared the two images and notice that the red area in
the Supervised image is much more extensive than that is in Unsupervised image.
If you make class 3 in the raster attribute table of the Unsupervised image red, then the red areas in the two images
will cover the same area.
6.  Now that you have classified your image based on spectral properties and field observations, you want to
determine how accurate your classification was by doing an Accuracy Assessment.  You will compare pixels in a
classified image to reference pixels, which are the pixel for which you know the correct class.  First open the

pnw98_bbc.img in a Viewer.  Click on the Classifier button  in the ERDAS menu bar.  Selected
Accuracy Assessment and close the Classifier menu.  The following window will appear:
In this window, go to File > Open and select the image that resulted from your supervised classification.  Navigate
to the file and then click OK to load it.  This will be that image that will be used in you accuracy assessment.

In the Accuracy Assessment window go to View > Select Viewer and then click on the Viewer that is displaying the
pnw98_bbc.img image.

Now go to View > Change Color and be sure that the Change colors window looks like this:

"Points with no reference" should be white and "Points with Reference" should be yellow.  Click OK when this is
done.

7.  Now that you have told ERDAS which image you want to use in your accuracy assessment (the supervised
classification image), you need to enter the GPS coordinates for the reference points (the points for which you are
sure of the class because you actually went out in the field and had a look).   When you actually do an accuracy
assessment for your project, you will have the GPS coordinates from your ground truth field exercise entered into
a .txt file to import.  Since you may not have this at the time that you are trying this exercise, use the GPS points for
your samples from the first field trip.  These points are currently in an Excel file along with the attributes for each
point (i.e. stem density, biomass, or whatever you collected in the field).  First you'll want to make a file in Excel
with ONLY the X-Y coordinates for your GPS points.  You should end up with an Excel file that looks something
like this:
Do not label the columns X and Y, but be sure that your X values are in the 1st column and your Y values are in the
2nd.  Now save this new file as a .txt file (not a .xls file).  You can do this by selecting the "Text (Tab delimited)"
option from the "Save as type:" field when you are saving the file.  Doing this will change the file from ".xls" to a
" .txt" file.  (See the below for example.)

Now that you have done this, go back into ERDAS.  From the Accuracy Assessment window go to Edit > Import
User-defined Points.  Select the file with your GPS coordinates from the ASCII Point File window that pops up and
hit OK.  You now see an Import Options window.
For Field Type, select Delimited by Separator and don't worry about the rest right now.  Click OK.  Your Accuracy
Assessment window should not look something like this:

In this window, go to View > Show All and notice that the points now appear on the pnw98_bbc.img image in your
Viewer.
If you Zoom In on the image, you can see that these points more clearly.

Now you can enter your class values for the reference numbers in the Accuracy Assessment table.  Lets just say, for
the purposes of this exercise, that the reference numbers for your points are as follows (Remember, the class values
that you put in the reference column are determined from field observations):

The value in the reference column probably do not mean anything to you now, but they will after you have done
you're field work.  Go to Edit > Show Class Values and the Class column will be filled in with the values that are the
actual values of the pixels located at the X-Y coordinates that you indicated.
You can see that the first three points have the same values in the Class and Reference columns.  This means that
your supervised classification was correct ion assigning values to these pixels.  Points four through nine do not
match, meaning that your supervised classification did not assign values correctly to these pixels.

Finally, you can look at the statistics of your accuracy assessment by first going to Report > Options.  Check to
make sure that the Error Matrix, Accuracy Totals, and Kappa Statistics options are all turned on.  If they are not, do
so.  Then go to Report > Accuracy Report and the following table will appear:

In this table, you can find the Confusion Matrix (call the Error Matrix in ERDAS) and the other report options that
you selected.

IMPORTANT:  When you close ERDAS, you will be asked if you want to print the Log File. 
Say NO.
 
 
 
 

 Georegistration
and
Correction
ERDAS Example

By this time you should be able to open for display an image file, select the band combinations
that work best for your purpose, and examine the spectral and spatial profiles of the data. If not,
you should refere to the earlier exercises.

The basic proceedure will be to:

 display files
 start Geometric Correction tool
 record Geometric (ground) Control Points (GCPs)
 resample the image
 verify

1.) Begin by opening the file which you wish to georegister. (This may be
r:\lawers\data\landsat\bbc\bbc_2000.img) and display it and "fit" it in a viewer. As your mouse
moves within the display window note that the coordinates displayed in the lower left of the
viewer are in "row" and "column".
 Again, note the pixel coordinates in the lower left.

