Classification Tutorial

Remote Sensing Techniques Using ArcView 3.2 and ERDAS Imagine 8.4

CEE 4984/5984

GPS/RS Applications in CEE

May 1, 2001

Mark Dougherty Colin Kraucunas

Jen Verwest

1

Table of Contents

Page Table of Contents……………………………………………………………...

1

Project Statement…………………...………………...…………….………..

2

Tutorial 1: Mosaicking images in ERDAS Imagine 8.4………..……….……. Tutorial 2: Creating image subsets in ERDAS Imagine 8.4….……...…..……

3 6

Tutorial 3: Stacking images in ERDAS Imagine 8.4…......…………………. Tutorial 4: Supervised classification in ERDAS Imagine 8.4….……………..

11

15

Tutorial 5: Unsupervised classification in ERDAS Imagine 8.4……………... Tutorial 6: Vegetative mapping in ArcView 3.2………......………………….

22

30

2

Project statement The purpose of this document is to present the techniques needed for basic analysis of

remotely sensed images, including mosaicking, subsetting, stacking, unsupervised and

supervised classification, and vegetative mapping. The document is written as a set of six

tutorials with step-by-step instructions. It is assumed that the student has access to a

working version of both ERDAS Imagine 8.4 and ArcView 3.2, and has a basic

familiarity with both software packages.

It is recommended that the two data sets provided with this document be used to

complete the tutorials. The data sets include both SPOT and Landsat images for the New

River Valley, located in southwestern Virginia. SPOT image files 37080SE.BIL and

37080SW.BIL are located in folders of the same name, as shown in the figure below.

These two SPOT images, when mosaicked together, provide remotely sensed coverage

for the entire New River Valley. The remaining files, shown below as nrvband1.img

through nrvband5.img, represent the five bands of light reflectance (visible through mid-

infrared) captured by Landsat instrumentation. The Landsat images, when stacked

together, provide remotely sensed data coverage for the same geographic area as the

SPOT images.

3

Tutorial 1: Mosaicking images in ERDAS Imagine 8.4 This tutorial will help you join two images together in a process called mosaicking. It is often helpful to mosaic two images together, especially when the two joined images result in a complete picture of an area being analyzed. Please note that the images must be in *.img format to complete the steps outlined in this tutorial. If the images are in another format, open them in the viewer and save them as *.img files. Complete the following steps to convert images to *.img files:

1. In ERDAS Imagine, click the Viewer button to open a new viewer. 2. Click the Open Layer button or choose File->Open->AOI Layer 3. Select the correct file type, select the file, and hit OK 4. Choose File->Save->Top Layer As… 5. Select a path and filename to save the image and hit OK 6. When the copying image process is finished, hit OK 7. Close the viewer

Complete the following steps to mosaic two IMAGINE images (*.img): 1. Click on the DataPrep button on the ERDAS Imagine

Toolbar 2. Click on Mosaic Images… 3. The Mosaic Tool Window will open. Choose Edit-

>Image List…

4. The Mosaic Image List will open. Click on Add… 5. The Add Images for Mosaic window will open, as shown below. Click Add… to add

images. Find and select the first image file and click Add. Find and select the second image and click Add. When finished, click Close, then click Close on the Mosaic Image List window.

4

6. The outlines of your two files

should be shown in the Mosaic Tool window, as shown at right. To edit the mosaicking options, choose Edit->Image Matching…

7. The Matching Options window

shown at right will open. Choose “For All Images” as the Matching Method and click OK.

8. To edit the overlap area

options, choose Edit->Set Overlap Function…

9. The Set Overlap Function

window shown at far right will open. Choose the “Feather” function and click Close

5

10. You are now ready to mosaic the images. Choose Process->Run Mosaic… 11. Specify an output path and filename for the combined image, then click OK 12. ERDAS Imagine will run the Mosaic procedure, which may take a few minutes.

When it is complete, click OK in the progress dialog box as shown below.

13. In the Mosaic Tool window, choose File->Save to save the mosaic options as a *.mos

file. Choose the path and filename and click OK 14. Close the Mosaic Tool window. Your mosaicked image is ready for viewing. 15. Click on Viewer in the ERDAS Imagine Toolbar to open a new viewer 16. In the viewer, choose File->Open->AOI Layer… or hit the Open Layer button 17. Change the file type to IMAGINE Image (*.img), select your image file and click OK 18. Your mosaicked image is displayed in the viewer.

6

Tutorial 2: Creating image subsets in ERDAS Imagine 8.4 Subsetting an image can be useful when working with large images. Subsetting is the process of “cropping” or cutting out a portion of an image for further processing. In this tutorial, you will create an image subset using the following steps.

1. Open the image in a viewer (any one of the nrvband*.img files can be used). 2. Using the rectangle selection tool, create an AOI (area of interest) rectangle

around the desired subset area. This can be done from the viewer menu by selecting AOI -> Tools.

