Simple Image Classification Using Raster Calculator: Work Flow
Uploaded by
Saaz ZeeSimple Image Classification Using Raster Calculator: Work Flow
Uploaded by
Saaz ZeeTrim Size: 216mm x 279mm Dixon e05.tex V3 - 08/24/2015 6:55 P.M.
Page 1
Simple Image Classification Using
Raster Calculator
Work Flow You will be learning these tools in this exercise
Copy raster to
GRID format
Record Pixel
Ranges for Band
Copy Raster Tool
Use Map Algebra
to Classify the
image
Raster Calculator
You will be working from this file:
C:\Workshop\Digitizing\Digitizing_Data4
ArcGIS is not capable of doing automated image analysis, but can be used to explore the basic steps of image analysis using the
Spatial Analyst Extension.
Start ArcMap and open a new empty map. Add Buckhorn_landsat.img from your C:\Workshop\Digitizing\Digitizing_Data4 folder.
GIS and Geocomputation for Water Resource Science and Engineering, First Edition. Edited by Barnali Dixon and Venkatesh Uddameri.
© 2016 John Wiley & Sons, Ltd. Published 2016 by John Wiley & Sons, Ltd.
Trim Size: 216mm x 279mm Dixon e05.tex V3 - 08/24/2015 6:55 P.M. Page 2
2 SIMPLE IMAGE CLASSIFICATION USING RASTER CALCULATOR
Take a moment to orient yourself to the features in the scene. There are many agricultural lands and open fields as well as wetland and
forested areas.
First, you will need to convert the Orthorectified Image to ESRI Grid data format.
1. Start the full ArcCatalog from the Start menu and navigate to the C:\Workshop\Digitizing\Digitizing_Data4 folder. Right-click on
the Busckhorn_landsat.img data layer, and select Export > Raster to different format.
• You will need to export the image to ESRI GRID file format so ArcGIS
can display individual bands.
• In the Copy Raster tools you can’t select this format, but you simply need
to type in an output file location C:\Workshop\Digitizing\scratch and name
(B_landsat) without a file extension.
• Leave the optional elements blank.
• Click OK to process.
This will create a copy of the six-band Landsat image in grid format; you will also
see six separate grid files in ArcCatalog for each of the six bands, but these are just
different views of the same data file. In ArcMap you can decide to open the six-band
version of the ortho2 grid or each band separately.
Trim Size: 216mm x 279mm Dixon e05.tex V3 - 08/24/2015 6:55 P.M. Page 3
SIMPLE IMAGE CLASSIFICATION USING RASTER CALCULATOR 3
Now that you have converted the image to grid format, in ArcMap, add the six separate bands from your scratch folder. NOTE:
You only need to add the ones that have c1, c2, and so on, at the end of “b_landsat.”
Let’s explore the pixel values a bit closer of both the original Ortho image and the
new grid layers.
• Zoom in close on the original image and use the Identify tool to click on a pixel. You
may want to move it above the six single bands you just loaded in the ArcMap TOC.
• Make sure the Identify from: is set to Top-most Layer.
Notice how the pixel values for each of the three color bands are shown in the Identify Results
window.
• Click on neighboring pixels of similar color to see how the pixel values change slightly.
• Click on pixels of different color to see how the pixel values are quite a bit different.
One of the limitations of ArcGIS is that you can only work with three bands of a multiband
image – but since you have converted the multiband image to six individual images, you can use all
six bands.
Now use the identify tool to examine the value of a single pixel for all of the visible layers. Be
sure <Visible Layers> is selected in the Identify from: drop down menu.
The fundamental concept in image analysis is that you can use the pixel values in the different bands
to identify similar features. You will now manually practice this concept in a very simple analysis
exercise using the Spatial Analyst extension.
If necessary in ArcMap, select Customize > Extensions, and in the Extensions dialog turn on the
Spatial Analyst Extension. Click Close. You are now ready to use Spatial Analyst Tools. You will
identify the appropriate pixel values for specific features and then use a query tool to find similar
areas.
Let’s use with an easily recognizable feature.
• Zoom in to an area of vegetation (light to dark green in a natural color Landsat).
• Explore the pixel values range for the vegetation for each of the six bands.
• The range for each band is often quite narrow – typically only a few pixel values.
Your task is to record (not in the workbook, on separate paper!!) pixel values ranges for vegeta-
tion in the table below. A few tips:
• Make sure you record the band number correctly.
• In theory it is possible to have more than one pixel range value within a single band for a particular
type of feature.
• To properly recognize what is vegetation and what is not, you will need to use simple visual inspection
of the original landsat image – it can help to use both a natural color image (3-2-1) as well as a setting
which is known to show vegetation more clearly (4-3-2). To do this just go to the properties of the
original landsat image choose Symbology.
• You will need to use a couple of different across the study area, to get an appropriate range of pixel
values
Band Min Pixel Value Max Pixel
# Value
Now you will use these pixel values to query the grid data 1 90
layers to find similar areas. The query tool in Spatial Analyst 2 43
3 48
is built into the Raster Calculator.
4 83
5 156
6 70
Trim Size: 216mm x 279mm Dixon e05.tex V3 - 08/24/2015 6:55 P.M. Page 4
4 SIMPLE IMAGE CLASSIFICATION USING RASTER CALCULATOR
In ArcToolbox select Spatial Analyst Tools > Map
Algebra >, Raster Calculator and create the query expression
for one of your bands. Querying a grid uses a similar logic
as querying vector data, but the tool works a bit differently.
Since each pixel only has one numerical value and no other
attribute data, there is no option to select attributes from a list.
Make sure your expression selects only those pixels from the
selected band that fall within the range (Min and Max) you just
identified. See the image below for an example – your actual
values could be different, depending on where you clicked in
the image.
Save your results to C:\Workshop\Digitizing\s cratch and
call it Reclass.
Click OK.
The result of a Raster Calculator operation on a grid data
layer is a new gird data layer. In this case the result is a value
of zero (for not meeting the specified condition) or a value of 1
(for meeting the specified condition). Symbolize this new grid
layer so that values of 0 are set to transparent. Your result so
far should look something like the image below.
Compare the result with the original Landsat image. Zoom back into the areas of vegetation you identified visually. Notice how the
resulting grid layer does a reasonable good job of describing the vegetation, but it is likely not perfect. Some areas appear to have been
misclassified as “vegetation,” while some parts within the vegetation were not classified. This could be cleaned up by incorporating
a more complex rater calculator formula that included the min and max for other bands as well, or creating an image for each band
range and then adding them and only keeping the common areas that were “1’s” (vegetation).
Close ArcMap. End of Exercise.