2.) Display an existing georegistered image in a second viewer. (This may be

r:\lawers\data\landsat\bbc\pnw98_bbc.img). Once again, note the coordinates in the lower left of
the viewer for this georegistered image. This image is georegistered to the UTM projection and
coordinate system. If you don't feel you understand map projections you may wish to visit the
USGS site call Map Projections

3.) Start the Geometric Correction tool from the viewer displaying the file to be retified.
Select RASTER > Geometric Correction from the viewer's menu bar and Polynomial from the
dialog box.

Each of these geometric models relate to specialized tasks. The

polynomial model has the most general application.

4.) Two dialogs boxes will appear. The Geo Correction Tool menu will be used now. You can
close the Polynomial model Properties (you will be selecting these parameters later)

5.) The next dialog box is the Reference Setup dialog. Note all the different way inwhich you can
collect reference points. Select (or accept) the option to use the Existing Viewer and click OK.
You will be prompted to click in the Viewer from which you wish to select the reference
cooridnates. Click in your second view - the pnw98_bbc.img. The refernce map information
dialog will open. Make sure the information is correct (which it is) and click OK.

6. You are now ready to begin assigning real world coordinates to the non-geographic registered
image. Your display will be filled with widows!
First, each viewer now has it's own "zoom" window above it. The zoom window display that part
of the image which is inside the "box" (some time called the link box). The Ground Control
Points (GCPs) are the small targets which are place on the image using the GCP tool. It is a
good idea to click on the "non-registered" image first. You can move the GCP with your mouse,
and the GCP can be placed in either the Viewer or in that viewer's zoom window. The
coordinates for each GCP is place in the GCP Tool CellArray (the table in the lower third of your
display). The Xinput and Yinput are the coordinates of the image to be registered and the
Xreference and Yreference are the coordinates of in the image that has already been registered.

Your proceedure is click on the GCP icon and then click on a location in the first viewer that
are easily identifiable in both images (the GCP #1 will be placed at that point and it's coordinates
will be filled into the CellArray). Then click again on the GCP icon and this time click on the
same location in viewer two (a coresponding GCP #1 will be placed at that point along with it's
real world coordinates in the CellArray) REMEMBER, you can move the GCPs in the zoom
window for better placement. Just click on the arrow icon .
You see that after you have collected Three or more matching pairs the software will begin to
place GCPs for you in an automatic mode. Also, you can right-hold you mouse button in the
Color column of the CellArray table to change the color of the GCP points.

Repeat this proceedure until you have all the points you want. (try to get 7 or 8 at a minium). To
delete a GCP, select the GCP in the CellArray and then right-hold in the POINT # column to
select Delete Selection.

When you have finished your points should spread out over the image.

7.) Clearly there is much more you can do with this tool. For now we will accept the limited
number of points and accept the solution to the transformation. Our next task is to resample the
image . Resampling is the process of calculatin the values for the rectified image and creating the
new file. All of the raster data layers (all bands) in the source file are resampled. The best known
algorithms for resampling are Nearest Neighbor, Bilinear Interpolation, and Cubic Convolution.
We'll use the Bilinear algorithm.

Click on the resampling icon from the Geo Correction Tool Menu . The resample dialog
opens. Enter a file name for the output (know where on the drive you will put the file) - select the
Bilinear Interpolation resampling method - click to exclude zero file values in the statistics. Also
set the Output cell size to 30.00 (we will be using a 30 meter DEM in the later exercises). Click
OK. This shouldn't take much more that a minute or two for this size of file.
A job status dialog opens to let you know when the processes is complete.

8.) Verify the rectification by displaying the new image in a viewer and Geo. Link the reference
image with the new image and use an Inquire Cursor to check that they are georegistered to each
other.
I did really well!

If you don't like you finished product, you may use the image I just made for other exercises.
(r:\lawers\data\landsat\bbc\my_resample.img)

 
 
 
 Simple corrections
ERDAS Example

1.) Begin by opening the file which you just finished georegistering (or use
r:\lawers\data\landsat\bbc\my_resample.img). Display the 432 band combination.

2.) Select INTERPRETER > Radiometric Enhancement > Haze Reduction option from the main
icon bar menu. Note along the way all the other options. The HAZE REDUCTION dialog will
appear.
This dialog offers you a chance to see the "model" that will preform a transformation on the data.
Click on the VIEW button to start the ERDAS Model build. The DeHaze model will be loaded.
This model is very simple. It illustrates the steps that ERDAS will take. The Input Raster is your
image, the PSF Kernel is a convolution filter. To understand how one pixel is convolved,
imagine that the convolution kernel is overlaid on the data file values of the image (in one band),
so that the pixel to be convolved is in the center of the window. The output value for the
convolution is, each value in the kernel is multiplied by the image pixel value that corresponds to
it. The products are summed, and the total is divided by the sum of the values in the kernel and
assigned back to the center pixel. The kernel is then moved to the next pixel.