3. Choose Utility -> Inquire Box ->Fit to AOI. This will automatically set the extents of the Inquire Box to match those of the AOI. Hit Apply.

Rectangle tool

7

4. Next, select Data Prep-> Subset Image. This will give you the following dialog box:

5. As shown at right,

select the input file using the browse button, then specify the output file name and location.

6. Hit the “From Inquire

Box” button. This sets the extents of the subset to match those of the inquire box selected in step 3.

7. Hit OK to create a

new subset image.

8. A progress bar, shown below, will display when the new subset image is created. Hit OK to accept the new subset image, which will be saved to the location specified in step 6.

9. Repeat steps 1 – 8 for each of the four remaining band files (nrvband*.img). [Note: For each image, first load, then select the rectangular AOI from step 2. When you repeat step 3, you will automatically fit the coordinates of the pre-selected AOI to the current image.]

8

10. The next step in the subsetting process is to combine three band images into one multispectral image. Begin by selecting Data Prep-> Create New Image. The following dialog box appears.

11. From the Create File dialog box, select the output file browse button and specify a path and output file for the combined image. Hit “From Inquire Box.” This will set the coordinates of the new image to match those of the subsets previously created.

12. The cell size

fields should be changed to 28.50 to match those of the subset images. [Note: you can find this information under Utilities-> Layer Info, as shown at right.]

9

13. When all values in the Create File dialog box for Date Type and Output Options are as shown above (step 10), hit OK. A progress bar will appear. When the process has completed, hit OK.

14. The next step is to

open the new image created in step 10. The image will appear all white because it does not yet have raster information in it. To do this, select Raster -> Band Combinations. The dialog box, below, will appear.

15. Use the

browse buttons to select the appropriate band subset files for each band.

Note: The image combination shown above represents the default band combinations representing visible blue, green, and red bandwidths.

16. Each subset image only has one

layer, therefore layer settings can be left at one. Hit apply and you should have a color image similar to the one shown at right.

10

17. In order to save your image with these settings, select File -> View to Image File, then select a name and output location.

18. Depending on the types of analysis you are doing, it may be necessary to re-order

the band combinations to obtain the correct appearance. This is done by once again selecting Raster -> Band Combinations.

Note: The image combination shown above represents the default band combinations representing visible green and red bandwidths, as well as near infra-red. The resulting image, shown below, is widely used in the classification of vegetation vs. bare surfaces because the healthy vegetation has a very high infra-red reflectance, while bare earth or asphalt has very little to zero infra-red or red reflectance.

11

Tutorial 3: Stacking images in ERDAS Imagine 8.4 In order to analyze remotely sensed images, the different images representing different bands must be stacked. This will allow for different combinations of RGB to be shown in the view. The following steps show how to stack images. 19. In ERDAS, click on the Interpreter button on the ERDAS Imagine Toolbar 20. When the Image Interpreter dialog box appears, select the Utilities. 21. I n the Utilities menu, select Layer Stack. The Layer Selection and Stacking dialog

box will appear.

22. In the Layer Selection and Stacking dialog box, select the first layer for the Input File by selecting the browse button. Navigate to the desired folder, and select the image that will be Layer 1 in the new image.

23. Click the Add button to create this file as Layer 1.

12

24. Continue to select the input files in order and click Add. The files will become the layers of the new image.

25. Once all the files are added, create an Output File by selecting the browse button and

navigate to the desired folder. Name the file and hit Ok. 26. Verify the remaining options, and click Ok.

13

27. When the Modeler dialog box is complete, click Ok.

28. Open a new Viewer and open the newly created raster image.

29. To change the layer being displayed, choose Raster then Band Combinations in the

Viewer dialog box. 30. Change the layers that are displayed for the respective colors in order to get the

desired bands visible. The selection shown below displays a combination commonly used for land use and vegetative mapping. The Red layer is near infrared, the Green layer is red, and the Blue layer is green.

14

31. Hit Ok to view the image, below.

15

Tutorial 4: Supervised classification in ERDAS Imagine 8.4 Image interpretation is the most important skill to be learned before producing accurate land use maps from remotely sensed data. Supervised classification allows the user to define the training data (or signature) that tells the software what types of pixels to select for certain land use. Facts about the area, knowledge about aerial photography, and experience in image interpretation permit pixels with specific characteristics to be selected for a better classification of the image. Through experience, supervised classification becomes easier and more accurate.

19. Open the raster image with the different bands stacked in layers as created in tutorial 3.

20. On the Viewer menu, choose Raster and select Band Combinations from the list.

21. When the Band Combinations dialog box appears, change the layers so that the

Red Layer is Near Infrared (Layer 4), the Green Layer is Red (Layer 3), and the Blue Layer is Green (Layer 2).

22. Hit Ok. The image that appears, shown on following page, is a common band combination used to evaluate land use and vegetation. In this image, green (band 2), red (band 3), and near infra-red (4) band are represented by blue, green, and red, respectively. This band combination and color selection make identification of bare surfaces easily distinguishable from healthy vegetation.

16

23. On the main ERDAS menu, select the Classifier button.

24. In the Classification menu, select Signature Editor. The signature dialog box will appear with a new file opened to begin defining training data. The signature editor allows the user to select areas of interest (AOI) to be used as training samples to categorize the photograph.

25. In the Viewer menu, select AOI and then choose Tools.

26. In the AOI Tools dialog box, select the polygon tool to create an AOI.

17

27. Zoom into an area to be classified, and draw a polygon around a specific region to be used for training data.

28. With the AOI still selected, hit the Select New Signature(s) from AOI button.

18

29. Change the signature name by clicking in the field and entering a more descriptive name. Change the color to be displayed that is defined by this signature by clicking in the color field and selecting a new color.

30. Continue to select more signatures until all desired land uses or areas are selected.

31. In the Signature Editor dialog box, select File then Save As. Browse to the desired folder. Name the file and click Ok. The resulting image classifications will distinguish deciduous from coniferous trees using different shades of green. Urbanized bare areas will be represented by red, water by blue, and agriculture by yellow. With careful pixel selection, bare soil can be distinguished from urbanized bare areas. In this example, based on knowledge of the area, bare soil is considered to be bare soil fields (as opposed to urbanized impervious areas such as asphalt).

19

32. Close the Signature Editor dialog box by selecting File then Close. Now that the Signature File has been created to select the different classifications, a supervised classification can be performed.

33. In the Classification dialog box, select the Supervised Classification button. The Supervised Classification dialog box will appear.

20

34. To select the Input Raster File, hit the browse button and navigate to the desired

folder. Select the file to be classified that is open in the viewer.

35. To select the Input Signature File, hit the browse button and move to the preferred folder. Select the Signature File previously created.

36. To create a Classified File, hit the browse button and navigate to the folder where

files are being saved. Name the file and hit Ok.

37. Verify the other settings below and click OK.

38. When the Status dialog box is complete, click Ok.

39. Open the newly classified file to observe the classifications and verify the signature file. Most of the land use is deciduous trees (light green), followed by agriculture (yellow), bare soil (pink), and urban (red). Water is shown in blue.

21

40. If the signature file needs to be edited, open the Signature Editor by clicking on it in the Classification dialog box.

41. In the Signature Editor dialog box, select File then Open. Navigate to the desired

file and open the previously created signature file.

42. Edit the signature file by using the add , replace , merge , and delete (in the edit menu) options.

43. Save the redefined signature file and repeat steps 15 through 21.

Through experience, supervised classification becomes easier and more accurate. Image interpretation is the most important skill to be learned before producing accurate land use maps.

22

Tutorial 5: Unsupervised classification in ERDAS Imagine 8.4 Performing an unsupervised classification, covered in this tutorial, is simpler than a supervised classification. Unfortunately, simplicity comes at a cost. In unsupervised classification, the signatures are automatically generated by an algorithm named ISODATA. The resulting classification has less discerning abilty than a supervised classification due to the lack of training data supplied to the clustering algorithm. In this tutorial, you will perform an unsupervised classification using the following steps.

44. On the main ERDAS menu, select the Classifier button, which will open the Classification menu.

45. On the Classification menu, shown at right, select Unsupervised Classification.

46. The Unsupervised Classification dialog box, shown on the next page, will appear.

23

47. Under Input Raster File,

place the name of the file and file location to classify. Under Output Cluster Layer, give the target destination and filename of the output file. Click the Output Signature Set to disable the Output Signature Set filename box (you will not be creating a signature set as you did with supervised classification).

48. Next set the clustering

options as follows; Maximum Iterations = 24 and Convergence Threshold = 0.950, as shown below.

49. Put 5 for number of

classes. Click OK to begin the classification process.

50. The job status dialog box,

shown below, will alert you as to progress. Hit OK when the process is 100% complete.

24

51. After the classification process is complete (yes, that is all there is to unsupervised classification processing), you will want to evaluate and test the accuracy of the classification. You may also want to reclassify your image using a different number of classes with a different number of iterations. Let’s see what the classified image looks like.

52. A good way to evaluate the

results of the unsupervised classification is to overlay the original image data with your *_isodata.img file. To do this, from the Viewer menu bar open the original image with File ? Open ? Raster Layer. The image at right should appear in your viewer.

53. Change the colors displayed to match those used in tutorial 4 by selecting from the Viewer menu bar Raster ? Band Combinations. Change the Layers to Colors 4,3, and 2. The following familiar image should appear.