A High-Frequency kernel (high pass) has the effect of increasing spatial frequency. That is to
say, when this kernel is used on a set of pexels in which a relatively low value is surrounded by
higher values the low value gets lower. When used with pixels in which a relatively high value is
surrounded by lower values the high value becomes higher. In this way a simple spatial
frequency of haze may be reduced.

3.) Now fill out the input and output fields of the Haze Reduction dialog. The software now
knows that the data is Landsat TM data and will now pick a "DeHaze" model for Landsat. Click
on VIEW to see it.

The grahic model illustrates the steps that ERDAS will go through. This "model" is specific to
Landsat TM data and as you can see it consist of picking between a matrix of weights for either
landsat 4 or 5 data and the then determining the majority value with a kernel, put that data out to
memory and the determining if any pixel has a value of zero before making the output file. I
wouldn't try to figure this one out. It is but one of the various approaches.

Fill out the dialog for input and output and click OK. Display your results side by side and look
at the spatial profiles of the reflectence values
I also did a Point_Spread model (not using the model for Landsat Data). Look at the results.
It looks like more haze, but what are the DN values like?

You should do one or both of these models and save a result.

5.) To preform topographic correction; Display the Digital Elevation Model for the site
(r:/users/lawers/data/landsat/bbc/bbc_dem30.img) and then select INTERPRETER >
Topographic Analysis > Topographic Normalization.
You will need the following information - SUN_ELEVATION=59.88;
SUN_AZIMUTH=137.92. Remember to look at the VIEW of the model which is being
preformed.

I'm glad someone else wrote this one for me. Click Okay and display your result.
Make some notes about what you see.

 
 
 
 
 

Extra Skills

VirtualGIS

1.) Open ERDAS and load your favoriate band combination. (I suggest the Landsat
pnw_bbc.img). Select the VirtualGIS Viewer option from the menu available under VirtualGIS

icon from the main icon menu bar

2.) In the VirtualGIS view window load the DEM for the associated image using the FILE >
OPEN > DEM option. Navigate to the directory where the DEM is located. If you have already
exported the DEM as an ERDAS image file then you may open it as an image (this is the case for
our class - the DEM for the pnw_bbc image is named bbc_dem30.img.) If the DEM is still in
Arc/Info grid format you may open it in that format. However you will have to change the "file
type" option in the file navigation widow to GRID.

3.) Drape either a classified image or a different band combination onto the 3-D representation of
the DEM, by selecting FILE > OPEN > Raster Layer. You may select the OPTION tab to define
the RGB bands.

4.) Link the two viewers by selecting VIEW > LINK/UNLINK Viewers. When prompted, select
the viewer with the 2-D image (i.e. viewer #1). The image will now shift to include a position
location for the "EYE" and for the "TARGET" that defines the viewing angle for the linked 3-D
image.
5.) Now move the location of the "EYE" and/or "Target" to change the perspective of the
VirtualGIS viewer.
6.) Select the NAVIGATION > POSITION EDITOR to adjust the viewing angle. It may be best
to alter only the PITCH attribute at this time (try -20). Make sure the Auto Apply checkbox is
selected.

7. Now reposition the "EYE" and "Target" in the linked 2-D window.
Expert Classifier

1. Select KNOWLEDGE ENGINEER from the Classifier icon menu

2.) Select EDIT > NEW Hypothesis to add the first hypothesis. Change the name to Vegetation,
Set the color to green and since you want this class as an output click on that checkbox. and

Click APPLY

3.) Repeat those steps and create a second hypothesis for the class water.

4.) Select the RULE graphic tool from the Knowledge Engineer icon bar. Click on the green
hypothesis rectangle for vegetation and then double-clik on the yellow NEW RULE rectangle to
open the RULE Propserties dialog. Change the name and leave the Compute from Conditions
button selected for Rule Confidence

5.) Click with the cell under Variable and select NEW VARIABLE. File this dialog out and
make sure you change the name, variable type (raster), select the input image the layer (4) and
use Cell Value
6.) Once you've applied this variable you can add others and edit them until you have a finished
knowledge map

7.) Once completed click on the RUN TEST icon. .

Here is how bad mine look.