25

54. The next step is to overlay the classified image over the original. Open a new file dialog box as described in step 9, above. Select the directory where you saved the *_isodata.img file previously. In order to add the new image without clearing the original image, click the Raster Options tab at the top of the Select Layer to Add dialog, as shown below, and clear the Clear Display box.

55. After loading the *_isodata.img image, select Raster ? Attributes then Edit? Column Properties. You will rearrange the columns of the following editor box as follows;

26

56. Select the column headings one at a time as shown and hit the Up key to rearrange the headings, as shown. When complete, hit OK.

57. The Raster Attribute Editor box, below will appear with columns as below.

27

58. Next, select the Opacity column to highlight all the values in blue (shown above). Select Edit ? Formula and place a zero in the Formula box function above as a value for all cells in the column. This makes the newly classified image effectively transparent, until you are ready to add colors, one at a time.

59. Change the color and opacity

in Class 1 to red and 1, respectivel, and note the change in the raster image displayed.

60. Continue editing the five classes one at a time, adding colors of your choice to represent what you think to be specific features. The appearance of the final drawing will depend on the color combination and number of classifications you choose. For example, the first color chosen above, red, does not seem appropriate for water, which is obviously displayed in the image. This color and heading can easily be edited to blue by clicking on the cell. The resulting drawing and legend may look like the following, at right. Note that the Classes have not been given names, yet. With unsupervised classification, it is often necessary to do

28

ground truthing after the classification is complete. As stated previously, several attempts at the unsupervised classification may be desirable to achieve a land classification that is understandable.

61. A somewhat more intuitive image display is presented below. The color scheme is the same, but the image has been re-classified using only four classes. Although the map below may be easier to interpret than the one above, it likely will have somewhat less discriminatory detail. Be aware that there are bound to be trade-offs in the selection of classes that will depend upon the use being made of the data and the land use being categorized.

29

62. A useful aid to evaluating unsupervised classifications is through the use of the Utility menu on the viewer menu, specifically the Blend, Swipe, and Flicker commands. Each of these commands will bring a control box that will either blend, or swipe, or flicker the upper-most image alternately with the lower image within the View. Recall from step 11 that you overlaid the classified image on top of the original, which you can now view in periodic “swipes” or “flickers” or “blends” to help evaluate the types of land cover beneath your classified image.

63. The following are the three control boxes that are activated by the Blend, Swipe,

and Flicker commands. We leave it to you to experiment with these handy tools as you gain more experience in your classification skills. Enjoy!

30

Tutorial 6: Vegetative mapping in ArcView 3.2 Vegetative mapping finds areas of healthy vegetation as well as stressed vegetation from a remotely sensed image. In order to do vegetative mapping, two bands are needed, visible red and near infrared. These bands are chosen because vegetation, especially healthy vegetation, is very reflective in the near infrared range and it provides good contrast with water. In order to distinguish healthy vegetation from other reflective sources, the visible red is chosen, which contrasts vegetation from bare soil, rocky surface, and man-made features. From the visible red and near infrared layer, a Normalized Difference Vegetation Index (NDVI) is calculated in ArcView using the formula NDVI = (IR-R) / (IR+R), where IR is infrared and R is visible red. A single band theme in grayscale is created that highlights vegetation. The following tutorial shows how to quickly build an NDVI vegetative mapping image in ArcView 3.2.

64. In ArcView, add the Image Analysis extension by selecting File then Extensions.

65. Check Image Analysis and click Ok.

66. Open a new view to add the raster image for the vegetation mapping with the green, visible red, and near infrared bands created in tutorial 2.

67. Open the image by clicking the Add Theme button.

68. Change the Data Source to Image Analysis Data Source and Navigate to the

desired folder. Click Ok.

31

69. Click the check box next to the theme to draw it.

70. Double-click the theme to bring up the Legend Editor. Change the Red Band to the Red Layer, the Green Band to the Green Layer, and the Blue Band to the Blue Layer. Click Apply and close the Legend Editor dialog box.

32

71. The image is displayed in the commonly used form for vegetative mapping.

33

72. In the Main menu, select Image Analysis the Vegetative Index. The Vegetative

Index dialog box appears.

73. Change the Near Infrared Layer and Visible Red Layer to Layer_Red and Layer_Green, respectively, or according to how the image was generated in ERDAS. Click Ok.

74. Click the check box next to the NDVI theme to draw it.

34

75. The bright areas represent areas of vegetation and the dark areas represent water, urban, and bare soil.

76. Save the image by selecting the NDVI theme. In the View menu, select Theme

then Save Image As.

77. Navigate to the desired folder, and name the image. Verify that the file type is the desired format, and click Ok. Choose IMAGINE Image as the file type to be able to use in ERDAS.

78. Select No to add the image as a theme to the view.