 ERDAS
Hybrid Approach
UW GIS@U
Search
Home W
HOME
MORE ON THE TOPIC OF:

1. How to clip and mosaic Images

2. An exercise in Unsupervised Classification
3. Evaluating Classifications
4. An exercise in Supervised Classification
5. Evaluate Again
6. Accuracy Assessment exercise

Follow the steps below for an example of a "single" path through the steps of the
Hybrid Approach

Remember that the hybrid approach is designed to let you first work with smaller subsets of the
image that have been clipped and mosaiced together to form a selection of the image
heterogeneity for a given theme (such as forest, or urban, or water). Second, these mosaiced
images are allowed to "self-orgainized" themselves into spectral clusters using an unsupervised
approach. Third, if the unsupervised approach to forming signature clusters fails to distinguish a
land or sea surface material which "YOU KNOW IS PRESENT", you may add to the signature
file the result of a "supervised" clustering approach. Fourth, all signatures either from the
unsupervised or supervised approach are appended together to form a single signature file which
is used to "classify" (as a supervised classification step) the complete and full extent of the
originals image. There are more then one path through this approach. Below are the basic steps
in more detail.

 CLIP Representative Data

There are multiple ways to clip an image, many of them use the subset command. The method
that I have used and found to work the fastest is the following:

1. Start with the image you want to clip in a viewer

2. Click on UTILITY and select INQUIRE BOX
3. Move the box and enlarge it include the area you want to clip
4. Click the DATAPREP button, and select SUBSET IMAGE
5. Fill out the options selecting the image input and output, and click on FROM INQUIRE
BOX button.

This will make a .img file that contains the area, layers, and data type you selected.

Group multiple Images blocks

Once you have all your clipped images, you will want to put them backtogether for your
unsupervised classification. Do the following:

1. Click the DATA PREP button

2. Select MOSAIC IMAGE
3. Click EDIT and ADD IMAGE or use the button
4. Add all your clipped images
5. Then click PROCESS and RUN MOSAIC 

This will make a new .img file with all your clipped images included.

Unsupervised clustering on representative data

1. With ERDAS running click on the classifier icon from the icon panel
2. Select Unsupervised classification, the dialog opens
3. Input raster file (*.img from the moasic step) and provide an output name
4. NOTE: you should output a signature file
5. Initial clusters can be generated either arbitrarily or from an existing signature mean;
select initial from statistic
6. Enter the number of different NUMBER OF CLASSES (ie. Try 3 times the number of
expected classes)
7. Set the maximum iterations up to 40 (this insure the process does not run on forever)
 8. Set the convergence threshold to .95 (ie. Stop processing when 95% of the pixels stay in
the same cluster between interations)
9. Click OK on the dialog and on the status dialog when the process is completed.

Evaluate

1. Check for spatial distribution

 file|open|raster layer and display your multi bands combination (ie. 4,5,3 - this can be
either the original "un-clipped" image of the mosiac image) it doesn't matter. All you are
going to do is display the classified product "on top" of the unclassified image to help
determine what surface material the various self-organized clusters relate to.
 from the viewer tool bar open a classified raster layer (this could be - or should be - the
unsupervised classified data layer you just created), make sure the "clear display" option
is turned off from the "raster options tab" found from that dialog box. Click OK. This will
add two data layers into a single viewer (one on top of the other). You will now go
through the steps to allow you to view the "true color", or false color image while also
viewing the unsupervised clusters.
 select Raster | Attributes from the viewer, the editor will be displayed
 NOTE: you can edit the way this editor appears by selecting Edit | Column Properties
 Start by setting opacity for all classes to "0". In the Raster Attribute Editor, click on the
word OPACITY at the top of the Opacity column, then right-click-hold on the word
Opacity and select formula from the column option menu, click on the "0" (the zero on
the number pad); and click on Apply.
 Now change the color for class 1 to something like Yellow or red so it is easier to see,
click on the color patch and change the color.
 Change the opacity for class 1 in the Cell Array to 1 and press return, this class will be
shown in the viewer.
 From the viewer menu bar select Utility | Flicker. Now the dialog opens, so turn on the
auto mode.
 The flashing pixels are the pixels in this class, click on the Class_Name in the editor and
give it meaningful name and assign it a meaningful color
 Repeat steps 6 thur 9 for the other classes.

2. Edit signature file

Based upon what you discovered when evaluating the location and pattern of the unsupervised
clusters you may wish to remove or combine some clusters from the signature file.

 Click on the Classifer button located in the main menu bar.  Select Signature Editor from
the menu and a Signature Editor table will appear.
 Go to the File menu in the Signature Editor window and open the .sig file that you named
in your unsupervised classification.  You should now get values in your Signature Editor
table.  The classes that you see here are those that were generated by the unsupervised
classification and are based on spectral properties.
  Now that your .sig file is open, you may begin to edit it.  This part of the exercise will
show you how to merge two classes.  You may want to do this after you go into the field
and decide that two of the classes that were separately grouped during the unsupervised
classification are really the same thing (i.e. both are alder).  To merge two classes, you
first need to select them from the table.  To do this, click on the row of one of the Class
#'s that you want to merge.  Hold down the shift key and then click on the other class(es)
being merged.  In this example we will be merging classes 2 and 3 that were generated
from an unsupervised classification into nine total classes
 Now go to Edit > Merge.  Notice that a new class containing the data that you merged has
been added to the last row in your table.  With the two merged classes still selected, go to
Edit > Delete to get rid them
 Save the new Signature File and close the Signature Editor table.

Supervised

You may also wish to add a signature cluster (a signature or cluster mean) to the signature file
for a surface material that you know is present in the image but that you feel is too generally
addressed by the unsupervised approach. To do this you may "supervise" the assignment of the
cluster signature by "training" the signature.

There are a number of options for creating supervised training areas (drawing AOI's, growing
AOI's, selecting AOI's from feature space) you can use one or any combinations of these to
define training signatures.

NOTE: Remember you can go to the View/Arrange layers from the viewer menu and "right-
mouse-hold" on the AOI layer and delete the AOI you don't need any longer.

1. Display

 Display file the representative mosaic as you like

 click on the Classifier icon on the ERDAS "icon" panel
 select signature editor to start the signature editor dialog box, you can close the
classification menu bar
 explore how to view selected columns by selecting the view/columns option (ie. you
don't need the RED, GREEN, BLUE columns displayed

2. AOI = area of interest

 from the viewer menu bar select AOI/TOOLS

 using the viewer ZOOM make sure you can see a feature well (ie. zoom in)
 In the AOI tool palette, click on the POLYGON icon.
 In the viewer draw a polygon around the feature
 In the signature editor click on the add signature icon or select edit/add
 In the signature editor click inside the signature name column you just added and change
the name to something you understand (ie. forest, urban, etc.)
  Repeat these steps for one or two other landcover types

 3. Options for making training area (AOI) > Growing areas

 Select AOI/Seed Properties from the viewer menu bar, the region growing dialog opens
 Under geographic constraints, the area should be check, enter a value (try 300) and press
return
 For spectral distance, enter something like 10
 Click on options and make sure "include island Polygons" is checked
 Close
 In the AOI tool palette click on the region grow icon (looks like a magnifying glass)
 Click inside a different landcover type (say residential) in the viewer. A polygon will be
created based upon the region growing properties you selected … you can play around
with those settings.
 When you're happy click on the added Signature (or edit/add) in the signature editor .
 Again do as many as you feel happy doing

 4. Options for making training areas (AOI) > from feature space

 Select feature/create/feature Space Layers from the signature editor, the dialog box will
come up
 Fill in the name of the input raster layer you are viewing (ie. *.img) and click the "output
to viewer" button.
 Scroll down until you see the <imagename_2_5.fsp.img> row and click on (ie. select)
that row. That will output band 2 with band 5, clearly you can select a different
combination
 Click OK. and wait.
 Click OK when the job is done and a feature space viewer will come up.
 Now link the feature by selecting feature/view/linked cursors from the signature editor. A
dialog will open
 Set the viewer to feature space viewer number (or click on the select)
 Click link, white cross hairs should come up in both the feature space viewer and the
image viewer
 Drage the inquire cursor around in either viewer and see where geographic objects fall in
feature space
 Use the AOI tool to draw a polygon IN THE FEATURE SPACE VIEWER around the
geographic object.
 Now Add signature in the signature editor
 The signature you have just added is a non-parametric signature so select feature/statistics
from the signature editor menu bar to generate statistics.
 Remember to change the names on all of you signatures.

 Append the signature files of the unspervised and supervised

  If a signature file is not currently openned - Click on the Classifer button located in the
main menu bar.  Select Signature Editor from the menu and a Signature Editor table will
appear.
 Open any of the signature files you wish to append.
 Select FILE > OPEN from the menu bar of the Signature Editor and locate a second
signature file. However, this time MAKE SURE the Append button is selected.
 Repeat these steps until all of the signature files have been append into a single file.

Do a supervised classification of whole image using appended signature file

 It may be necessary to "associate" the signature file with the image you wish to classify.
Image association ensures that the signature file be applied to the "raw" (the original)
image. In the signature editor menu bar select EDIT > Image Association. Navigate to the
whole raw multispectral original image.
 In the signature editor menu bar select classify/supervised to perform a supervised
classification
 Enter a new name for both the output and the output distance file
 Click Attribute options and select minimum, maximum, mean and st.d.
 Under the NON-parametric rule, select feature space
 Click OK wait